# INTEGRATIVE RESEARCH ON ORGANIC MATTER CYCLING ACROSS AQUATIC GRADIENTS, 2nd Edition

EDITED BY : Nicholas D. Ward, Thomas S. Bianchi, Patricia M. Medeiros, Michael Seidel, Richard G. Keil and Carol Robinson PUBLISHED IN : Frontiers in Marine Science and Frontiers in Earth Science

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88966-154-1 DOI 10.3389/978-2-88966-154-1

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# INTEGRATIVE RESEARCH ON ORGANIC MATTER CYCLING ACROSS AQUATIC GRADIENTS, 2nd Edition

Topic Editors:

Nicholas D. Ward, Pacific Northwest National Laboratory, University of Washington, United States Thomas S. Bianchi, University of Florida, United States Patricia M. Medeiros, University of Georgia, United States Michael Seidel, University of Oldenburg, Germany Richard G. Keil, University of Washington, United States Carol Robinson, University of East Anglia, United Kingdom

From mountains to the sea, fjords represent one of the earth's steepest aquatic gradients Photo: Patricia Medeiros, Norway

The goal of this research topic was to motivate innovative research that blurs traditional disciplinary and geographical boundaries. As the scientific community continues to gain momentum and knowledge about how the natural world functions, it is increasingly important that we recognize the interconnected nature of earth systems and embrace the complexities of ecosystem transitions. We are pleased to present this body of work, which embodies the spirit of research spanning across the terrestrial-aquatic continuum, from mountains to the sea.

Publisher's note: In this 2nd edition, the following article has been updated: Sawakuchi HO, Neu V, Ward ND, Barros MdLC, Valerio AM, Gagne-Maynard W, Cunha AC, Less DFS, Diniz JEM, Brito DC, Krusche AV and Richey JE (2017) Carbon Dioxide Emissions along the Lower Amazon River. *Front. Mar. Sci.* 4:76. doi: 10.3389/fmars.2017.00076

Citation: Ward, N. D., Bianchi, T. S., Medeiros, P. M., Seidel, M., Keil, R. G., Robinson, C., eds. (2020). Integrative Research on Organic Matter Cycling Across Aquatic Gradients, 2nd Edition. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-154-1

# Table of Contents

## CHAPTER 1

#### INTRODUCTION

*06 Editorial: Integrative Research on Organic Matter Cycling Across Aquatic Gradients*

Nicholas D. Ward

## CHAPTER 2

### TERRESTRIAL FLOW PATHS

*09 Dissolved Organic and Inorganic Carbon Flow Paths in an Amazonian Transitional Forest*

Vania Neu, Nicholas D. Ward, Alex V. Krusche and Christopher Neill

*24 The Genesis and Exodus of Vascular Plant DOM From an Oak Woodland Landscape*

Peter J. Hernes, Robert G. M. Spencer, Rachael Y. Dyda, Anthony T. O'Geen and Randy A. Dahlgren

*41 Molecular and Optical Properties of Tree-Derived Dissolved Organic Matter in Throughfall and Stemflow From Live Oaks and Eastern Red Cedar*

Aron Stubbins, Leticia M. Silva, Thorsten Dittmar and John T. Van Stan

# CHAPTER 3

#### RIVERS AND STREAMS


William C. Gagne-Maynard, Nicholas D. Ward, Richard G. Keil, Henrique O. Sawakuchi, Alan C. Da Cunha, Vania Neu, Daimio C. Brito, Diani F. Da Silva Less, Joel E. M. Diniz, Aline De Matos Valerio, Milton Kampel, Alex V. Krusche and Jeffrey E. Richey

*78 The Fate of Carbon in Sediments of the Xingu and Tapajós Clearwater Rivers, Eastern Amazon*

Dailson J. Bertassoli Jr., André O. Sawakuchi, Henrique O. Sawakuchi, Fabiano N. Pupim, Gelvam A. Hartmann, Michael M. McGlue, Cristiano M. Chiessi, Matthias Zabel, Enno Schefuß, Tatiana S. Pereira, Rudney A. Santos, Samantha B. Faustino, Paulo E. Oliveira and Denise C. Bicudo

*92 Flux of Dissolved and Particulate Low-Temperature Pyrogenic Carbon From Two High-Latitude Rivers Across the Spring Freshet Hydrograph* Allison N. Myers-Pigg, Patrick Louchouarn and Roman Teisserenc

*103 Impact of Wetland Decline on Decreasing Dissolved Organic Carbon Concentrations Along the Mississippi River Continuum* Shuiwang Duan, Yuxiang He, Sujay S. Kaushal, Thomas S. Bianchi, Nicholas D. Ward and Laodong Guo

#### CHAPTER 4

#### ESTUARIES AND OCEANS

*115 Microbially-Mediated Transformations of Estuarine Dissolved Organic Matter*

Patricia M. Medeiros, Michael Seidel, Scott M. Gifford, Ford Ballantyne, Thorsten Dittmar, William B. Whitman and Mary Ann Moran

*126 Composition and Transformation of Dissolved Organic Matter in the Baltic Sea*

Michael Seidel, Marcus Manecki, Daniel P. R. Herlemann, Barbara Deutsch, Detlef Schulz-Bull, Klaus Jürgens and Thorsten Dittmar


Nicolas E. Reyna, Amber K. Hardison and Zhanfei Liu


Eurico J. D'Sa and Hyun-cheol Kim

# Editorial: Integrative Research on Organic Matter Cycling across Aquatic Gradients

Nicholas D. Ward\*

*Marine Sciences Laboratory, Pacific Northwest National Laboratory, Sequim, WA, USA*

Keywords: carbon, cycling, aquatic, terrestrial, marine, continuum, gradients, organic

**Editorial on the Research Topic**

**Integrative Research on Organic Matter Cycling across Aquatic Gradients**

### INTRODUCTION

The interface between freshwater and marine ecosystems provides a unique setting to examine the evolution of biogeochemical components derived from the landscape, inland waters, estuaries, and the ocean across distinct physiochemical gradients. A diverse body of work exploring this research topic is highlighted here with the goal of integrating our understanding of how organic matter (OM) is transported and transformed along the terrestrial-aquatic continuum and sparking interdisciplinary discussions on future research needs. The movement of water ultimately controls the transport and transformation of geochemical components as they move from land to sea (Ward et al.), and as such, contributions to this research topic will be described within the context of the hydrological cycle, starting with rainfall.

#### TERRESTRIAL FLOW PATHS

#### Edited and reviewed by:

*Eric 'Pieter Achterberg, GEOMAR Helmholtz Centre for Ocean Research Kiel (HZ), Germany*

\*Correspondence:

*Nicholas D. Ward nicholas.ward@pnnl.gov*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *06 April 2017* Accepted: *20 April 2017* Published: *04 May 2017*

#### Citation:

*Ward ND (2017) Editorial: Integrative Research on Organic Matter Cycling across Aquatic Gradients. Front. Mar. Sci. 4:131. doi: 10.3389/fmars.2017.00131* Rainfall passes through terrestrial vegetation and soil layers prior to entering rivers and streams, translocating and transforming carbon in the process. Neu et al. examined the concentration and flux of dissolved organic and inorganic carbon (DOC and DIC, respectively) in each flow path (i.e., rainfall, throughfall, stemflow, soil solution, groundwater, overland flow, and stream flow) in an Amazonian transitional forest. The highest DOC concentrations and fluxes were observed in throughfall and rainfall, particularly during the first storms after the dry season. DOC concentrations were order(s) of magnitude lower in the soils and streams compared to aboveground flowpaths while DIC levels were generally higher, illustrating that rapid transformations begin occurring to DOC before even entering a river network (Neu et al.).

Stubbins et al. evaluated the dissolved organic matter (DOM) composition of throughfall and stemflow in the Southeastern USA based on ultrahigh resolution mass spectrometry and optical properties. DOM composition was primarily related to surface characteristics of the terrestrial vegetation (i.e., tree type and the abundance of epiphytes) and also to the presence of atmospherederived particles, which presumably accumulate during dry periods and are mobilized during the first rainfall events of a water year as observed by Neu et al.

Hernes et al.similarly assessed source inputs of DOM from the landscape to riversin a temperate watershed in northern California. Leached DOM was extensively transformed as it traveled through the subsurface as indicated by optical analyses and lignin phenol measurements. Hernes et al. further argued that interpretations of lignin biomarkers in water are obscured by complex source signatures and fractionation/transformation of DOM before reaching a main river channel. The take home point from these studies is that a considerable amount of "action" occurs before water and associated OM ever enters a river or stream.

#### RIVERS AND STREAMS

The composition and abundance of OM in river systems is not only linked to the landscape, as described above, but also to processes occurring in situ. For example, Duan et al. evaluated downstream variability of DOC concentrations in the Mississippi River over a ∼20 year time period. DOC concentrations consistently decreased from the headlands to the river mouth, which was attributed to an imbalance between in situ processing and DOM inputs from floodplains and the landscape. The downstream decrease in DOC concentrations in the Mississippi River is in contrast to the behavior of other large river systems, which was attributed to the historic loss of wetlands in the Mississippi basin (Duan et al.).

The lower reach of the Amazon River is an example of a floodplain-rich system in contrast to the Mississippi River. Sawakuchi et al. assessed CO<sup>2</sup> emissions from the lower Amazon River to the atmosphere, which, along with revised estimates for the upper/central Amazon basin, increased the global inland water CO<sup>2</sup> emission estimate by 43%. This study highlighted two major gaps in current global CO<sup>2</sup> data coverage—tidally-influenced lower rivers and near-shore coastal oceans.

Although the Amazon River is well-known for high rates of CO<sup>2</sup> evasion, Gagne-Maynard et al. evaluated the importance of primary production in the Amazon River mainstem, which has typically been assumed to be irrelevant. However, based on an oxygen mass balance, primary production was estimated to occur at roughly 50% the rate of microbial respiration despite limited light attenuation. Further, constraining this interplay between photosynthesis and respiration is an important step in understanding the underlying mechanisms for both CO<sup>2</sup> outgassing and OM accumulation in large rivers. For example, Bertassoli et al. observed high potential for OM burial since the mid-Holocene in the Amazon's two major lowland clearwater tributaries, the Tapajós and Xingu rivers. Sediment cores showed sharp changes between oxic and anoxic periods attributed to hydrologic patterns, illustrating the potential sensitivity of OM burial to variable conditions caused by factors such as climate change and hydropower operations.

The studies described in Section Terrestrial Flow Paths focused largely on the importance of vegetation on determining the composition and abundance of OM that is ultimately mobilized into river networks. OM derived from biomass burning (i.e., pyrogenic OM) is another important source of OM to the aquatic environment but is less well-studied than vascular plant biomarkers. Myers-Pigg et al. assessed the transport and fate of both dissolved and particulate pyrogenic OM (levoglucosan, specifically) in two Arctic rivers in order to determine the role of sorption and selective partioning on pyrogenic OM decomposition and/or storage. Although certain types of pyrogenic carbon such as levoglucosan have been shown to be bioreactive over short time scales, associations with the particulate phase may result in burial, meaning the signatures of these biomarkers could have potential utility as wildfire indicators in the sediment record (Myers-Pigg et al.). The studies described above collectively illustrate the importance of hydrology on controlling the export, storage, and transformation of OM in inland waters.

#### ESTUARIES AND OCEANS

Various studies examined the transformation of DOM along estuarine and marine gradients. In a southeastern USA marsh-dominated estuary, microbes exhibited a preferential decomposition of marine-derived DOM, resulting in a postincubation predominance of terrestrially-derived DOM, suggesting that this system is effective at exporting terrigenous DOM to the ocean (Medeiros et al.). Terrigenous DOM from sub-Arctic rivers was estimated to have a residence time of 5–7 years in the brackish Baltic Sea, with 50–67% being exported to the North Sea (Seidel et al.). Similarly, DOM in the temperate Delaware Bay estuary was observed to be relatively unaltered by photo and microbial oxidation on a molecular level along its salinity gradient likely due to high turbidity levels (Osterholz et al.). The relative stability of terrigenous DOM in these estuaries and shelf seas is perhaps indicative of the extensive decomposition that occurs in rivers prior to export or may also illustrate unfavorable environmental conditions for the decomposition of these molecules.

Short-term variability in fluorescent DOM (fDOM) concentrations were evaluated in high resolution throughout a year-long period in the subtropical Shark River estuary, which receives drainage inputs from the Everglades (Regier et al.). fDOM varied from 10 to 50% during tidal cycles, indicating rapid exchange with mangrove marshes and varied by more than 100% over multi-day periods due to both precipitation events and water management, illustrating the dynamic behavior of DOM sources, export, and transformation in estuaries. Storm events following long dry periods drastically altered the biogeochemistry of the subtropical Mission-Aransas estuary in Texas. The estuary became significantly less salty and warmer following storm events and large cyanobacteria blooms were observed concurrently with increased nutrient and carbon concentrations, demonstrating the potential role of storms on stimulating estuarine hypoxia (Reyna et al.).

Finally, the behavior of chromophoric DOM (CDOM), DOC, and chlorophyll was evaluated in the Southern Ocean from New Zealand to the western Ross Sea based on satellite remote sensing; all three parameters had the highest concentrations near New Zealand and consistently decreased toward the core of the Antarctic Circumpolar Current (ACC) at higher latitudes (D'Sa and Kim). Temperature and salinity gradients along this transect were primarily controlled by fronts associated with the ACC and the measured parameters were tightly coupled to these physical gradients illustrating the linkage between physical and biogeochemical processes.

The collection of work presented in this Research Topic illustrates the dynamic behavior of OM as it travels with water through the terrestrial biosphere, river networks, estuaries, and oceans. The fate of OM in the aquatic environment is closely tied to both its previous history and its present physical and biological surroundings. As such, it is essential to consider how OM has already evolved prior to and after being exported from any given location along the continuum.

## AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

## FUNDING

This work was supported by NSF DEB Grant #1256724, FAPESP Grant #08/58089-9, the Gordon and Betty Moore Foundation Marine Microbial Initiative, and the University of Florida Jon L. and Beverly A. Thompson Endowment.

## ACKNOWLEDGMENTS

I would like to thank all of the authors who contributed to the Frontiers in Marine Science Research Topic: Integrative Research on Organic Matter Cycling Across Aquatic Gradients, the reviewers who kindly dedicated their time to the peerreview process, the Frontiers editorial team, and the editors of the Research Topic: Thomas Bianchi, Richard Keil, Patricia Medeiros, Michael Seidel, and Carol Robinson. I would also like to thank the Gordon and Betty Moore Foundation Marine Microbial Initiative for their generous donation to support open access publication fees for the Research Topic.

**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Ward. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dissolved Organic and Inorganic Carbon Flow Paths in an Amazonian Transitional Forest

#### Vania Neu<sup>1</sup> \*, Nicholas D. Ward<sup>2</sup> \*, Alex V. Krusche<sup>3</sup> and Christopher Neill <sup>4</sup>

1 Instituto Socio Ambiental e dos Recursos Hidricos, Universidade Federal Rural da Amazônia, Belem, Brazil, <sup>2</sup> Department of Geological Sciences, University of Florida, Gainesville, FL, USA, <sup>3</sup> Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Piracicaba, Brazil, <sup>4</sup> The Ecosystems Center, Marine Biological Laboratory, Woods Hole, MA, USA

As a raindrop falls from the atmosphere, over vegetation and forest canopies, and enters soils and streams it experiences a dynamic exchange of carbon constituents with the surrounding environment. Understanding the magnitude and dynamics of carbon export in above and below ground flow paths is critical for constraining the influence of terrestrial and aquatic ecosystems on global carbon cycling. Here we examine the concentration and flux of dissolved organic and inorganic carbon (DOC and DIC) in rainfall, throughfall, stemflow, overland flow, soil solution, groundwater, and stream water in an Amazonian transitional forest located near the arc of deforestation. Rainfall was enriched in DOC by interactions with atmospheric particles derived from both biogenic and anthropogenic emissions, resulting in a flux of 82.3 kg C ha−<sup>1</sup> yr−<sup>1</sup> , which was the largest flux from each respective flow path. Forest throughfall, stemflow, and direct overland flow mobilized DOC from products of terrestrial primary production and decomposition. Net throughfall represented the second largest DOC flux (68.4 kg C ha−<sup>1</sup> yr−<sup>1</sup> ), whereas stemflow and overland flow only had a flux of 1.5 and 3.9 kg C ha−<sup>1</sup> yr−<sup>1</sup> , respectively. Much of the DOC in above ground flow paths was removed from solution as the rain percolated through soil layers due to both biological decomposition and sorption/desorption to mineral surfaces, resulting in low concentrations in stream and groundwater (2.6 ± 2.4 mg L−<sup>1</sup> and 1.45 ± 0.3 mg L−<sup>1</sup> , respectively) relative to throughfall (43.9 ± 5.2 mg L−<sup>1</sup> ) and stemflow (30.6 ± 2.7 mg L−<sup>1</sup> ). The flux of DIC in each respective flow path was generally lower than for DOC, and likely driven by atmospheric gas exchange and inputs from respiration and decomposition. DOC concentrations in above ground flow paths were highest during the first rainfall after a dry period, whereas DIC concentrations generally increased throughout the rainy season as soil moisture increased.

Keywords: tropical forest, carbon, organic, inorganic, flow path, flux, rainfall, terrestrial

## INTRODUCTION

Carbon derived from the terrestrial biosphere is mobilized by rainfall into soils, streams, and groundwater through numerous pathways. Before hitting a forest canopy or uncovered ground, a raindrop interacts with organic carbon (OC) rich particles in the atmosphere (Artaxo et al., 1988; Talbot et al., 1990). This interaction results in the enhancement of dissolved organic carbon (DOC) concentrations in precipitation (Andreae et al., 1990; Artaxo et al., 2005; Waterloo et al., 2006; Germer et al., 2007). Precipitation drives a flux of OC from the atmosphere to the landscape

#### Edited by:

Selvaraj Kandasamy, Xiamen University, China

#### Reviewed by:

Wei-Dong Zhai, Shandong University, China K. Suresh Kumar, University of Allahabad, India

\*Correspondence:

Vania Neu vania.neu@ufra.edu.br; Nicholas D. Ward nickdward@gmail.com

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 14 April 2016 Accepted: 16 June 2016 Published: 28 June 2016

#### Citation:

Neu V, Ward ND, Krusche AV and Neill C (2016) Dissolved Organic and Inorganic Carbon Flow Paths in an Amazonian Transitional Forest. Front. Mar. Sci. 3:114. doi: 10.3389/fmars.2016.00114 via dry deposition, atmosphere washing, and incorporation of aerosols in rain drop condensation nuclei (Artaxo et al., 2005). Rainfall receives additional inputs of carbon as it travels through forest canopies (e.g., throughfall) and down trunks and stems (e.g., stem flow); litterfall also provides a significant source of fresh OC as rainfall percolates through subsurface soil layers into streams and groundwater (Qualls et al., 2002).

Much of the OC flux described above is removed from the water while traveling through soil layers as a result of both sorptive and biodegradative processes, leaving stream and groundwater with low DOC concentrations relative to above ground flow paths (Qualls and Haines, 1992; Schmidt et al., 2011). Remineralization rates of OC in soils are high and may be enhanced by global climate change (Feng et al., 2008). OC derived from terrestrial primary production and dissolved inorganic carbon (DIC) derived from root and soil respiration is mobilized from soil layers into streams and groundwater reservoirs. Rivers and streams are typically supersaturated with respect to CO<sup>2</sup> as a result of direct inputs of DIC and remineralization of terrestrially-derived OC (Raymond et al., 1997; Richey et al., 2002; Mayorga et al., 2005; Ward et al., 2013). Inland waters emit 2.1 Pg C yr−<sup>1</sup> , globally, of which 70% comes from the tropics and 25% from the Amazon River, alone (Raymond et al., 2013).

Carbon fluxes to both the atmosphere and ocean are generally higher in tropical regions relative to temperate and high-latitude counterparts (Aufdenkampe et al., 2011). Covering a region of over 6 million km<sup>2</sup> , the highly productive Amazon basin plays a significant role in the global carbon cycle. The Amazon rainforest is responsible for roughly 10% of global terrestrial primary production and the Amazon River represents roughly 20% of the world's freshwater flux (Field et al., 1998; Malhi et al., 2008). The extensive basin contains regions with relatively little human perturbation, and regions highly impacted by deforestation and burning. The Amazon rainforest naturally produces large amounts of OC-rich aerosols through biogenic emissions that are eventually entrained in rainfall (Talbot et al., 1990). The size and morphology of these particles is highly diverse with origins from plant sources such as pollen grains, fungi, algae, and plant fragments (Artaxo et al., 1988).

Deforestation and biomass burning, along with other anthropogenic emissions, can also add aerosols to the atmosphere. For example, aerosol concentrations in Amazonian regions situated in the arc of deforestation have increased during recent years as a result in enhanced anthropogenic activities (Artaxo et al., 2005). Aerosol concentrations were generally highest during the dry season in these regions due to large amounts of grass and forest burning in addition to dust derived from agricultural and exposed soil sites (Artaxo et al., 2005). The majority of the carbon from these aerosols is transported from the atmosphere to the land at the beginning of the wet season, with rainfall DOC concentrations decreasing with additional precipitation (Filoso et al., 1999; Johnson et al., 2006a; Germer et al., 2007).

Throughfall is an important component of forest carbon cycling. The primary carbon input pathway is canopy washing. In the Amazon rainforest, throughfall has been observed to represent roughly 75–90% of total rainfall (Lloyd and Marques, 1988; Filoso et al., 1999; Ferreira et al., 2005). Stemflow generally has even higher DOC concentrations than rainfall and throughflow, but the total flux is low since stemflow only represents 2–8% of total rainfall in the Amazon (Jordan, 1978; Lloyd and Marques, 1988). Overland flow occurs either when rainfall occurs more rapidly than soil infiltration (Horton, 1933) or when soils are completely saturated (Linsley et al., 1975). Overland flow generally has greater DOC concentrations than throughfall and is transported rapidly. However, as with stemflow, overland flow is generally low compared to other flow paths in preserved forests, resulting in a relatively small overland DOC flux (Johnson et al., 2006a; Waterloo et al., 2006). In surface soils and organic matter layers DOC concentrations are generally high as a result of initial enrichment of throughfall and leaching of decomposing organic matter; soil solution concentrations typically decrease with soil depth as a result of mineral sorption and OC remineralization (McClain et al., 1997; Liu and Sheu, 2003). The removal of OC in soils leaves stream and ground water with low DOC concentrations relative to the preceding flow paths (Qualls et al., 2002; Liu and Sheu, 2003; Waterloo et al., 2006; Johnson et al., 2006a).

The objective of the present study is to determine the concentration and flux of DOC and DIC in each respective flow path: rainfall, throughfall, stemflow, overlandflow, soil solution, groundwater, and streamwater in an Amazonian forest. Water samples were collected monthly between February 2007 and January 2008 using a series of collectors designed for each flow path. The forest studied here is transitional between tropical rainforest and savanna, and is situated in the arc of deforestation in Brazil (**Figure 1**). In recent decades this region has lost large amounts of forested area due to deforestation and land use change.

## METHODS

### Study Site

The study was conducted at Tanguro Ranch in the southern Amazonian state of Mato Grosso, Brazil. Tanguro Ranch (12◦ 53′ S, 52◦ 21′ W) lies southeast of the Xingu Indigenous Park in the municipalities of Canarana and Querência at an elevation of ∼350 m above sea level in the upper region of the Xingu drainage basin between the Tanguro and Darro Rivers (**Figure 1**). Topography of the region consists of wide plains and gentle slopes to the waterways in a landscape with little other pronounced relief. It lies on a portion of the Brazilian Shield underlain by Precambrian Gneisses of the Xingu Complex. Soils consist of Oxisols (Haplustox/Latossolo vermelho-amarelo distrófico) on the broad, flat uplands away from the rivers. These soils have a water table depth of 12−15 m and no impeding layers. Throughout the year the water table depth varies by 0.27 m in the lowland and 1.07 m in the plateau. Wetter, and occasionally inundated, soils (Gleissols) of similar origin are found along the river margins (Projeto Radambrasil, 1981).

The pre-agricultural vegetation of the Tanguro region is evergreen tropical forest (Walter, 1979; Ivanauskas et al., 2004) typical of the transition from Cerrado to the central Amazon rainforest. The vegetation of this transition zone has lower species

FIGURE 1 | Study site. Samples were collected in the Tanguro Ranch watershed in the Xingu drainage basin between the Tanguro and Darro Rivers.

diversity, smaller trees, and lower canopy heights compared to many previously studied Amazonian forests (Ivanauskas et al., 2004). The climate at Tanguro is humid tropical. Average annual precipitation from 1987 to 2007 was 1905 ± 271 mm and is strongly seasonal, with <5% of total annual precipitation falling between May and September (Vianello and Alves, 2002). June, July and August are the three driest months with typically less than 10 mm rain per month. Mean annual temperature is 25◦C. During this study, hourly rainfall and rain intensity were recorded at a weather station 27 km south of the study boundaries. Daily rainfall was also measured in two collectors 2 and 5 km from the study site.

The completely forested watershed studied here covers an area of 13.20 km<sup>2</sup> . The watershed is drained by a perennial first-order stream. Broad, flat plateaus typical of the region dominate the majority of the watershed's topography, which transitions into a distinct shoulder hillslope zone and narrower/flatter lowland, near-stream zone. The watershed boundaries were delineated from Shuttle Mission Radar Topography data and validated with field measurements with a high-precision differential GPS (**Figure 1**). This watershed was divided into three parcels and nine sub-parcels. Each parcel lies along a topographic gradient, with three sub-parcels (lowland, hillside and plateau).

#### Sample Collection

Water samples were collected to assess the organic and inorganic carbon content of rainfall, throughfall, stemflow, overland flow, soil solution, groundwater, and streamwater between February 2007 and January 2008. Soil solution samples were collected with tension lysimeters on a monthly basis, however, soils were too dry to extract water from June to November. The lysimeter was rinsed with distilled water before use, and after installation the first collection was discarded. Tension lysimeters were installed at depths of 10, 50, 100, and 200 cm, in each sub-parcel (36 collectors total). Soil solution was collected from the lysimeters after setting the vacuum for 24 h at 0.6 bars. Samples were stored in acid-washed 250 mL glass bottles and sealed with silicone stoppers.

Overland flow samples were collected using 27 collectors installed along the nine sub-parcels (three collectors in each). The collectors consisted of a PVC pipe (200 cm wide with 15 cm diameter) that was closed in the two extremities with PVC caps. In one extremity the cap had a small opening connected to a reservoir (10 L capacity) to store the sample. The collectors have a transversal opening (3 × 190 cm), that allows the collection of water originating from overland flow. The collectors were inserted horizontally into the soil such that the opening of the collector was near the soil surface. Below the soil surface (3 cm), a small horizontal cut in the soil parallel to the collector was made in order to install a polystyrene plate (10 × 190 cm). This plate was fixed into the collector, and into the soil, making a connection between the soil and collector to permit the overland flow to enter the collector. The captivating area of each collector was delimited to 6 m<sup>2</sup> plots with wooden boards inserted into the soil. This was done to prevent the collection of water from areas uphill and lateral of each plot. The overland flow collectors were prepared for collection every 15 days by connecting the collector to the reservoir. Samples were collected from the reservoirs 1 day after each sampled rain event. The collectors were set to bypass the reservoir in between samplings, automatically discarding any water. During each sampling the volume of water collected in each collector was measured using a digital dynamometer (Insthutherm, Model DD-200).

Throughfall samples were collected biweekly with 27 collectors installed along the nine sub-parcels (three collectors in each). Collectors were dispersed randomly to avoid bias due to canopy density. The collector is a suspended channel made with a PVC pipe (150 cm wide with 15 cm diameter) that is closed on one end with PVC caps. The other end had a funnel connected to a reservoir (20 L capacity) to store the sample. The funnel was covered with plastic so as not to increase the capture area of water. The channel had a transversal opening (7 × 140 cm) in the direction of the tree's canopy. The captivating area of each collector was 980 cm<sup>2</sup> . The collector was suspended 1 m above the soil surface. Rainfall was collected in the same type of collectors installed in triplicate in an open area.

Groundwater was collected every 15 days from nine wells in the watershed during the rainy season and monthly during the dry season. Screened groundwater wells made of 10 cm diameter PVC pipes were installed by hand at different depths depending on their position along the topographic gradient. Transversal cuts were made in the sides of the pipe 1.5 m from the bottom so that groundwater could flow in and out of the well. Lowland wells were between 1.69 and 2.14 m deep, hill slope wells were 4.84 to 6.25 m deep, and plateau wells were 7.07 to 9.65 m deep. Groundwater was collected with a manual collector, made with a PVC pipe (82 cm wide and 5 cm diameter) and a movable glass bottom. Roughly 3 L of water was removed prior to sample collection to insure that water that had most recently arrived at the well was collected.

Stemflow was collected along the catchment in three randomly located 10 × 10 m plots along the topographic gradient. In each plot collectors (48 total) were installed on trees with a diameter ≥5 cm. Polystyrene plates (2.5 cm wide) were fitted around the tree trunks in a downward spiral connected to a 20 L reservoir, starting 1.5 m above the forest floor, similar to the collectors described by Likens and Eaton (1970). To assess possible contamination from the collectors, plastic components of the collectors were immersed in deionized water for 3 days and the DOC was measured. There was no detectable organic contamination from the collectors.

## Sample Analysis

The pH and conductivity of water samples was determined with Orion (model 250A) and Amber Science (model 2052) probes, respectively (Supplementary Tables 1–7). For DOC samples 30 mL of water was filtered through 0.7 µm pore size combusted GFF filters (Whatman) into combusted glass vials and preserved in the field with 75 mL of 6N HCl. DOC concentrations were measured on a Shimadzu total carbon analyzer (Model TOC-VCPH). For DIC samples 60 ml of water was filtered through 0.47 µm cellulose acetate membrane filters (Millipore) into acid-washed 60 mL HDPE bottles with no headspace to avoid degassing and preserved with thymol (100 mg/1000 ml of solution). DIC samples were analyzed for total organic carbon before acidification on a Shimadzu total carbon analyzer (Model TOC-VCPH). DIC samples were then acidified using 6N HCl and sparged to remove gas and reanalyzed for total organic carbon. DIC was calculated as the difference between non-acidified and acidified/sparged samples. This method may slightly underestimate DIC concentrations due to degassing during filtration. However, this method has been compared to DIC calculations from CO<sup>2</sup> values and the variation is typically minimal, whereas alkalinity titrations result in high amounts of error in DOC-rich acidic waters (Neu et al., 2011). Samples were stored in the dark at room temperature for roughly 1 week prior to analysis. Samples could not be kept cold since there was no electricity available at the field site.

#### Rainfall and Stream Discharge

Hourly rainfall and rain intensity was recorded at the Tanguro Ranch weather station 27 km south of the sampling location. Daily rainfall was also recorded in two collectors roughly 2 km from the watershed sampling point. Stream stage was recorded every 15 min from May to August 2007 with an ISCO 6200 automatic sampler equipped with a pressure transducer module. From August 2007 to August 2008 stream stage was recorded hourly with an Onset water level logger. A rating curve was developed by periodic measurements of stream discharge with a Global Water FP-101 current meter across a range of stream flows. Hourly stream flows for January to May 2007 were estimated from daily rainfall and rainfall-stream flow relationships from the same location derived from the 16-month stream flow record that began in May.

#### Flux Estimates

We estimated total annual DOC and DIC fluxes in throughfall based on our record of rainfall and intensity from Tanguro Ranch, measured DOC and DIC concentrations from the sampled rain events, extrapolated DOC and DIC concentrations for non-sampled events, and an estimate of canopy interception for a tropical forest. Precipitation during sampled events (n = 10) ranged from 0 to 115 mm and unsampled events (n = 51) ranged from 0 to 75 mm. We assumed that throughfall was on average 85% of precipitation (Lloyd and Marques, 1988; Filoso et al., 1999; Ferreira et al., 2005) to calculate total througfall volume duirng unsampled events. Average DOC and DIC concentrations from sampled events were used to estimate concentrations during unsampled events with similar amounts of precipitiation.

We estimated water volumes for stemflow during unsampled events from relationships of measured volumes (in mm) between throughfall and stemflow (**Figure 2A**):

$$\text{SF} = 6.35 \times 10^{-5} \text{ (TF)}^2 + 0.0063 \text{ (TF)} - 0.0252; \text{ r}^2 = 0.86 \text{ (l)} \tag{1}$$

where SF is stemflow and TF is throughfall. Likewise, overland flow (OF) was calculated based on the following relationship between overland flow and throughlow (**Figure 2B**):

$$\text{OF} = 0.0116 \text{(TF)} + 0.0383; \text{ r}^2 = 0.84 \tag{2}$$

We then used these calculated volumes and mothly mean DOC concentrations in stemflow and overland flow from the nearest sampled event to calculate DOC fluxes for each unsampled event. We then summed DOC fluxes for all sampled and unsampled events to obtain an annual flux.

#### Statistical Analysis

The results were submitted to analysis of variance and subsequently Tukey median comparison test for the variables with statistically significant differences, using the GLM (General Linear Models) procedure, using SAS 9.1 software (Statistical Analysis System). The tests were considered significantly different for p < 0.05.

FIGURE 2 | Correlation between throughfall, stemflow, and overland flow. The correlation between throughfall and stemflow (A) and throughfall and overland flow (B) was used to calculate stemflow and overland flow for unsampled events.

## RESULTS

#### Precipitation and Stream Discharge

Precipitation in 2007 totaled 1641 mm and was lower than the 1987–2007 mean of 1905 mm. The greatest difference between measured and historical mean precipitation occurred in March 2007 when precipitation totaled 34 mm in contrast to the longterm monthly average of 250 mm (**Figure 3**). Total precipitation from February 2007 to January 2008 was 1829 mm, of which roughly 476 mm occurred during times of sample collection. Precipitation and stream discharge were maximal from October to April. Annual average stream discharge was 17.9 ± 3.6 L s−<sup>1</sup> (**Table 1**). Stream discharge throughout the year was primarily dominated by base flow (**Figure 3**).

#### Dissolved Organic Carbon

Concentrations of DOC showed the highest seasonal variation in above-ground flow paths (e.g., rainfall, throughfall, stemflow, and overland flow). The concentration of DOC in rainfall was highest at the beginning of the rainy season, reaching a maximum of 15.6 ± 0.3 mg C L−<sup>1</sup> during the October 18, 2007 event, and declined throughout the rainy season, reaching a minimum of 3.3 ±

0.2 mg C L−<sup>1</sup> (**Figure 4**). The annual average DOC concentration in rainfall was 6.7 ± 3.7 mg C L−<sup>1</sup> (**Table 1**). The concentration of DOC in throughfall was highest at the end of the dry season during the September 2007 rain event (126 ± 68 mg C L−<sup>1</sup> ), and declined throughout the rainy season. In April 2007, after a brief dry period the throughfall DOC concentrations increased from 35.7 ± 20.0 mg C L−<sup>1</sup> during the March 2007 sampling to 77.5 ± 28.7 mg C L−<sup>1</sup> (**Figure 4**). The annual average throughfall DOC concentration for lowland, hillside and plateau sub-parcels was 47.8 ± 6.5, 41.9 ± 4.6, and 42 ± 4.4 mg C L−<sup>1</sup> , respectively, showing minimal variation between the different topographic regimes (p = 0.66).

Stemflow DOC concentrations varied both seasonally and as a function of tree diversity and the stem surface. In particular, stem rugosity and the presence or lack of epiphytes had a large influence on stemflow DOC in this study. The following species, characterized by a rugose bark and/or lichens on the stem, represent the majority of the trees sampled for stemflow and provided the greatest amount of DOC enrichment: Amaioua guianensis (bark with fissures and lichens), Trattinickia burseraefolia (bark with lichens), Hirtelar bicornis, Trattinickia glaziovii, Xylopia amazonica, Miconia punctate, Pseudolmedia murure, and Miconia pyrifolia (rugose bark and lichens on the stem). The following species, characterized by a smooth bark, were a minor fraction of the trees sampled and generally had low stemflow DOC concentrations: Licania egleri, Protium guianense, Protium unifoliolatum, Vochysia vismifolia, and Envira cana. As with rainfall DOC, stemflow DOC concentrations were highest at the beginning of the rainy season, reaching a maximum of 87.7 ± 53 mg C L−<sup>1</sup> during the October 16, 2007 sampling, and steadily decreased throughout the rainy season (**Figure 4**). Annual average stemflow DOC concentrations in lowland, hillside and


TABLE 1 | Average concentrations and annual fluxes of DOC, DIC, and water across the topographic gradient.

<sup>a</sup>Water fluxes measured during the sampling period from February 15, 2007 to January 25, 2008.

<sup>b</sup>Values represent the gross throughfall DOC/DIC contribution (e.g., the sum of rainfall and net throughfall).

<sup>c</sup>Values represent the average of 10, 50, 100, and 200 cm lysimeter depths.

plateau sub-parcels were 31.7 ± 3.09, 32.5 ± 3.3, and 27.6 ± 1.6 mg C L−<sup>1</sup> , respectively. Whereas lowland and hillside parcels had similar stemflow DOC concentrations, concentrations in plateau parcels were generally lower (p < 0.05).

Overland flow did not occur in events with low precipitation (e.g., March and April 2007 samplings, with 5 and 6 mm of precipitation, respectively). Lowland parcels had the highest overland flow volume compared to the oxisoils of the hillside and plateau parcels. Compared to the other above-ground flow paths, overland flow had the highest DOC concentrations. As with other above-ground flow paths, overland flow DOC concentrations were highest at the beginning of the rainy season, reaching a maximum of 160 ± 110 mg C L−<sup>1</sup> during the October 16, 2007 sampling. Annual average overland flow DOC concentrations in lowland, hillside and plateau sub-parcels were 57.2 ± 10.6, 39.6 ± 7.0, and 40.9 ± 5.7 mg C L−<sup>1</sup> , respectively. Concentrations did not vary significantly along the elevation gradient (p = 0.25). Although overland flow had the greatest DOC concentrations, this flow path represented a small fraction of the total water flow.

In contrast to the high DOC concentrations observed aboveground, below-ground DOC concentrations were generally lower. Soil solution DOC concentrations were highest at 10 cm depth, with an average of 17.7 ± 2.0 mg C L−<sup>1</sup> and decreased to 11.9 ± 1.0 mg L−<sup>1</sup> , 16.4 ± 1.5 mg L−<sup>1</sup> , and 14.9 ± 1.3 mg C L−<sup>1</sup> at depths of 50, 100, and 200 cm depths, respectively. Differences with depth were statistically significant (p < 0.05), whereas there was little variation along the topographic gradient (p = 0.33). It was not possible to extract soil solution samples during the long dry season and at the beginning of the rainy season with the methodology used here, however, as with the above-ground flow paths, soil solution DOC concentrations appeared to peak and decrease throughout the rainy season (**Figure 4**). The highest observed soil solution DOC concentrations occurred during the December 2007 event, with a value of 24.6 ± 17 mg C L−<sup>1</sup> . Litterfall typically accumulates and decomposes above ground during the dry season. Above ground flow paths transport DOC rich waters to the sub-surface, particularly during the first several periods of rainfall in a season.

Groundwater generally had the lowest concentration of DOC, and contrary to other flow paths, the highest values were observed during the dry season. The maximum groundwater DOC concentration was 4.9 ± 4.0 mg C L−<sup>1</sup> during August 2007. Annual average concentrations of groundwater DOC in lowland, hillside, and plateau sub-parcels were 2.0 ± 0.5, 1.4 ± 0.3, and 1.0 ± 0.2 mg C L−<sup>1</sup> , respectively, with little variation across the topographic gradient (**Table 1**; p = 0.19). However, the lowlands had the highest carbon stock in the soil profile (119 Mg C ha−<sup>1</sup> at 140 cm depth) compared to hillside (119 Mg C ha−<sup>1</sup> ) and plateau (105 Mg C ha−<sup>1</sup> ) sub-parcels. Stream flow also had low DOC concentrations relative to the different flow paths, with an annual average concentration of 2.9 ± 3.5 mg C L−<sup>1</sup> (**Table 1**). Stream DOC concentrations reached a maximum of 15.8 ± 0.2 mg L−<sup>1</sup>

during the February 16, 2007 sampling when discharge was high, and decreased throughout the rainy season.

Based on the above measurements, the annual DOC flux in the watershed was estimated to be 82.4 ± 18.4 kg C ha−<sup>1</sup> yr−<sup>1</sup> for rainfall (**Table 1**). The gross flux of DOC for throughfall was 142.6 ± 75.6 kg C ha−<sup>1</sup> yr−<sup>1</sup> , or 60.2 ± 77.8 kg C ha−<sup>1</sup> yr−<sup>1</sup> for net throughfall (i.e., the difference between gross throughfall and rainfall fluxes). DOC fluxes for stemflow, overland flow, and stream flow were 1.5 ± 0.8, 3.9 ± 1.6 and 1.7 ± 0.2 kg C ha−<sup>1</sup> yr−<sup>1</sup> , respectively (**Table 1**). It is important to note that the fluxes estimated for overland flow do not represent the actual flux to the stream, per se, since it is possible that the water collected with our methods could have potentially infiltrated less saturated soils further downslope before reaching the stream.

#### Dissolved Inorganic Carbon

Above-ground flow paths (e.g., rainfall, throughfall, stemflow, and overland flow) had relatively low DIC concentrations in contrast to the high DOC concentrations observed in these flow paths. The concentration of DIC in rainfall reached a maximum of 1.5 mg C L−<sup>1</sup> during the February 28, 2007 sampling (**Figure 5**), with an annual average of 0.8 ± 0.7 mg C L −1 (**Table 1**). Throughfall DIC concentrations were generally highest near the end of the rainy season, with a maximum value of 4.5 ± 2.4 mg C L−<sup>1</sup> during the April 6, 2007 sampling period. The annual average throughfall DIC concentration was 2.5 ± 2.3 mg C L−<sup>1</sup> . Stemflow DIC concentrations increased at the beginning of the rainy period, reaching a maximum of 6.7 ± 3.2 mg C L <sup>−</sup><sup>1</sup> during the October 16, 2007 sampling period. The annual average DIC concentration in stemflow was 1.5 ± 1.1 mg C L−<sup>1</sup> . Overland flow DIC concentrations reached a maximum of 2.4 ± 0.3 mg C L−<sup>1</sup> during the February 6, 2007 sampling period, and had annual average of 1.3 ± 1.3 mg C L−<sup>1</sup> (**Table 1**). The concentration of DIC in each of these above-ground flow paths did not significantly vary between lowland, hillside and plateau sub-parcels.

In contrast to observations of above-ground flow paths, below-ground flow paths had relatively high DIC concentrations that were above atmospheric CO<sup>2</sup> saturation levels. Soil solution DIC generally increased with soil depth, indicating CO<sup>2</sup> accumulation along the profile. Soil solution DIC did not vary significantly among the different sites (p = 0.33). The annual average soil solution DIC concentration was 3.2 ± 1.2 mg C L−<sup>1</sup> at 10 cm depth, 2.9 ± 0.5 mg C L−<sup>1</sup> at 50 cm depth, 6.3 ± 1.1 mg C L−<sup>1</sup> at 100 cm depth, and 11.9 ± 1.6 mg C L−<sup>1</sup> at 200 cm depth. Soil solution DIC concentrations were maximal during the February 16, 2007 sampling period with a value of 14.2 ± 13.9 mg C L−<sup>1</sup> . In groundwater the annual average concentration of DIC in lowland, hillside and plateau sub-parcels was 2.7 ± 0.3, 2.7 ± 0.3, and 3.7 ± 0.4 mg C L−<sup>1</sup> , respectively, differing along the topographic gradient, with the highest concentrations observed in plateau sub-parcels. Groundwater DIC concentrations reached a maximum of 8.7 ± 2.1 mg C L−<sup>1</sup> during the March 13, 2007 sampling period (**Figure 5**). The annual average DIC concentration in stream flow was 1.6 ± 1.0 mg C L−<sup>1</sup> , with a maximum of 6.5 mg L−<sup>1</sup> observed during the March 13, 2007 sampling period.

The annual flux of DIC through above-ground flow paths was significantly lower than DOC fluxes (p < 0.05). Based on the above measurements, the annual DIC flux in the watershed was estimated to be 38.8 ± 4.4 kg C ha−<sup>1</sup> yr−<sup>1</sup> for rainfall. The DIC flux for gross throughfall was 16.6 ± 8.2 kg C ha−<sup>1</sup> yr−<sup>1</sup> resulting in a net throughfall flux of 22.2 ± 9.3 kg C ha−<sup>1</sup> yr−<sup>1</sup> . Stemflow, overland flow, and stream flow had DIC fluxes 0.6 ± 0.1, 0.1 ± 0.1, and 0.9 ± 0.2 kg C ha−<sup>1</sup> yr−<sup>1</sup> , respectively (**Table 1**).

## DISCUSSION

#### Dissolved Organic Carbon

The concentration of DOC in rainfall was highest during the first large rainfall events after a long dry period and decreased throughout the wet season. This phenomenon is most likely a result of the accumulation of carbon-rich particulate material in the atmosphere during periods with minimal rainfall. This particulate material can be "washed" from the atmosphere during rainfall events (Greenfield, 1957) and a fraction of the particulate organic matter is solubilized into the dissolved phase in raindrops (Monteith et al., 2007). The atmospheric particulate material can be derived from a variety of natural and anthropogenic origins including aeolian transport, agricultural activities, and biomass burning (Ketseridis et al., 1976; Simoneit and Elias, 2000; Pio et al., 2001; Snyder et al., 2009). As with similar studies, our results indicate that the source of rainfall DOC that accumulates during dry periods is depleted during subsequent rainfall events, resulting in a decrease in rainfall DOC concentrations throughout the wet season (Filoso et al., 1999; Johnson et al., 2006a; Germer et al., 2007). Rainfall DOC fluxes in other regions studied in the Central Amazon were generally lower than our estimates. For example, rainfall fluxes in the Rio Negro region were roughly three times lower than the fluxes measured here (**Table 2**). Compared to the Xingu basin studied here, the Central Amazon generally experiences greater annual precipitation and is less impacted by deforestation and land use change (Andreae et al., 1990; Filoso et al., 1999; Waterloo et al., 2006). Rainfall DOC fluxes exceeding estimates made here were observed in other Amazonian regions that are highly influenced by biomass burning and land use alterations. For example a rainfall DOC flux of 106.5 kg ha−<sup>1</sup> yr−<sup>1</sup> was observed in Rondonia, Brazil, a site that is strongly affected by burning and land use alteration (**Table 2**; Artaxo et al., 2005; Germer et al., 2007).

As rainfall drains through the forest canopy, the throughfall becomes enhanced with DOC derived from leaf and plant surfaces, surficial decomposition processes, and animal excrement (Parker, 1983; Levia and Frost, 2003). Similar to rainfall, the enrichment of throughfall DOC was greatest during the first large rain events following a long dry period (**Figure 4**). Throughfall DOC concentrations dropped after subsequent rain events. Similar to the accumulation of carbon-rich particles in the atmosphere, organic carbon can accumulate on surfaces of the forest canopy during dry periods and be subsequently mobilized by throughfall during rain events. We not only observed enhanced throughfall DOC concentrations after a long dry season, but also after a month of dry weather from

groundwater, and stream water during rain events from February 2007 to January 2008.


TABLE 2 | Annual DOC fluxes for rainfall, throughfall, net throughfall, stemflow, and overland flow in the Amazon basin.

March to April 2007 (**Figures 2**, **3**). These results illustrate the importance of atmospheric deposition and forest metabolism on throughfall organic carbon content. Rainfall DOC, on the other hand, was elevated during the March 2007 sampling, and once again depleted by the April 2007 samplings, illustrating that the accumulation and solubilization of atmospheric particulates with rainfall may, in some cases, occur more quickly than the accumulation of OC in the forest canopy and mobilization by throughfall. This is likely because throughfall DOC originates from a range of sources including atmospheric particles and surficial processes and characteristics. A similar magnitude of DOC in throughfall (96.1 ± 68.0 mg C L−<sup>1</sup> ) was observed during the first rain after a long dry period in other Amazonian forests (Johnson et al., 2006a).

Stemflow and overland flow both had seasonal trends similar to throughfall (i.e., maximal DOC after the first rain event following a dry period). Stemflow generally represents a relatively minor flux of carbon. For example, here we measured an average DOC flux in stem flow of only 1.5 ± 1.1 kg C ha−<sup>1</sup> yr−<sup>1</sup> , roughly an order of magnitude less than the throughflow and rainfall fluxes (**Table 1**; **Figure 6**). Stemflow is primarily controlled by tree diameter, bark rugosity, epiphyte presence, and the forest ground flora biodiversity (Dalva and Moore, 1991; Liu and Sheu, 2003; Wang et al., 2004; Johnson and Lehmann, 2006). For example, forests with extensive palm undergrowth generally have increased stemflow due to enhanced water collection and dispersal by the large palm leaves (Lloyd and Marques, 1988; Germer et al., 2007). In this study, stemflow was largely influenced by bark rugosity and the presence of epiphytes, with much higher stemflow DOC concentrations observed in trees with rugose bark and lichen coverage than trees with a smooth bark. Surface complexities such as these can allow enhanced accumulation of organic matter both from production and decomposition processes.

Lowland and hillside sub-parcels generally contained greater coverage of stem epiphytes and higher stemflow DOC concentrations than plateau sites. The lowland and hillside parcels were in close proximity to the stream, resulting in enhanced air humidity and epiphyte production, whereas plateau sites were more exposed to wind and further from the stream. Although stemflow DOC concentrations were higher in lowland and hillside parcels, the DOC flux was greatest in plateau sites, which had a larger proportion of trees with a diameter at breast height >5 cm than lowland and hillside parcels (**Table 1**). The larger diameter trees provide a greater surface area for water to flow and organic carbon to solubilize (Dalva and Moore, 1991). Similar to our observations that stemflow represents a small DOC flux relative to total rainfall, other studies in the Amazon have shown that stemflow represents from roughly 2% (Lloyd and Marques, 1988) to 7–8% (Jordan, 1978) of the total rainfall. Although the stemflow DOC concentrations measured here and elsewhere were relatively high, the volume of water comprising stem flow was low, resulting in a relatively small DOC flux (Dalva and Moore, 1991; Hinton et al., 1998).

Overland flow was generally low throughout the study period. The forest soils were characterized by high clay content and aggregate stability, allowing a high degree of infiltration (Brandão, 2006). There was no overland flow present during rain events with less than 6 mm of precipitation. Overland flow was generally highest in the lowland sub-parcels since the riparian zone was primarily composed of saturated gleysols. These soils are generally slow draining and sloped toward the stream channel (Deckers et al., 1998). The highest concentrations and fluxes of DOC in overland flow were observed in the lowland sites (**Table 1**), most likely because, relative to plateau and hillside sites, higher levels of soil saturation allow a greater volume of water to be in contact with litter layers. The overland DOC flux from plateau and hillside sites, on the other hand, was low relative to other flow paths due to the soil type and relief characteristics. Overland flow represents a rapid pathway for carbon transport, and often contains the highest concentration of DOC relative to other flowpaths in regions with highly saturated soils, however the overland DOC flux is generally low because overland flow represents a relatively small fraction of total water flow (Johnson et al., 2006a; **Table 1**). As with other above ground flow paths, overland flow DOC concentrations were highest during the first rainfall following a long dry period (**Figure 4**).

A large amount of the DOC that enters soils as throughfall/stemflow is removed as it travels through soils into streams (Qualls et al., 2002; Liu and Sheu, 2003). For example, in the temperate Appalachian mountains, roughly 95% of the DOC in throughfall was lost in soils prior to entering streams; roughly 10–33% of this DOC loss was attributed to biodegradation, while the remaining DOC was hypothesized to be removed by mineral sorption (Qualls and Haines, 1992). In this study, soil surface layers generally had high DOC concentrations, which decreased with soil depth. Groundwater had much lower DOC concentrations than the preceding flow paths during the rainy season, similar to other observations in the

Ranch watershed.

Amazon basin (Johnson et al., 2006a). Contrary to above-ground flow paths, groundwater DOC concentrations were highest during dry periods (**Figure 4**), indicating that groundwater DOC is primarily mobilized by baseflow and, perhaps, diluted by rapid rainfall. As rainfall percolates through subsoil layers, sorption of molecules onto iron and aluminum-rich mineral surfaces can remove a large fraction of the DOC. Sorption generally occurs rapidly relative to decomposition processes (McClain et al., 1997; Qualls et al., 2002; Liu and Sheu, 2003; Johnson et al., 2006a). Microbial degradation of OC in soils is another important pathway for DOC loss, and has potential to be enhanced by a warming climate (Feng et al., 2008). An emerging view is that the degree of OC decomposition in litterfall and soils is largely a component of environmental and biological conditions rather than intrinsic molecular properties (Schmidt et al., 2011).

Concentrations of DOC in the stream were a similar magnitude as in groundwater, on average, but increased during heavy rainfall after a dry period (e.g., February 2007). This is similar to other observations of DOC concentrations in temperate and tropical settings made during brief storm periods. For example, high frequency measurements have illustrated a strong correlation between stream discharge and DOC concentrations during rapid rain events (Boyer et al., 1997; Hinton et al., 1998; Buffam et al., 2001; Ward et al., 2012). These types of rapid loading events have been estimated to contribute to as much as 86% of annual OC export in temperate forested watersheds (Raymond and Saiers, 2010) and 71–85% of OC export in temperate agricultural watersheds (Dalzell et al., 2007). Similar to our observations of accumulation and mobilization of DOC in above ground flow paths, others have observed a seasonal accumulation of OC in surface soil and litter layers, which is mobilized at the beginning of the wet season (Buffam et al., 2001; Raymond and Saiers, 2010; Ward et al., 2012). In addition to enhanced concentrations, it has been observed in both temperate and high-latitude river systems that OC biodegradability can increases with peak discharge during rain events (Buffam et al., 2001; Fellman et al., 2009). Although we observed enhanced DOC concentrations during periods of increasing discharge, rapid loading events were not as important with regards to annual stream DOC fluxes as in the cases described above because stream flow was largely dominated by baseflow (**Figure 3**). In our study site, stream flow was low relative to the amount of precipitation, likely as a result of high evapotranspiration rates (not measured here). As a result, the estimated stream DOC flux of 1.6 kg C ha−<sup>1</sup> yr−<sup>1</sup> was considerably lower than in other Amazonian streams (**Table 2**). The flux of DOC in other Amazonian streams has been shown to be as high as 31.5 kg C ha−<sup>1</sup> yr−<sup>1</sup> for white water streams and 190 kg C ha−<sup>1</sup> yr−<sup>1</sup> for black water streams (Johnson et al., 2006a). Stream flow DOC fluxes are generally lower in subtropical forests, ranging from 4.1 kg C ha−<sup>1</sup> yr−<sup>1</sup> (Qualls et al., 2002) to 25 kg C ha−<sup>1</sup> yr−<sup>1</sup> (Liu and Sheu, 2003).

#### Dissolved Inorganic Carbon

Contrary to the relatively high levels of DOC in above ground compartments, the concentration of DIC was generally lowest in above-ground flow paths. Similar to other observations, above-ground DIC concentrations were generally near atmospheric CO<sup>2</sup> saturation (Johnson et al., 2006b). The highest DIC concentrations were observed in the soil solution and groundwater. DIC concentrations generally increased throughout the rainy season for each respective flow path (**Figure 5**). In contrast to DOC concentrations, above-ground DIC concentrations progressively increased throughout the wet season (**Figures 3**, **4**). The observed increase in soil solution DIC with precipitation is likely a result of enhanced biological decomposition and root respiration with increasing soil moisture. Soil solution DIC likely increased with soil depth since biological CO<sup>2</sup> production is rapid relative to atmospheric gas exchange (Bekele et al., 2007). Plateau parcels had higher DIC concentrations at 200 cm depth (3.7 ± 0.4 mg C L−<sup>1</sup> ) than hillside and lowland parcels (2.7 ± 0.3 mg C L−<sup>1</sup> ) likely due to a combination of relatively thicker soil layers that inhibit gas exchange and less soil moisture in plateau parcels.

Root respiration and remineralization of organic matter to CO<sup>2</sup> are the primary sources of DIC in soils and subsequent transport to groundwater reservoirs (Hope et al., 2004; Johnson et al., 2006b). Environmental conditions such as soil moisture/temperature, abiotic processes such as mineral sorption/desorption, and biological community composition/activity all influence rates of OC decomposition in soils (Schmidt et al., 2011). The fuels for OC remineralization range from simple substrates (e.g., carbohydrates) to molecularly complex compounds such as lignin macromolecules (Otto and Simpson, 2006; Feng et al., 2008; Kuzyakov, 2010; Thevenot et al., 2010). A similar range of organic carbon fuels are decomposed and remineralized to CO<sup>2</sup> in streams and river channels (Benner et al., 1986, 1995; Mayorga et al., 2005; Ward et al., 2013). Photo-oxidative processes represent an additional pathway for remineralization of OC to CO<sup>2</sup> in stream waters, and in some circumstances may enhance biological degradation rates (Hernes and Benner, 2003; Rodríguez-Zuniga et al., 2008; Spencer et al., 2009). Stream DIC concentrations were generally lower than in soil solution and groundwater because gas exchange occurs more rapidly in the aquatic environment (Hope et al., 2004; Johnson et al., 2006b).

## REFERENCES

Andreae, M. O., Talbot, R. W., Berresheim, H., and Beecher, K. M. (1990). Precipitation chemistry in Central Amazonia. J. Geophys. Res. 95, 16987–16999. doi: 10.1029/JD095iD10p16987

## CONCLUDING REMARKS

Understanding the magnitude and dynamics of carbon export in above and below ground flow paths is critical for constraining the influence of terrestrial and aquatic ecosystems on global carbon cycling. The evolution of carbon constituents along the continuum of terrestrial landscapes, inland waters, and ultimately the ocean is potentially largely dependent on initial interactions of raindrops with the surrounding environment. The results of this study illustrate the seasonality of carbon concentrations in above and below ground flow paths in tropical forests. Calculating a complete watershed carbon budget is logistically difficult, and, as such, flux estimates made here and elsewhere contain large uncertainties and methodological caveats that should be considered when making global comparisons. The flow paths examined here represent spatially-heterogeneous and rapid conduits, and as such, should be examined with increasingly high spatiotemporal resolution in order to fully unravel the complex underlying processes and inherent variability.

## AUTHOR CONTRIBUTIONS

VN, AK, and CN conceived the experimental design and completed the field collections and sample processing. VN and NW interpreted the data and prepared the manuscript. All authors critically revised the manuscript and approved of the final submission.

## FUNDING

This work was supported by FAPESP grant 03/13172-2, a FAPESP fellowship to VN and NSF Grant DEB 0640661 to CN.

## ACKNOWLEDGMENTS

We thank Alexandra Montebelo and Gustavo Baldi for help with field sampling and laboratory analyses. We thank Dárlisson Nunes, Roberto Baena, Sandro, Ebes, Raimundo, and Artemizia Moitta of the Instituto de Pesquisa Ambiental na Amazonia (IPAM) for field help and IPAM and Daniel Nepstad for use of field facilities at Tanguro. Paul Lefebvre helped with watershed delineation, Shelby Hayhoe and Richard McHorney contributed stream discharge data. Access to Tanguro Rancho and rainfall data were provided by Grupo A. Maggi.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2016.00114

Artaxo, P., Gatti, L. V., Leal, A. M. C., Longo, K. M., Freitas, S. R., Lara, L. L., et al. (2005). Química atmosférica na Amazônia: a floresta e as emissões de queimadas controlando a composição da atmosfera amazônica. Acta Amazonica 35, 185–196. doi: 10.1590/S0044-596720050002 00008


dissolved lignin phenols from the Congo River. J. Geophys. Res. 114, G3. doi: 10.1029/2009JG000968


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Neu, Ward, Krusche and Neill. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Genesis and Exodus of Vascular Plant DOM from an Oak Woodland Landscape

Peter J. Hernes <sup>1</sup> \*, Robert G. M. Spencer <sup>2</sup> , Rachael Y. Dyda<sup>1</sup> , Anthony T. O'Geen<sup>1</sup> and Randy A. Dahlgren<sup>1</sup>

<sup>1</sup> Department of Land, Air and Water Resources, University of California, Davis, Davis, CA, USA, <sup>2</sup> Department of Earth, Ocean and Atmospheric Science, Florida State University, Tallahassee, FL, USA

Evaluating the collective impact of small source inputs to larger rivers is a constant challenge in riverine biogeochemistry. In this study, we investigated the generation of dissolved organic matter (DOM) in a small oak woodland catchment in the foothills of northern California, the subsequent transformation in lignin biomarkers and chromophoric DOM (CDOM) parameters during transport through the landscape to an exporting stream, and finally the overall compositional impact on the larger receiving stream and river. Our study included a natural leaching experiment in which precipitation passing through oak, pine, and grass litter and duff samples was collected after each of a series of storms. Also included were soil trench samples to capture subsurface lateral flow, stream samples along with point-source reservoir inputs, and samples of canopy throughfall, stemflow, and gopher hole (bypass) flow. The litter/duff leaching study demonstrated changing DOM fractionation patterns throughout the season, as evidenced by changing lignin compositions in the leachates with each successive storm. This adds a necessary seasonal component to interpreting lignin compositions in streams, as the source signatures are constantly changing. Released DOM from leaching was modified extensively during transit through the subsurface to the stream, with preferential increases in aromaticity as evidenced by increases in carbon-normalized absorbance at 254 nm, yet preferential decreases in lignin phenols, as evidence by carbon-normalized lignin yields in the headwater stream that was less than half that of the litter/duff leachates. Our extensive number of lignin measurements for source materials reveals a much more complex perspective on using lignin as a source indicator, as many riverine values for syringyl:vanillyl and cinnamyl:vanillyl ratios that have previously been interpreted as degraded lignin signatures are also possible as unmodified source signatures. Finally, this study demonstrated that the impact of numerous small headwater streams can significantly overprint the DOM signatures of much larger rivers over relatively short distances spanning several to tens of kilometers. This finding in particular challenges the assumption that river studies can be adequately conducted by focusing only on the main tributaries.

Keywords: dissolved organic matter, lignin biomarkers, hydrologic flowpaths, overprinting, oak woodland, stream order, SUVA

#### Edited by:

Rick Keil, University of Washington, USA

#### Reviewed by:

Mar Nieto-Cid, Spanish National Research Council, Spain Cedric G. Fichot, Boston University, USA Youhei Yamashita, Hokkaido University, Japan

> \*Correspondence: Peter J. Hernes pjhernes@ucdavis.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Earth Science

Received: 03 September 2016 Accepted: 23 January 2017 Published: 13 February 2017

#### Citation:

Hernes PJ, Spencer RGM, Dyda RY, O'Geen AT and Dahlgren RA (2017) The Genesis and Exodus of Vascular Plant DOM from an Oak Woodland Landscape. Front. Earth Sci. 5:9. doi: 10.3389/feart.2017.00009

#### INTRODUCTION

The allure of pipe flow models for DOM in large rivers is unmistakable: large loads and rapid transit would seem to preclude the opportunity for any significant change. Over the past decade, this model has largely been debunked in terms of DOM processing and degradation, as we now know that considerable DOM transformation and remineralization occurs in situ, thereby enhancing the CO<sup>2</sup> outgassing signal from these rivers (Richey et al., 2002; Bouillon et al., 2012). However, the idea of pipe flow or unchanging large loads still persists in terms of tributary allochthonous sources to large rivers. It seems impossible, for example, to consider that streams with discharge <5% of the mainstem can significantly alter the mainstem DOM quality and quantity. Yet riverine chemistry represents an integrated signature of all upstream sources and processes within the catchment/basin, and even those small inputs must be incorporated into models of how riverine chemistry evolves.

Although rivers are grand integrators, it is clear that not all sources and processes are equally represented. Rather, some element of hydrologic connectivity must be incorporated into models of riverine chemistry, such that a soil or vegetation perched on top of a hill is much less connected than a soil or vegetation in the riparian zone. For example, the distinct lignin signature of conifer forests in the headwaters of the Sacramento River basin is overwhelmed by the lignin signature of wetlands and agricultural islands in the Sacramento/San Joaquin River Delta upon passage through the latter (Eckard et al., 2007). However, in terms of discharge, water movement, and land area, this is not a surprising finding as tidal pumping in the Delta essentially throws riverine discharge into reverse during the transition from low tide to high tide, and the land area of the Delta approaches 3000 km<sup>2</sup> . In contrast, what happens to DOM dynamics in a non-tidally influenced system with stream inputs that are so small as to be considered insignificant sources?

Evaluating sources and processing of DOM in rivers requires a multitude of experimental, analytical, and sampling approaches, as well as attention to hydrologic flowpaths. Biomarkers such as lignin offer the specificity to trace DOM derived from angiosperm vs. gymnosperms, and woody vs. non-woody tissues (Hedges and Mann, 1979; Spencer et al., 2016). Optical parameters derived from UV-visible absorbance can trace the bulk properties of chromophoric DOM (CDOM) such as aromaticity, microbial vs. terrigenous sources, and high- vs. lowmolecular weight properties (Traina et al., 1990; Helms et al., 2008; Spencer et al., 2013). Leaching experiments generate a posteriori sources of DOM that can be traced through the study region and evaluated for processing effects (Aufdenkampe et al., 2001; Hernes et al., 2007; Spencer et al., 2008). And of course, demarcating and sampling representative hydrologic flowpaths can be instrumental in determining how the bulk of DOM enters headwater streams, which in turn helps to understand which processes and depths within the soil column are most influential in transforming the initial leachate compositions (Sanderman et al., 2009; Swarowsky et al., 2012).

In this study, we took the approach of characterizing sources, transport, and processing of DOM within an oak woodland landscape from the nanoscale to tens of kilometers with the goal of quantifying the ways in which headwater catchment processes can influence the DOM chemistry of a large river. Because DOM export in the Mediterranean climate of the Sierra Nevada foothills is dominated by storm events, our leaching experiment and sampling strategy revolved around a time series approach during the rainy portion of the water year in January through April. Our suite of analyses included molecular lignin analyses as a biomarker for vascular plant sources and diagenetic change, optical parameters that capture the aromatic component of DOM, and bulk carbon analyses. We collected samples from a perched water monitoring system (trench) that was constructed to intercept hydrologic flowpaths at different depths within the soil profile, from samples of opportunity such as flow from gopher holes and stem/canopy flow, and from streams, reservoirs, and the Yuba River to which all DOM in our study region eventually flows. The central premise of our study was that the cumulative effects of small streams can have a quantitative and traceable impact on the DOM chemistry of a large river.

### MATERIALS AND METHODS

#### Site Description

The litter/duff leaching experiment was conducted in the northern Sierra Nevada foothills at the University of California Sierra Foothill Research and Extension Center (SFREC), approximately 80 km north of Sacramento, CA (**Figure 1**). Additional water sampling occurred in and around SFREC (**Figure 1**), which experiences a Mediterranean climate with cool, moist winters and hot, dry summers. Elevation at the site ranges from 250 to 450 m. The mean annual air temperature is 15◦C (January mean of 8.4◦C, July mean of 25.9◦C). Annual precipitation is 734 mm (307—1235 mm during the past 20 years) and falls exclusively as rainfall, primarily between November and March (Lewis et al., 2006). The dominant ecosystem at the site is oak woodland, which regionally constitutes a mixture of winterdeciduous blue oak (Quercus douglasii H.&A.; ∼45% coverage), an evergreen interior live oak (Quercus wislizenii A. DC.; ∼5% coverage), foothill pine (Pinus sabiniana Dougl.; ∼5% coverage), and mixed annual grasses (∼45% coverage, common species include soft chess [Bromus hordeaceus], ripgut brome [Bromus diandrus], red brome [Bromus madritensis spp. rubens], annual fescue [Vulpia sp.], wild oats [Avena fatua and Avena barbata], and medusahead [Taeniantherum caput-medusae]) (Chow et al., 2009; O'Geen et al., 2010).

#### Field Incubation and Measurements

Storm-associated leachates (eleven events) were collected during a 182-day field incubation (December 1, 2006 through May 31, 2007 comprising the entire rainy season) using fresh litterfall (equivalent to Oi horizon in soil taxonomy) and associated duff (i.e., litter decomposed at the soil surface over 2–4 years, or Oe/Oa horizons) for the four dominant vegetation components (Chow et al., 2009). Litter materials were collected during the first week of November, 2006 prior to any major rainfall events, and triplicate incubations were initiated in polyethylene trays (54 × 43 × 13 cm) using 200–300 g dry weight equivalent

of most materials, but ∼900 g for grass duff and ∼500 g for foothill pine duff. Trays were covered with 0.1 cm mesh stainless steel screen to minimize particulate loss or inputs, then slightly tilted to allow rapid drainage. The trays were suspended one meter above the ground with polyethylene tubes connected to 50-L high density polyethylene (HDPE) carboys (**Supplementary Figure 1A**). Leachate volumes were measured from each tray the day following each rainfall event and oneliter subsamples (HDPE bottles) were collected for analyses after vigorous mixing. Carboys were emptied and rinsed after each collection. The mass of leachate DOC yielded from each tray for each rainstorm was calculated from the volume of leachate collected times the measured DOC concentration, and this was normalized by the original mass of material according to each tray (i.e., leached C per mass litter in units of mg g−<sup>1</sup> ). Remaining mass of material in each tray was measured at the end of the experiment. More details about the experiment and bulk measurements can be found in Chow et al. (2009). Although the experiment was conducted in triplicate, in most cases only a single lignin analysis was conducted due to time and expense. Those that were analyzed in duplicate or triplicate are reported in **Supplementary Table 1**.

Creek into the Yuba River, which is labeled for context.

While there are likely "bottle effects" in conducting a leaching/incubation experiment in this manner (e.g., differences in microbial activity, light conditions, and wetness compared to litters and duffs in contact with the natural soil), containing the materials in trays was necessary to maintain reasonable controls on the experiment. Although the experiment may not be a perfect proxy for natural leaching and degradation, the results are still informative as to overall dynamics.

### Flowpath Continuum Sampling

In addition to the litter/duff leachates, a longitudinal continuum of samples was collected in order to evaluate the evolution of the primary DOM leachate signal along the hydrologic flowpath (**Figure 1**, **Supplementary Figure 2**). Sampling began in an ungrazed 36 ha catchment that contains an ephemeral stream, hereafter abbreviated as HS for "headwater stream," and is equipped with a perched water collection system (trench) to monitor and collect subsurface water flowing out of several distinct soil horizons, with collection trays inserted at the bottom of each of these horizons within the trench (**Supplementary Figure 1B**). Sampling depths included 0–10, 10–30, 30–60, and 60–100 cm, although we did not include any samples from the 60–100 cm depth in our analyses as flow was smaller and more sporadic in this layer than the other three. More details about the perched water sampling system can be gleaned from O'Geen et al. (2010). For this study, stream and trench samples were collected during two different rain events in mid-February 2007 and late January 2008, and the stream alone was sampled in late February 2007. Also included in the January 2008 rain event were stemflow and canopy throughfall for the three tree species as well as gopher hole samples in which water was rapidly flowing out of the gopher holes during peak flow. The latter represent bypass flow as water flowing through gopher holes does not have the typical contact time with soils. The overall sampling within the HS catchment aimed at capturing prominent hydrologic flowpaths depicted in **Supplementary Figure 2**. The larger synoptic study linkages are depicted in **Figure 1**. Briefly, the Collins reservoir discharges into Dry Creek, for which HS is a representative tributary, and Dry Creek flows into the Yuba River. Englebright reservoir is on the Yuba River and its outflow was sampled along with downstream water from the Yuba River above the confluence with Dry Creek and again near Marysville, CA at Simpson Lane, ∼20 km downstream of the Dry Creek confluence. Both reservoirs, Dry Creek, and the Yuba River were sampled in conjunction with the late January 2008 storm event.

## DOC and C-specific UV Absorbance (SUVA254) Analyses

Leachates were pre-filtered through glass fiber filters (Whatman GF/F) and all leachates and water samples passed through a 0.2µm polycarbonate membrane filters (Millipore), prior to DOC and UV-visible analyses. All samples were filtered within 24 h and analyzed within 72 h. DOC concentrations were measured on acidified samples (pH ∼ 2) using a Dohrmann UV-enhanced persulfate TOC analyzer (Phoenix 8000; MDL = 0.1 mg L−<sup>1</sup> ). UV absorbance at 254 nm was measured on a Helios Gamma UV/VIS Spectrophotometer (Thermo Fisher Scientific, Inc.). C-specific UV absorbance at 254 nm (SUVA254) was determined by normalizing UV absorbance at 254 nm with DOC concentration and is reported in the units of liter per milligram carbon per meter (L mg C−<sup>1</sup> m−<sup>1</sup> ).

#### Lignin Analyses

All dissolved lignin samples were filtered through a 0.2µm membrane filter (Millipore polycarbonate) then frozen. Prior to cupric oxide (CuO) oxidation for lignin measurements, water was thawed (20 mL for leachates, 100–500 mL for water samples), acidified to pH 2 to minimize precipitation, then rotary evaporated to ∼3 mL. The concentrate was transferred to Monel reaction vessels (Prime Focus, Inc.) and dried under vacuum centrifugation. Weighed and ground litter and duff materials (10–80 mg) both pre- and post-experiment were transferred directly into the reaction vessels. Lignin analyses were carried out by CuO oxidation following a modified version as outlined by Hedges and Ertel (1982), Spencer et al. (2010a) and Hernes et al. (2013b). Following oxidation in 8% NaOH in the presence of excess CuO at 155◦C for 3 h, samples were acidified and extracted three times with ethyl acetate, with the latter fraction evaporated under a gentle stream of ultrapure nitrogen. Samples were then stored frozen until analysis. Lignin phenols were trimethylsilyl derivatized using bistrimethylsilyltrifluoromethylacetamide (BSTFA); separation of phenols was achieved using an Agilent 6890 gas chromatograph fitted with a DB5-MS capillary column (30 m, 0.25 mm inner diameter, J&W Scientific) and attached to an Agilent 5973 mass selective detector. Quantification was achieved using selected ion monitoring with cinnamic acid as an internal standard following the five-point calibration scheme of Hernes and Benner (2002). All samples were blank-corrected due to the presence of trace amounts of contamination in the NaOH reagent. At least one blank was run for every 10 sample oxidations performed. Blank concentrations of lignin phenols were low (40–55 ng) for the eight lignin phenols measured in this study (three vanillyl phenols: vanillin, acetovanillone, vanillic acid; three syringyl phenols: syringaldehyde, acetosyringone, syringic acid; and two cinnamyl phenols: p-coumaric acid, ferulic acid). Total blanks for these eight compounds never exceeded 5% of the sample yields (average 1.3%).

#### Lignin Mixing Model Calculations

Calculations of lignin overprinting were conducted using linear mixing models of lignin parameters. Depending on the question (i.e., is the goal to determine volumes of water mixed, or amounts of DOC mixed, or amounts of lignin mixed?), this is not strictly correct for ratios, which may not mix linearly over their entire range. In our case, we are primarily asking the question about how much lignin comes from each source, and with similar weighting (i.e., concentrations), linear mixing is a reasonable approximation. This exercise is not presented as a quantitatively precise calculation of mixing, but rather to demonstrate the potential for overprinting to exert significant controls on river DOC compositions. In light of this, the simplest approach that makes the point was chosen. Accurately modeling mixing of ratios necessitates an iterative approach that involves its own uncertainties related to assumptions about weighting, but ultimately leads to greater calculated proportions for the endmember with the most weight. In our case, we are interested in how much the HS stream as a representative endmember overprints larger creeks/rivers, and because lignin concentrations are higher in the composite HS sample than receiving waters, calculating mixed ratios more accurately (i.e., linear mixing of terms in the numerator separately from those in the denominator, and then calculating the ratio) would lead to higher calculated HS proportions than linear mixing of the entire ratio, hence our approach is conservative as to the question of overprinting.

Linear mixing models are typically used when a sample value is bracketed by two endmembers believed to be the primary sources for that sample. If the sample value is equidistant from either endmember value, then the sample is a 50% mixture of each endmember. Determining the percentage of either endmember is possible by calculating the relative offset between the sample value and an endmember and dividing this by the total difference between the two endmembers. For example, if Endmember A has a value of 0.9, Endmember B has a value of 1.0, and the sample value is 0.93, intuitively we know that there is a higher percentage of Endmember A represented in that sample than Endmember B because the sample value is closer to Endmember A. A linear mixing model calculates the percentage of Endmember A as follows: |Endmember B – Sample|<sup>∗</sup> 100%/|Endmember A – Endmember B|, or in this example, |1.0 – 9.3|<sup>∗</sup> 100%/|0.9 – 1.0| = 70% Endmember A. Calculated values >100% are possible if the endmembers do not bracket the sample, which is less helpful for attributing sources, but can still provide context for the magnitude of compositional change that occurs.

#### RESULTS

Due to the large volume of data from this study, much of the data in figures consists of averages and does not capture the full range of values. However, many of these ranges are described below and the comprehensive dataset is available in **Table 1** and **Supplementary Tables 1**, **2**. Weighted averages presented for leachates in **Table 2** were computed only across a single tray/replicate, even when replicate samples were analyzed occasionally from other trays. Because there was variation in mass loss between replicate trays in any given storm, incorporating those replications in weighted averages would have been unnecessarily complicated with minimal added value.

Carbon loss from the leachate experiment has previously been reported by Chow et al. (2009). Briefly, total carbon lost, as determined by difference between pre- and post-litter/duff C mass, ranged from 8.8 to 34% (**Table 1**). The total lost as leached DOC (volume of leachate collected multiplied by DOC concentration) ranged from 1.4 to 12.8%, and after accounting for a small amount lost as particulate organic carbon in the leachates, the remainder by difference was attributed to remineralized organic carbon (OC) lost as carbon dioxide (CO2), which ranged from 5.9 to 31.8% (**Table 1**; Chow et al., 2009).

Lignin concentrations in litter/duff leachates were highly variable, varying by nearly an order of magnitude from 0.15 to 12.1 mg L−<sup>1</sup> (**Supplementary Table 1**). These values were one to three orders of magnitude larger than typical riverine values (Spencer et al., 2008; Hernes et al., 2013b; Mann et al., 2014). On average, lignin concentrations were 2- to 3-fold higher in the litter-derived leachates compared to duff-derived leachates. In general, there was some evidence of a "first flush" in the two December storm events, as lignin concentrations in all litter leachates were 2- to 4-fold higher in averaged December samples than the average over the remainder of the sampling period (**Supplementary Table 1**). Annual grass duff lignin was approximately 2-fold higher in December, while the other three duffs were similar across the sampling period.

Across the 15 month sampling period, HS DOC concentrations ranged from 2.2 to 17.2 mg L−<sup>1</sup> , with all values >10 associated with a single storm on 1/4/2008 which was not part of the leaching experiment (**Supplementary Table 2**). During the snapshot synoptic sampling, DOC for the Collins reservoir sample was 2.9 mg L−<sup>1</sup> , Englebright reservoir was 1.6 mg L−<sup>1</sup> , Dry Creek was 4.8 mg L−<sup>1</sup> , Yuba River near Dry Creek was 2.0 mg L−<sup>1</sup> , and Yuba River at Simpson Lane was 2.4 mg L−<sup>1</sup> (**Supplementary Table 1**). Calculations involving


Negative values in the "Composition of respired lignin" columns indicate a calculated production of either the term in the numerator or the denominator. Average value for S:V and (Ad:Al)s in these columns does not include foothill pine due to minimal syringyl content.

AGD, annual grasses duff; AGL, annual grasses litter; FPD, foothill pine duff; FPL, foothill pine litter; BOD, blue oak duff; BOL, blue oak litter; LOD, live oak duff; LOL, live oak litter; S:V, sum of three syringyl phenols divided by sum of three vanillyl phenols; C:V, sum of two cinnamyl phenols divided by sum of three vanillyl phenols; (Ad:Al)v, vanillic acid divided by vanillin; (Ad:Al)s, syringic acid divided by syringaldehyde; 38, sum of eight lignin phenols divided by organic carbon; OC, organic carbon; DOC, dissolved organic carbon.


TABLE 2 | Average lignin parameters and SUVA254 values for oak woodland leachates, trench samples, and stream in comparison to snapshot parameters for down and upstream sources of water/DOM.

Leachate averages were weighted according to the mass of leached carbon during each storm event, hence are more likely to reflect the earlier storms. Number of samples averaged is in parentheses (two values indicate difference in number of lignin vs. SUVA254), however replicate leachate samples are only counted as one. Canopy/stemflow lignin was not averaged due to distinct compositional differences between the pine and oak samples.

AGD, annual grasses duff; AGL, annual grasses litter; FPD, foothill pine duff; FPL, foothill pine litter; BOD, blue oak duff; BOL, blue oak litter; LOD, live oak duff; LOL, live oak litter; T, trench; HS, headwater stream in Sierra Foothills Research and Extension Center; S:V, sum of three syringyl phenols divided by sum of three vanillyl phenols; C:V, sum of two cinnamyl phenols divided by sum of three vanillyl phenols; (Ad:Al)v, vanillic acid divided by vanillin; (Ad:Al)s, syringic acid divided by syringaldehyde; Λ8, sum of eight lignin phenols divided by organic carbon; SUVA254, carbon-specific ultraviolet absorbance at 254 nm.

DOC utilized at least two decimal points and tabulated DOC is reported at two decimal points, however one decimal point is more representative of the accuracy and therefore the convention adopted for the text.

Lignin concentrations in the trench and HS samples were up to three orders of magnitude smaller than in the leachates (**Supplementary Table 1**: note units of mg L−<sup>1</sup> for leachates vs. µg L−<sup>1</sup> for everything else), ranging from 4.4 to 84.5 µg L−<sup>1</sup> . HS lignin concentrations on average were significantly higher than trench samples (41.2 vs. 20.6 µg L−<sup>1</sup> ), suggesting that lignin in the streams does not solely come from subsurface sources. Lignin concentrations in the Collins reservoir was 16.7 µg L−<sup>1</sup> and increased into Dry Creek at 27.6 µg L−<sup>1</sup> , while the Englebright-to-Yuba River transition also resulted in increased lignin concentrations, from 5.6 µg L−<sup>1</sup> in the former to 11.9 µg L −1 at the Dry Creek confluence to 13.9 µg L−<sup>1</sup> at Simpson Lane (**Supplementary Table 1**).

In contrast to lignin concentrations, carbon-normalized yields of lignin were less variable between duff and litter leachates, with the exception of annual grasses, which averaged 5.68 mg 100 mg OC−<sup>1</sup> in the litters compared to 2.16 mg 100 mg OC−<sup>1</sup> in the duffs (**Table 2**). Average weighted carbon-normalized yields in the other six sample types ranged from 1.72 to 2.56 mg 100 mg OC−<sup>1</sup> . These values are consistent with previous studies involving plant leachates (Hernes et al., 2007, 2013a; Pellerin et al., 2010). Carbon-normalized yields in the litters and duffs prior to leaching ranged from 0.72 mg 100 mg OC−<sup>1</sup> in the annual grass duff to 6.05 mg 100 mg OC−<sup>1</sup> in the annual grass litter, and averaged 3.82 mg 100 mg OC−<sup>1</sup> , while post-leaching values were 1.03–6.87 mg 100 mg OC−<sup>1</sup> with an average of 4.17 mg 100 mg OC−<sup>1</sup> , indicating preferential leaching/degradation of non-lignin plant constituents (**Table 1**).

Carbon-normalized lignin yields in the trench and HS samples were consistently lower than that of the leachates, ranging from 0.12 mg 100 mg OC−<sup>1</sup> (a trench sample) to 3.06 mg 100 mg OC−<sup>1</sup> (a HS sample) with a combined average of 0.93 mg 100 mg OC−<sup>1</sup> . Carbon-normalized yields in the Collins-to-Dry Creek system were essentially identical at 0.58 mg 100 mg OC−<sup>1</sup> , while experiencing an increase from 0.35 mg 100 mg OC−<sup>1</sup> to 0.59 mg 100 mg OC−<sup>1</sup> in the Englebright-to-Yuba River system (**Table 2**).

Diagnostic lignin ratios include source indicators, S:V (indicator of angiosperms, or deciduous plants) and C:V (indicator of non-woody tissue), and diagenetic acid:aldehyde ratios for both vanillyl and syringyl phenols. The lowest dissolved S:V ratios were in the two reservoirs, at 0.63 for Collins (with Dry Creek at 0.81) and 0.49 for Englebright (with Yuba River at 0.69 and 0.89) (**Table 2**). HS and trench samples ranged from 0.51 to 1.17, while angiosperm (i.e., the oaks and grasses) leachates were somewhat higher at 0.75–1.62 (**Supplementary Table 1**). Angiosperm litters and duffs ranged from 0.72 to 1.21 prior to leaching and 0.86 to 1.37 post-leaching. Trends in S:V were largely mirrored in C:V, with the lowest surface water dissolved values (0.07) in the two reservoirs (stemflow and canopy throughfall values were comparable or lower) and increasing into Dry Creek (0.09) and Yuba River (0.14 and 0.17). C:V in HS and trench samples ranged from 0.09 to 0.48, with an average of 0.21, while leachates ranged from 0.04 to 0.34 and on average was higher in the litter leachate than the duff leachate for all four plant types. Litter and duff C:V ranged from 0.03 to 0.59 prior to leaching and 0.05 to 0.58 post-leaching (**Table 1**).

Average dissolved vanillyl acid:aldehyde ratios, (Ad:Al)v, were lowest in the litter leachates (0.41–1.05), and generally higher in all other sample types (1.09–2.00) (**Table 2**). Litter and duff (Ad:Al)v ranged from 0.20 to 0.41 prior to leaching and 0.27 to 0.42 post-leaching (**Table 1**). Average dissolved syringic acid:aldehyde ratios, (Ad:Al)s, were less distinguishable between litters and duffs, ranging from 0.48 to 1.26, while all other samples varied from 0.99 to 1.21 (**Table 2**). Angiosperm litter and duff (Ad:Al)s ranged from 0.19 to 0.57 prior to leaching and 0.26 to 0.51 post-leaching (**Table 1**).

Absorbance of UV-visible light by CDOM can offer complementary information to lignin, as unsaturated compounds such as lignin polyphenols are largely responsible for absorbance patterns (Traina et al., 1990; Blough et al., 1993; Fooken and Liebezeit, 2000). In particular, carbon-specific UV absorbance (SUVA) exhibits strong correlations to aromaticity in DOM (Traina et al., 1990). In this study, SUVA<sup>254</sup> in any single leachate sample ranged from 0.45 L mg C−<sup>1</sup> m−<sup>1</sup> in the foothill pine litter leachate to 4.47 L mg C−<sup>1</sup> m−<sup>1</sup> in the blue oak litter leachate (**Supplementary Table 1**), corresponding to an approximation from past studies of <10% aromaticity to >30% aromaticity, respectively (Weishaar et al., 2003). Blue oak duff leachate on average had the highest SUVA<sup>254</sup> value at 2.58 L mg C <sup>−</sup><sup>1</sup> m−<sup>1</sup> while foothill pine litter leachate was the lowest with an average SUVA<sup>254</sup> of 0.95 L mg C−<sup>1</sup> <sup>m</sup>−<sup>1</sup> (**Table 2**). The litter leachate SUVA<sup>254</sup> values (1.64 L mg C−<sup>1</sup> <sup>m</sup>−<sup>1</sup> ) were on average ∼25% lower than the duff leachates (2.22 L mg C−<sup>1</sup> m−<sup>1</sup> ). Across the continuum, SUVA<sup>254</sup> values were highest in the gopher hole water flows (4.43 L mg C−<sup>1</sup> m−<sup>1</sup> on average), lower in the stem flow and canopy throughfall samples (3.54 L mg C−<sup>1</sup> m−<sup>1</sup> on average) and HS samples (3.75 L mg C−<sup>1</sup> m−<sup>1</sup> ), still lower in the 0–10 and 10–30 cm trench samples (3.13 and 2.83 L mg C <sup>−</sup><sup>1</sup> m−<sup>1</sup> , respectively), and then exhibited a range of values in the reservoir, Yuba River, and Dry Creek samples (2.89–4.38 L mg C−<sup>1</sup> m−<sup>1</sup> ) (**Table 2** and **Supplementary Tables 1**, **2**). For comparison, 21 site-averaged SUVA<sup>254</sup> values measured throughout the Sacramento River basin in 2002-3 ranged from 2.56 to 3.72 L mg C−<sup>1</sup> m−<sup>1</sup> , including an average value of 2.93 L mg C−<sup>1</sup> m−<sup>1</sup> at Yuba River at Simpson Lane (Chow et al., 2007).

#### DISCUSSION

#### Lignin Relativity As Demonstrated by Fractionation Effects in Leachates

Characterizing fractionation effects in lignin biomarkers has significantly improved interpretation of DOM and lignin parameters in various environments, explaining apparent degradation trends that previously were conundrums (Hernes et al., 2007, 2013a; Spencer et al., 2008, 2016). Lignin fractionation studies to date have primarily consisted of snapshot measurements, in which dissolved and particulate lignin is allowed to reach equilibrium, then a single measurement of each is compared to assess the magnitude of fractionation (Hernes et al., 2007, 2013a). Our system was much more dynamic, with repeated leaching with every rainstorm. We observed changing trends in fractionation across the season in 12 out of 24 plant material/diagnostic lignin parameter combinations (p < 0.1, **Figures 2**, **3**). Acid to aldehyde ratios, (Ad:Al)v, increased in five of the eight plant material leachates, S:V increased in live oak and foothill pine litter leachates and decreased in blue oak and annual grasses duff leachates, and C:V decreased in all the duff leachates except foothill pine duff. These seasonal changes add considerable complexity to interpretations of streamwater lignin measurements, but more broadly highlight the need for seasonal measurements of rivers and streams themselves—as well as source materials for DOM—as snapshots are simply inadequate for capturing all the dynamics of integrated river basin biogeochemistry.

Three different mechanisms separately or in combination could be responsible for the changing trends in leachates that we observed: (1) in the absence of significant degradation, the changing trends could simply represent a continuum of solubilities, with the early season ratios representing more soluble lignin and the late season ratios representing less soluble lignin, (2) in the absence of significant solubility differences in the initial litter/duff materials, the changing trends could be due to degradation of litter/duff lignin, which typically results in elevated (Ad:Al)v ratios and decreased S:V and C:V ratios, and (3) if neither solubility nor degradation of lignin is a significant factor, then degradation of cellulose or other organic matter that binds with lignin and inhibits leaching could be slowly releasing lignin of changing compositions. Although none of the mechanisms can be ruled out with certainty, there are some clues as to the overall reactivity of each material. For example, foothill pine litter, annual grasses duff, and foothill pine duff lost <10% of their initial mass due to mineralization of OC to CO2, which would appear to favor a solubility control over a degradation control. On the other hand, live oak duff and live oak litter lost 23 and 32%, respectively, of OC to CO2, which could favor a degradation control over solubility. Perhaps not surprisingly, four out of six lignin parameters for the two live oak materials were linearly correlated to OC loss, with the remaining two parameters showing more complex change (**Figures 2**, **3**). Blue oak litter lost 27% of its carbon over the course of the experiment, with roughly equal proportions to respiration and to DOC leaching (**Table 1**), and the linear relationship between (Ad:Al)v and leached carbon was the strongest of all correlations (**Figure 2**), perhaps a reflection of synergistic effects between leaching and respiration. At the molecular level, (Ad:Al)v in duff leachates are consistently higher than in litter leachates (**Figures 2**, **3**), which clearly indicates that degradation plays a role over the course of 2–4 years (i.e., the approximate age of the duffs), but it remains to be seen if degradation over a few months has the same impact.

Some uncertainty in interpreting fractionation effects on dissolved lignin can be mitigated by measuring particulate and dissolved lignin simultaneously, as it can be reasonably assumed that partitioning between DOM and POM in higher stream order systems is at equilibrium. For example, DOM- derived lignin measured in four rivers in Papau New Guinea demonstrated higher C:V, S:V, and (Ad:Al)v than in POM in 11 of 12 pairings by up to an order of magnitude (Burns et al., 2008), whereas in the Amazon River basin, ultrafiltered DOM C:V and S:V values were consistently lower than fine or coarse POM while (Ad:Al)v values were two to three times higher (Hedges et al., 2000). Our leachate study indicates that both plant sources and timing must be considered when interpreting the direction and magnitude of fractionated lignin measurements in riverine DOM and POM. However, we hypothesize that DOC:POC ratios will exert some level of control on whether the nature of partitioning in rivers is net sorption to or desorption from POM. Ultimately, interpreting DOM lignin data in absolute terms is simply not possible: lignin parameters are a relative accounting system and there must be points of comparison either internally across gradients or externally with particulate sources of dissolved lignin.

#### Leaching and Degradation

In all systems, mass balance must apply: the total amount of any individual lignin phenol in the initial parent materials must be accounted for in either the residual parent material, the cumulative leachate produced, the cumulate particulate material produced, remineralization to CO2, and chemical transformation to other compounds, which could include other lignin phenols (i.e., conversion from aldehyde precursors to acid precursors). The latter two processes are challenging to determine experimentally/analytically and generally must be inferred. Carbon-normalized yields of lignin (38, mg lignin per 100 mg OC) increased in the residual material compared to the starting materials in seven out of the eight treatments (**Table 1**), indicating that lignin was more resistant to leaching and degradation than bulk carbon.

Composition of lignin remineralized to CO<sup>2</sup> can be approximated by difference using mass balance, in which carbon-normalized lignin yields can be used to calculate the mass of each lignin phenol in the pre-litter/duff materials, and then the mass from the post-litter/duff materials along with leached DOC lignin are subtracted to get respired lignin. Although our experimental design was not optimized for this analysis, the general trend indicates that remineralized lignin was slightly depleted in syringyl phenols, enriched in cinnamyl phenols, and depleted in vanillic and syringic acid relative to litter/duff materials (**Table 1**). However, the latter trend could also be explained by the conversion of aldehyde precursors to acid precursors (i.e., production of acid precursors in the

remaining litter/duff materials would lead to smaller amount-bydifference calculations), as is commonly thought to occur during degradation of lignin.

It is now well-documented that the process of leaching releases dissolved lignin with a different composition than the original plant material (Hernes et al., 2007, 2013a; Spencer et al., 2008). The magnitude of the fractionation can readily be seen in this study by comparing the time-series leachate data in **Figures 2**, **3** to the solid colored lines that represent the initial litter/duff material. However, a plot of S:V vs. C:V, which can be used to distinguish angiosperm (deciduous) tissues from gymnosperm (conifer) tissues and woody tissues from nonwoody tissues (Hedges and Mann, 1979), puts this fractionation into interpretive context. For example, grass litter and duff leachates span C:V ratios from 0.1 to 0.6, with the lower value traditionally interpreted as degraded plant material while the higher value is considered fresh (**Figure 4**). (In general, litter C:V do appear higher than duff values (**Figure 4**), lending credence to the idea that C:V is responsive to degradation, however this does not appear the case for annual grasses, while systematic differences between litter and duff leachate S:V values are ambiguous.) In the bigger picture, all of the foothill pine, blue oak, and live oak C:V ratios, including fresh litters, are below the ranges originally considered (Hedges and Mann, 1979) as either gymnosperm or angiosperm non-woody tissues (**Figure 4**). Similar to the lower end of the grasses C:V values, all of these samples would traditionally be defined as degraded. Hence it is clear that the range of values for S:V and C:V in plant materials themselves are broader than once thought, let alone the values of leachates, which places even more importance on adequately characterizing source materials to rivers and streams before attempting to interpret dissolved lignin values.

#### DOM Export to Streams

In California oak woodlands, most DOM export from catchments is driven by storm events that rapidly mobilize DOM from the adjacent landscape which in turn flows to streams via several dynamic hydrologic flowpaths (Swarowsky et al., 2012). Scaling our litter/duff leachate yields (mass DOC released per mass of litter/duff) to a total annual amount of litter and duff on the landscape of HS (field observations of litter and duff horizon depths multiplied times density and area) indicates a net production of approximately 450 kg ha−<sup>1</sup> DOC (Chow et al., 2009). In contrast, calculated stream yields from HS and other similar streams at the research site indicate a DOC export of approximately 0.03 kg ha−<sup>1</sup> DOC. Rapid microbial remineralization of plant leachates has previously been demonstrated (Cleveland et al., 2004; Pellerin et al., 2010). Our study shows that DOM produced on the landscape is nearly quantitatively retained or remineralized, and DOM that does flow into adjacent streams has the potential to be highly modified by biotic and abiotic processes.

Our study did not capture subsurface flow (i.e., trench samples) or stream dynamics for all storms across the leaching experiment, but were instead focused on two storm events in February 2007 and an additional sampling in January 2008 as part of an effort to capture a snapshot of DOM at multiple scales (stream orders) simultaneously. Hysteresis effects, in which DOC concentrations at a given discharge are significantly different between the rising and falling limbs of a storm, are well-known for streams and rivers (McGlynn and McDonnell, 2003), and similar variability extended to DOM composition as represented by lignin parameters (**Figure 5**). Short time-series sampling during the storm events revealed little in the way of consistent trends in how lignin parameters respond to DOM mobilization in HS (**Figure 5**), as S:V in any individual sampling event varied by up to 0.2, C:V by 0.1, (Ad:Al)v and (Ad:Al)s by 0.4, and 3<sup>8</sup> by >2.0.

A portion of compositional variability is likely due to firstflush phenomena, which can be delayed by the necessity of wetting up in Mediterranean climates (O'Geen et al., 2010), however, changing contributions from different hydrologic flowpaths also must play a role (Swarowsky et al., 2012). The soil trench samples that isolate water transport through different soil horizons can be potentially helpful in parsing out these differences. Our most complete set of trench samples was collected in conjunction with the January 2008 storm, and it is clear that in terms of lignin compositions, the 10–30 cm depth was the most volatile, with large changes in every lignin parameter during the storm, while the 0–10 and 30–60 cm depths were more stable (**Figure 6**). The 10–30 cm depth during this storm event accounted for over half of the total flow measured in the trench (Swarowsky et al., 2012). Based on DOC (and to some extent lignin concentrations) in HS stream water that was higher than the trench samples, it is clear that there must be additional sources of DOM to the stream. Overland flow is generally not observed in this system due to rapid infiltration. However, during the January 2008 storm, water gushed out of gopher holes in the catchment, indicating a potential role for horizontal bypass flow, and the six samples collected during this study yielded DOC values of 3.9–8.7 mg L−<sup>1</sup> , with a mean of 5.8 mg L−<sup>1</sup> , which could partially account for the higher HS DOC values. Alternatively, if our leachate samples are proxies for bypass flow, then the potential contributions for bypass vs. subsurface DOC sources to HS can be evaluated compositionally with lignin property:property plots. In terms of S:V vs. C:V, HS most resembles the 30–60 cm depth trench sample and the blue oak duff leachate sample (**Figure 7A**), while (Ad:Al)v vs. (Ad:Al)s favors the 0–10 and 10–30 cm depths, along with the annual grass duff and blue oak duff leachates (**Figure 7B**). Of the two, S:V vs. C:V is more conducive to identifying sources (Hedges and Mann, 1979), and since blue oak comprises ∼45% of the vegetation in the catchment, it is not surprising that HS lignin would more closely resemble blue oak leachates. Yet the results of (Ad:Al)v vs. (Ad:Al)s actually seem to fit intuition, as the landscape is dominated by blue oak and annual grasses, and the 0–10 and 10–30 cm trench depths generate the most flow. This raises the possibility that in certain environments, acid:aldehyde ratios may have utility as a secondary source indicator, i.e., not useful for identifying the original plant materials, but becoming a "signature" of a water body. This has previously been postulated for water masses in marine environments where lignin diagenesis below the surface appears conservative on the timescale of centuries (Hernes and Benner, 2002). Given that annual grasses cover ∼45% of the landscape, it is a bit surprising that they are not more representative of HS in S:V vs. C:V space. However, percent coverage is not the critical factor, but rather, the total amount of litter/duff that accumulates from each vegetation type. Blue oak and foothill pine accumulate >20 times the amount in litter/duff materials per unit area compared to annual grasses (Chow et al., 2009), hence blue oak is the dominant vegetative source in this landscape and even foothill pine at ∼5% coverage has similar amounts of litter/duff materials compared to annual grasses.

In some systems, lignin compositions can be used quantitatively to determine relative contributions from different sources of vascular plant DOM (Eckard et al., 2007), however in this system the compositions of all the potential contributors are too similar for this type of analysis. Carbon-normalized yields of lignin are another quantitative tool that can be used to differentiate between vascular plant sources and non-vascular plant sources such as microbes or algae. Values of 3<sup>8</sup> for HS averaged 0.95 mg lignin 100 mg OC−<sup>1</sup> , which is significantly lower than weighted averages for all leachates (**Table 2**), including a <sup>3</sup><sup>8</sup> value of 2.04 mg lignin 100 mg OC−<sup>1</sup> for blue oak duff leachate, which appears to be the predominant source compositionally. This indicates that all leachate sources contain more than sufficient lignin to account for HS yields and that either lignin is preferentially lost relative to bulk carbon, or that HS DOM has been significantly augmented with algal/microbial

sources. Again, the trench samples give an indication of where this modification may take place, as all trench 3<sup>8</sup> values are significantly lower than the leachates, but the 0–10 and 30–60 cm depths are even lower than HS. Microbial communities are known contributors to the soil organic matter pool, so lower trench 3<sup>8</sup> values could reflect dilution by non-vascular plant sources such as microbially-derived DOM. On the other hand, lignin phenols are also surface active and hence subject to sorption to soils (Kaiser et al., 1996; Hernes et al., 2007; Klotzbücher et al., 2016), which could preferentially remove lignin relative to bulk carbon. However, soils do not have infinite sites for lignin sorption, so over time the sorbed lignin would need to be degraded by microbes to free up those sites for the next storm event. Regardless of the mechanism for reducing 3<sup>8</sup> values between leachates and passage through various soil horizons, the 3<sup>8</sup> values of the three trench samples indicate that the 10–30 cm depth must constitute at least half the DOM in HS, if no other sources are considered, which is consistent with the fact that the 10–30 cm depth was the dominant hydrologic flowpath during the storm event (Swarowsky et al., 2012).

Optical properties do not have the specificity of biomarkers, but likely represent a larger proportion of the total DOM than any individual biomarker, and as such, can offer other clues as to the sources of DOM to streams. Carbon-normalized UV absorbance shows strong correlations to aromaticity (Traina et al., 1990), and in this study, HS SUVA<sup>254</sup> values averaged 3.75 L mg C−<sup>1</sup> m−<sup>1</sup> . In contrast, the weighted average SUVA<sup>254</sup> value for blue oak duff leachate was the highest of all the leachates at 2.58 L mg C−<sup>1</sup> m−<sup>1</sup> , yet still considerably lower than the stream values. Trench samples are closer in magnitude to HS streams, with lower SUVA<sup>254</sup> values (average 2.98 L mg C−<sup>1</sup> m−<sup>1</sup> ) compared to streams indicating that DOC modification must occur in conjunction with other higher SUVA<sup>254</sup> sources, such as gopher holes (SUVA<sup>254</sup> values from 4.29 to 4.70 L mg C−<sup>1</sup> m−<sup>1</sup> ) or stem-flow and canopy throughfall (SUVA<sup>254</sup> ranging from 3.04 to 3.83 L mg C−<sup>1</sup> m−<sup>1</sup> ). Overall, the SUVA<sup>254</sup> values highlight a more complex picture of sources than perhaps lignin values alone. While blue oak is clearly an important contributor to stream DOM in this catchment, the optical properties of leachate DOM from blue oak duff still must undergo considerable modification before export to streams, including the preferential loss of non-aromatic carbon. This appears somewhat contradictory to the 3<sup>8</sup> analysis in which leachates are also likely modified by preferential loss of lignin, which is an aromatic carbon component. However, lignin is just one of many different classes of aromatic compounds in organic matter, and the more straightforward conclusion is that SUVA<sup>254</sup> and lignin are tracing different components of DOM.

#### DOM Dynamics across a Continuum of Stream Orders

DOM chemistry at any given point in a river represents the sum of all sources and processes upstream. Those sources and processes are not weighted equally, as local landscape features tend to overprint more remote signatures (Eckard et al., 2007). In the regional system studied here, HS flows into Dry Creek, which in turn flows into the Yuba River. Dry Creek flows out of the Collins reservoir, so a stream like HS would act to overprint the Collins reservoir signature. Similarly, the Yuba River flows out of Englebright reservoir, and Dry Creek would act to overprint that signature. In reality, a complete picture of DOM dynamics would require continuous composition and concentration monitoring at all these sites over an extended time period, but our snapshot still provides insight into the evolution of DOM from catchments to higher order streams and rivers.

Among the many impacts of building dams, impounding river water in reservoirs can have a significant effect on both DOM quantity and quality (Oliver et al., 2016). Increased residence time allows for more input from algal sources and greater degradation of allochthonous DOM, the latter which is enhanced by sedimentation that increases optical transparency and allows photochemical degradation of aromatic-rich terrigenous DOM (Kraus et al., 2011). The dissolved lignin signatures of both the Collins and Englebright reservoirs appear consistent with these effects, as 3<sup>8</sup> values are low (0.58 and 0.36 mg lignin 100 mg OC−<sup>1</sup> , respectively), which can be achieved by both algal production and lignin degradation, and the (Ad:Al)v values are the highest measured in our dataset (1.92 and 2.00). SUVA<sup>254</sup> values are more complicated to interpret: the SUVA<sup>254</sup> value in Collins of 3.12 L mg C−<sup>1</sup> m−<sup>1</sup> falls within the range that seems typical of rivers and streams in this region and is similar to values tracked in the San Luis Reservoir (2.94 L mg C−<sup>1</sup> m−<sup>1</sup> average) associated with the State Water Project (Kraus et al., 2011). In contrast, Englebright SUVA<sup>254</sup> of 4.38 L mg C−<sup>1</sup> m−<sup>1</sup> is considerably higher, but with low total DOC of 1.6 mg L −1 , it seems more likely that this high SUVA<sup>254</sup> results from conservative concentration of non-lignin aromatics, as opposed to production from new sources. In natural river systems, source inputs for river DOC are challenging to determine as a starting point, but the presence of these large and well-homogenized reservoirs serves as a strong backdrop for evolving DOM signatures downstream.

Dry Creek DOM clearly carries the signature of the Collins reservoir out of which it flows, as evidenced by identical 38, as well as other lignin compositional parameters and SUVA<sup>254</sup> values that are all within 10–20% of Collins (**Table 2**). Although HS itself may not constitute a high enough discharge to overprint Dry Creek all by itself, it is representative of many small streams that discharge into Dry Creek along its transit from Collins reservoir to the Yuba River. In this context, the lignin compositional parameters of Dry Creek are all intermediate between Collins reservoir and HS, indicating a clear compositional shift to the Dry Creek signature imparted by HS and similar streams over the ∼7 km transit distance between Collins and the Dry Creek sampling point. A linear mixing model, |Dry Creek – Collins|<sup>∗</sup> 100%/|HS – Collins|, applied to each of the four lignin compositional parameters suggests that the proportion of Dry Creek DOM derived from HS and similar streams ranges from 13 to 101% (**Table 3**). Granted, this does not account for variations from stream to stream, nor does it account for in situ processing, but it provides a clear indication as to the potential collective impact of seemingly insignificant streams (in terms of discharge) on larger streams. Simply by DOC mass balance alone, the difference between Collins DOC at 2.9 mg L−<sup>1</sup> and Dry Creek DOC at 4.7 mg L−<sup>1</sup> indicates that small streams must account for ∼40% of the DOC in Dry Creek.

In comparison to the Collins to Dry Creek transformation, the Englebright to Yuba River at Dry Creek transformation appears greater over the ∼15 km reach between sampling points: 3<sup>8</sup> values increase by >60% from 0.36 to 0.59 mg lignin 100 mg OC−<sup>1</sup> , SUVA<sup>254</sup> values drop from 4.38 to 3.08 L mg C−<sup>1</sup> m−<sup>1</sup> , C:V values double, S:V values increase by ∼40%, and (Ad:Al)v values appear less degraded (**Table 2**). A ∼25% increase in DOC from 1.6 mg L−<sup>1</sup> in Englebright compared to 2.0 mg L −1 in the Yuba River at Dry Creek reflects distinct inputs of carbon. Only one notable tributary, Deer Creek, enters the Yuba River in between Englebright reservoir and the Yuba River sampling point at Dry Creek, and discharge from Deer Creek was ∼8% that of the Yuba River at the time of sampling (**Figure 8**), although lag times between precipitation runoff and management decisions to release water from reservoirs adds nuance to interpretation. Hence the transformation in DOM in the Yuba River is not coming from a single point source, but

TABLE 3 | Linear mixing models to demonstrate the magnitude of overprinting over 7–35 km reaches of Dry Creek and the Yuba River.


All changes in downstream values were in the direction of HS composition, however, some Yuba River at Simpson values were beyond even the HS compositions, and these are represented by percentages >100%. See Materials and Methods for more explanation of the linear mixing models.

HS, headwater stream in Sierra Foothills Research and Extension Center; S:V, sum of three syringyl phenols divided by sum of three vanillyl phenols; C:V, sum of two cinnamyl phenols divided by sum of three vanillyl phenols; (Ad:Al)v, vanillic acid divided by vanillin; (Ad:Al)s, syringic acid divided by syringaldehyde; Λ8, sum of eight lignin phenols divided by organic carbon.

must also be driven by the collective inputs from multiple smaller streams as well as DOM exchange with local landscape features and geomorphological features within the river channel itself. Given the relative magnitudes of the Yuba River compared to Dry Creek (<5% of the discharge of the Yuba River), it is somewhat surprising that Dry Creek is less impacted during transit than the Yuba River. However, this also reflects the challenges in making generalizations between processes in streams and rivers of significantly different orders (8th order for Yuba, 3rd or 4th order for Dry Creek).

Ultimately, there is great interest in gauging how well the DOM signature of a larger river like the Yuba River reflects the headwater sources. Clearly, the more diversity in landscapes within a river basin, the less likely river chemistry will mimic any single headwater source. In this case, there are distinct similarities in lignin chemistry and SUVA<sup>254</sup> between HS and the Yuba River despite the influence of reservoir effects on the Yuba River. Similar to the Dry Creek analysis, the Yuba River is bracketed well by Englebright vs. HS compositions, with linear mixing model calculations, |Yuba River – Englebright|<sup>∗</sup> 100%/|HS – Englebright|, indicating that 35–70% of the DOM at the Yuba River sampling point is derived from streams similar to HS (**Table 3**). This analysis can be extended even further with the Yuba River at Simpson Lane measurement ∼35 km downstream of Englebright. DOC has increased from 1.6 to 2.4 mg L−<sup>1</sup> , and essentially, Yuba River at Simpson Lane lignin has completely converted to an HS-like signature, with three parameters at or near 100% and one greatly exceeding even the HS composite (**Table 3**). While parameterizing small source signatures as a single average is overly simplistic (among the many HS lignin measurements, there are individual samples that would sufficiently bracket Yuba River at Simpson with Englebright), this exercise overall speaks to the dominance of oak woodlands in the lower Yuba River basin.

#### Implications

This study captured DOM sources and processing at multiple scales, highlighting (1) the dynamic temporal changes in DOM source signatures even from the same source materials, (2) extensive modification and attenuation of DOM during transport through the landscape to the exporting stream, (3) more diversity in lignin source signatures than has previously been reported, and (4) the significant DOM overprinting capacity of small streams to much larger rivers. While these finding unquestionably address significant gaps in our knowledge of river basin biogeochemistry, they also highlight the greater challenges of river biogeochemistry research overall.

It is clear that novel and more comprehensive experimental designs must be brought into river research. A recent emphasis on time-series sampling in the past decade, for example, has led to significant modifications in measurements of river fluxes and seasonally evolving DOM and POM compositions (Hernes et al., 2008; Spencer et al., 2010b, 2016; Ellis et al., 2012). However, there is a great need for parsing out in situ sources and processing of riverine OM vs. external inputs, and one potential approach could be to pair studies of concrete-lined aquaducts with nearby rivers to try to constrain the relative role of external vs. internal. Lagrangian sampling of rivers could be useful for determining OM transformation hotspots (both new sources and processing) within a river channel, that could then lead to more intense investigations of sources and processing at those hotspots. Continual advancements in sensors that allow high temporal and spatial monitoring of colored DOM (CDOM) properties in rivers will undoubtedly help in capturing the complexity of river systems (Downing et al., 2012; Pellerin et al., 2012), but in spite of their potential, CDOM and fluorescent DOM (FDOM) by themselves have proven elusive in giving the chemical specificity that is needed to truly understand DOM dynamics. On the other hand, such sensors could also be used as targeting agents to point toward regions where chemical analyses can be put to the greatest use (Hernes et al., 2009).

River systems are incredibly complex both spatially and temporally, with flows that vary over several orders of magnitude, varying degrees of exchange in both directions with groundwater, varying landscapes through which they flow, changes in sediment loads that impact primary production and photochemical oxidation as well as sorption/desorption processes, and a myriad of point and non-point sources to deconvolute in order to interpret chemistry. In the collective attempts by the research community to narrow the scope of river studies to scales that are conceivably more accessible to interpretation, we lose the overall richness and complexity of the entire system. The overall connectedness of river chemistry to the landscape through which it flows re-emphasizes the incredible potential of rivers as canaries in the coal mine for capturing the effects of climate and land use change.

### AUTHOR CONTRIBUTIONS

RAD and AO designed leaching experiment and field sampling scheme and conducted DOC analyses, hydrology measurements, and select UV-visible absorbance measurements. Lignin analyses were conducted by or under the supervision of RD and PH. RS conducted additional UV-visible measurements. The manuscript was written by PH, with the contribution of all authors.

## FUNDING

This work was funded by the California State Water Resources Control Board (agreement no. 04-120-555-01), with additional contributions from National Science Foundation grant 1335622 (PH). Open access publication fees were supported in part by the Gordon and Betty Moore Foundation.

## ACKNOWLEDGMENTS

The authors would like to thank Jiayou Deng, Alex Swarowsky, Tony Orozco, and Jeannie Evatt at the Department of Land, Air, and Water Resource for their field and lab support. Sarah Flores, Daniel Jara, and Jennifer Harfmann conducted lignin analyses on various subsets of the samples.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/feart. 2017.00009/full#supplementary-material

Supplementary Figure 1 | Experimental setup for (A) litter and duff leaching during storms, and (B) subsurface lateral flow sampling via the trench.

Supplementary Figure 2 | Conceptual diagram of hydrologic flowpaths within the HS catchment at SFREC ultimately leading to export via HS. Percentages refer to the total lateral flow contributed by each soil layer during a typical storm event. The top three horizons correspond to the three perched water (trench) samples analyzed in this study. This figure is reproduced from O'Geen et al. (2010). ©The Regents of the University of California.

Supplementary Table 1 | All lignin data from the study, along with ancillary bulk data (when measured) and information, including leached volumes from the leachate study, DOC, SUVA254, times, and dates.

Supplementary Table 2 | Additional DOC and SUVA254 data from the study on samples that were not analyzed for lignin.

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Hernes, Spencer, Dyda, O'Geen and Dahlgren. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Molecular and Optical Properties of Tree-Derived Dissolved Organic Matter in Throughfall and Stemflow from Live Oaks and Eastern Red Cedar

#### Aron Stubbins <sup>1</sup> \*, Leticia M. Silva<sup>1</sup> , Thorsten Dittmar <sup>2</sup> and John T. Van Stan<sup>3</sup>

*<sup>1</sup> Department of Marine Sciences, Skidaway Institute of Oceanography, University of Georgia, Savannah, GA, USA, <sup>2</sup> Research Group for Marine Geochemistry, Institute for Chemistry and Biology of the Marine Environment, University of Oldenburg, Oldenburg, Germany, <sup>3</sup> Department of Geology and Geography, Georgia Southern University, Statesboro, GA, USA*

#### Edited by:

*Richard G. Keil, University of Washington, USA*

#### Reviewed by:

*David Christopher Podgorski, Florida State University, USA Christian Schlosser, GEOMAR Kiel, Germany*

#### \*Correspondence:

*Aron Stubbins aron.stubbins@skio.uga.edu*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Earth Science*

Received: *28 November 2016* Accepted: *20 February 2017* Published: *28 March 2017*

#### Citation:

*Stubbins A, Silva LM, Dittmar T and Van Stan JT (2017) Molecular and Optical Properties of Tree-Derived Dissolved Organic Matter in Throughfall and Stemflow from Live Oaks and Eastern Red Cedar. Front. Earth Sci. 5:22. doi: 10.3389/feart.2017.00022* Studies of dissolved organic matter (DOM) transport through terrestrial aquatic systems usually start at the stream. However, the interception of rainwater by vegetation marks the beginning of the terrestrial hydrological cycle making trees the headwaters of aquatic carbon cycling. Rainwater interacts with trees picking up tree-DOM, which is then exported from the tree in stemflow and throughfall. Stemflow denotes water flowing down the tree trunk, while throughfall is the water that drips through the leaves of the canopy. We report the concentrations, optical properties (light absorbance) and molecular signatures (ultrahigh resolution mass spectrometry) of tree-DOM in throughfall and stemflow from two tree species (live oak and eastern red cedar) with varying epiphyte cover on Skidaway Island, Savannah, Georgia, USA. Both stemflow and throughfall were enriched in DOM compared to rainwater, indicating trees were a significant source of DOM. The optical and molecular properties of tree-DOM were broadly consistent with those of DOM in other aquatic ecosystems. Stemflow was enriched in highly colored DOM compared to throughfall. Elemental formulas identified clustered the samples into three groups: oak stemflow, oak throughfall and cedar. The molecular properties of each cluster are consistent with an autochthonous aromatic-rich source associated with the trees, their epiphytes and the microhabitats they support. Elemental formulas enriched in oak stemflow were more diverse, enriched in aromatic formulas, and of higher molecular mass than for other tree-DOM classes, suggesting greater contributions from fresh and partially modified plant-derived organics. Oak throughfall was enriched in lower molecular weight, aliphatic and sugar formulas, suggesting greater contributions from foliar surfaces. While the optical properties and the majority of the elemental formulas within tree-DOM were consistent with vascular plant-derived organics, condensed aromatic formulas were also identified. As condensed aromatics are generally interpreted

**41**

as deriving from partially combusted organics, some of the tree-DOM may have derived from the atmospheric deposition of thermogenic and other windblown organics. These initial findings should prove useful as future studies seek to track tree-DOM across the aquatic gradient from canopy roof, through soils and into fluvial networks.

Keywords: Tree-DOM, dissolved organic matter (DOM), carbon, CDOM, deposition, dissolved organic carbon (DOC), stemflow, throughfall

#### INTRODUCTION

In forested catchments, trees represent the first interceptors of precipitation and the first potential source of dissolved organic matter (DOM) to the aquatic carbon cycle. The earliest trees appear in the fossil record approximately 385 million years ago (Stein et al., 2007), since when they have fundamentally altered terrestrial (Algeo et al., 2001; Gensel, 2001) and wetland ecosystems (Greb et al., 2006). Forests are estimated to have covered close to 50 million km<sup>2</sup> of the planet 5,000 years ago (FAO, 2016) equivalent to approximately 1/3rd of the earth's land surface. Just as forests transformed the global ecosystem, humans now have a similarly profound influence upon global ecology and biogeochemistry. Deforestation during the Anthropocene (Crutzen, 2002) has seen forest land cover reduced by approximately 50% to 31.7 million km<sup>2</sup> as of 2005 (Hansen et al., 2010) and was continuing at a rate of approximately 1.5 million km<sup>2</sup> year−<sup>1</sup> between 2000 and 2012 (Hansen et al., 2013).

Despite the vast and rapidly changing expanse of land covered and volume of precipitation intercepted by trees, only modest attention has been focused upon the DOM delivered by trees to downstream ecosystems (Kolka et al., 1999; Michalzik et al., 2001; Neff and Asner, 2001; Levia et al., 2011; Inamdar et al., 2012). Once intercepted, rainwater takes one of two hydrological flow paths to the forest floor: throughfall (water that drips from the canopy or falls directly through canopy gaps) and stemflow (water funneled by the canopy to the stem). Both stemflow (5– 200 mg-C L−<sup>1</sup> ) (Moore, 2003; Tobón et al., 2004; Levia et al., 2011) and throughfall (1–100 mg-C L−<sup>1</sup> ) (Michalzik et al., 2001; Neff and Asner, 2001; Le Mellec et al., 2010; Inamdar et al., 2012) are enriched in DOM relative to rainwater (0.3–2 mg-C L−<sup>1</sup> ) (Willey et al., 2000).

The fate of exported tree-derived DOM (tree-DOM) will depend upon the chemistry of the tree-DOM, the nature of the receiving ecosystem and hydrological considerations. Significant losses and alteration of tree-DOM occurs as stemflow and throughfall enter soils due to sorption to mineral soils, which preferentially retain hydrophobic DOM fractions (Jardine et al., 1989; Kaiser and Zech, 1998, 2000). The spatially and temporally uneven delivery of biolabile tree-DOM to soil ecosystems during storms may fuel biogeochemical hot spots and hot moments (McClain et al., 2003; Vidon et al., 2010) and microbial utilization of biolabile organics in soils further modifies tree-DOM (Aitkenhead-Peterson et al., 2003). It has also been suggested that sunlight driven photoreactions (Mopper et al., 2015) may act as a sink for tree-DOM (Aitkenhead-Peterson et al., 2003). Photoreactions would be expected to preferentially remove aromatics, including black carbon (Stubbins et al., 2012b), while preserving and producing high H/C compounds such as aliphatics (Stubbins et al., 2010; Stubbins and Dittmar, 2015).

The degree to which tree-DOM is lost and altered by these processes before reaching a downstream aquatic ecosystem also depends upon the flow path traveled down the tree and from there to an inland water body (Inamdar et al., 2012). Direct input of stemflow or throughfall into a stream or lake will presumably result in negligible alteration prior to delivery. High levels of minimally modified tree-DOM may also reach inland waters during periods of heavy rainfall and resultant high flow when residence times within modifying ecosystems (e.g., soils) are reduced or bypassed completely in the case of overland flow. High flow pulses driven by heavy precipitation shunt reactive DOM downstream through river networks and are increasingly recognized as significant components of the fluvial carbon cycle (Raymond et al., 2016). These extreme, short lived pulses can account for the majority of annual river DOC loads being exported in just a few days per year (Raymond and Saiers, 2010). It remains unclear whether tree-DOM is delivered efficiently to fluvial systems during these pulse-shunt events.

To further understand the quality of tree-DOM, we collected stemflow and throughfall samples from broadleaved (oak) and needleleaved (cedar) trees with or without epiphytes during two storm events. The concentrations, optical properties and molecular signatures of the sampled tree-DOM are presented.

#### METHODS

#### Sample Site

Samples were collected on the Skidaway Institute of Oceanography (SkIO) campus, Georgia, USA (31.9885◦N, 81.0212◦W) (**Figure 1a**) during two rain events: Storm A on June 27th 2015 and Strom B on 28th June 2015 (**Table 1**). SkIO is in a subtropical climate zone (Köppen Cfa), with 30-year mean annual precipitation ranging from 750 to 1,200 mm that occurs as rainfall and mostly during the summer months (GA Office of the State Climatologist). Average daily temperatures in summer range between 30 and 35◦C (Georgia Office of the State Climatologist, 2012). The elevation of SkIO ranges from 0 to 10 m above mean sea level. The sampling sites were flat (0–5% slopes) and underlain by Chipley fine sandy soils (https://websoilsurvey. sc.egov.usda.gov). Two species of tree were sampled: Quercus virginiana Mill. (southern live oak) referred to here as oak for brevity; and, Juniperus virginiana L. (eastern red cedar) referred to as cedar. The epiphytes Tillandsia usneoides L. (Spanish moss) and Pleopeltis polypodioides (resurrection fern) can be found to

cover these trees at high densities on SkIO (**Figures 1b–e**). A Spanish moss-covered cedar tree (cedar moss, **Figure 1b**) and an epiphyte-free cedar tree (bare cedar, **Figure 1c**) were chosen to assess whether the influence of epiphytes were apparent in the concentrations or quality of tree-DOM. Four oak trees were sampled, each with highly variable epiphyte coverage including, both resurrection ferns and Spanish moss (**Figures 1d,e**), which is typical of live oaks in the maritime southeastern US. Both stemflow and throughfall samples were collected for each tree type.

#### Sample Collection and Processing

All plastic and glassware were pre-cleaned by rinsing five times with ultrapure water (MilliQ), soaking in pH 2 ultrapure water (2 ppt 6N hydrochloric acid), re-rinsing five times with ultrapure water, and dried. Once dry, glassware was further baked at 450◦C for 8 h. Twenty throughfall samplers (0.18 m<sup>2</sup> , 0.5 m height, high density polyethylene (HDPE) bins) were deployed for each storm, three beneath each of four oaks (TF Oak 1–4), and four beneath the bare (TF Bare Cedar) and four beneath the epiphyte-covered cedars (TF Cedar Moss). An additional four throughfall samplers were placed upon open ground to sample rainwater (Rain). Stemflow samplers, which consisted of collars cut from polyethylene tubing wrapped about the trunk at 1.4 m height and connected to 10 L HDPE carboys, were installed on four oaks (SF Oak 1–4), the bare cedar (SF Bare Cedar) and the epiphyte-covered cedar (SF Cedar Moss). Throughfall and stemflow collectors were deployed approximately 1 h before rainfall commenced and collected within 1 h of rainfall ceasing. All sampling sites are within 10 min walk of Stubbins' laboratory at SkIO. Samples were rapidly returned to the laboratory and 0.2 µm filtered within 4 h of collection. Sample volumes were measured. Throughfall and rainwater volume fluxes (mm) were calculated by dividing the sample volumes by the surface area of the samplers (0.18 m<sup>2</sup> ).

#### Dissolved Organic Carbon Concentrations

After filtration, aliquots of sample were transferred to pre-combusted 40 mL glass vials, acidified to pH 2 (hydrochloric acid), and analyzed for non-purgable organic carbon using a Shimadzu TOC-VCPH analyzer fitted with a Shimadzu ASI-V autosampler. In addition to potassium hydrogen phthalate standards, aliquots of deep seawater reference material, Batch 10, Lot# 05-10, from the Consensus Reference Material Project (CRM) were analyzed to check the precision and accuracy of the DOC analyses. Analyses of the CRM deviated by <5% from the reported value for these standards (41–44 µM-DOC). Routine minimum detection limits in the investigator's laboratory using the above configuration are 2.8 ± 0.3 µM-C and standard errors are typically 1.7 ± 0.5% of the DOC concentration (Stubbins and Dittmar, 2012).

#### Colored Dissolved Organic Matter

Filtered sample (non-acidified) was placed in a 1 cm quartz absorbance cell situated in the light path of an Agilent 8453 ultraviolet-visible spectrophotometer and CDOM absorbance spectra were recorded from 190 to 800 nm. Ultrapure water provided a blank. Blank corrected absorbance spectra were corrected for offsets due to scattering and instrument drift by subtraction of the average absorbance between 700 and 800 nm (Stubbins et al., 2011). Data output from the spectrophotometer were in the form of dimensionless absorbance (i.e., optical density, OD) and were subsequently converted to the Napierian absorption coefficient, a (m−<sup>1</sup> ) (Hu et al., 2002). If sample absorbance (OD) exceeded 2 at 250 nm, samples were diluted 10 fold with ultrapure water and reanalyzed. Specific UV absorbance at 254 nm (SUVA254; L mg-C−<sup>1</sup> <sup>m</sup>−<sup>1</sup> ), an indicator of DOM aromaticity defined as the Decadic absorption coefficient at 254 nm (m−<sup>1</sup> ) normalized to DOC (mg-C L−<sup>1</sup> ) (Weishaar et al., 2003) was calculated along with spectral slope over the range 275–295 nm (S275−295) (Helms et al., 2008). Spectral slope values are reported as positive values.

### Fourier Transform Ion Cyclotron Resonance Mass Spectrometry

In the current study, whole water samples were analyzed without extraction or purification to provide the broadest analytical window prior to mass spectral analysis. Each stemflow sample (SF Oak 1–4; SF Cedar Moss; SF Bare Cedar) was analyzed. For throughfall samples, carbon-weighted composite samples were generated for each rainfall and throughfall sample by combining carbon-dependent volumes of sample (i.e., all four aliquots were combined for each of the rain samples and each cedar sampled for throughfall; three aliquots were combined for each of the four oak trees sampled for throughfall; the volume that each aliquot contributed to a composite sample was adjusted in order that the final composite sample contained an equal fraction of carbon from each aliquot). To generate consistent FT-ICR mass spectra all samples were analyzed under the same conditions, including DOC concentration. Therefore, all tree-DOM samples were diluted to the identical DOC concentration with ultrapure water (10 mg-C L−<sup>1</sup> ) and then further diluted 1:1 with methanol. However, as rainwater DOC (1–2 mg-C



*Values present are means* ± *one standard deviation.*

L −1 ; **Table 1**) was significantly lower than for tree-DOM and reducing all sample DOC concentrations to <1 mg-C L−<sup>1</sup> would have impaired FT-ICR MS performance, rainwater samples were diluted 1:1 with methanol and run at their resulting DOC concentrations (Storm A: 1.1 mg-C L−<sup>1</sup> ; Storm B: 0.6 mg-C L −1 ). As this impaired the quality of the rainwater FT-ICR MS data, this data is only used to contrast with the tree-DOM data in a cluster analysis and the molecular quality of rainwater DOM is not presented. In order to compare rainwater DOM to tree-DOM directly, the study design would have needed to include a DOM isolation and concentration step in order to allow all samples, rainwater included, to be analyzed by FT-ICR MS at the same concentrations. This option was not chosen as it would have reduced the analytical window for our focus of study: tree-DOM.

Once mixed 1:1 with methanol, samples were analyzed in negative mode electrospray ionization using a 15 Tesla FT-ICRMS (Bruker Solarix) at the University of Oldenburg, Germany. 500 broadband scans were accumulated for the mass spectra. After internal calibration, mass accuracies were within an error of <0.2 ppm. Elemental formulas were assigned to peaks with signal to noise ratios greater than five based on published rules (Koch et al., 2007; Stubbins et al., 2010; Singer et al., 2012). Peaks detected in the procedural blank (PPL extracted ultrapure water) were removed. Peak detection limits were standardized between samples by adjusting the dynamic range of each sample to that of the sample with the lowest dynamic range (dynamic range = average of the largest 20% of peaks assigned a formula divided by the signal to noise threshold intensity; standardized detection limit = average of largest 20% of peaks assigned a formula within a sample divided by the lowest dynamic range within the sample set; Spencer et al., 2014; Stubbins et al., 2014). Peaks below the standardized detection limit were removed. These peaks were removed in order to prevent false negatives within samples with low dynamic range.

For each elemental formula, we calculated the modified Aromaticity Index (AImod) (Koch and Dittmar, 2006, 2016), which indicates the likelihood of an elemental formula representing aromatic structures, from an AImod of zero, where formulas are aliphatic, through an intermediate range, where an elemental formula could indicate aromatic or non-aromatic isomers, to AImod values above 0.5, where an elemental formula is highly likely to represent aromatic isomers (Koch and Dittmar, 2006). These AImod values were calculated as:

$$\text{Almod} = (1 + \text{C} - 0.5\text{O} - \text{S} - 0.5(\text{N} + \text{P} + \text{H})) \tag{1}$$

$$\text{/(C} - 0.5\text{O} - \text{S} - \text{N} - \text{P}) \tag{1}$$

AImod values 0.5–0.67 and >0.67 were assigned as aromatic and condensed aromatic structures, respectively (Koch and Dittmar, 2006). Compound classes were further defined as highly unsaturated (AImod < 0.5, H/C < 1.5, O/C < 0.9), unsaturated aliphatics (1.5 ≤ H/C < 2, O/C < 0.9, N = 0), saturated fatty acids (H/C ≥ 2, (O/C < 0.9), sugars (O/C ≥ 0.9) and peptides (1.5 ≤ H/C < 2, O/C < 0.9, N < 0). Since an individual formula could occur as multiple isomeric structures, these classifications only serve as a guide to the structures present within DOM. For instance, "peptides" have the elemental formulas of peptides, but their actual structure may differ.

Standardized peak intensities (z) within a sample were calculated following:

$$\mathbf{z} = \frac{\mathbf{x} - \boldsymbol{\mu}}{\sigma} \tag{2}$$

where, x is the measured peak intensity, µ is mean peak intensity within the sample, and σ is the standard deviation in peak intensity within the sample (Spencer et al., 2014). Cluster analysis of the standardized peak intensities of assigned formulas (Ward clustering in JMP <sup>R</sup> ) was then performed (Spencer et al., 2014).

## RESULTS

#### Dissolved Organic Matter Concentrations and Optical Properties

DOM concentrations, quantified as DOC, ranged from 1.1 to 2.4 mg-C L−<sup>1</sup> in rainwater, 40 to 95 mg-C L−<sup>1</sup> in oak stemflow, 2.2 to 28 mg-C L−<sup>1</sup> in oak throughfall, 52 to 71 mg-C L−<sup>1</sup> in epiphytecovered cedar stemflow, 31 to 62 mg-C L−<sup>1</sup> in epiphyte-covered cedar throughfall, 25 to 30 mg-C L−<sup>1</sup> in bare cedar stemflow, and 12 to 16 mg-C L−<sup>1</sup> in bare cedar throughfall across the two storms sampled.

Absorbance spectra decayed exponentially with increasing wavelength (**Figure 2**). Napierian absorption coefficients for CDOM at 300 nm ranged from 0.1 to 3.2 m−<sup>1</sup> in rainwater, 101 to 526 m−<sup>1</sup> in oak stemflow, 6 to 81 m−<sup>1</sup> in oak throughfall, 321 to 378 m−<sup>1</sup> in epiphyte-covered cedar stemflow, 95 to 148 m−<sup>1</sup> in epiphyte-covered cedar throughfall, 129 to 159 m−<sup>1</sup> in bare cedar stemflow, and 36 to 67 m−<sup>1</sup> in bare cedar throughfall across the two storms sampled.

Spectral slope values for the range 275–295 nm for CDOM ranged from 0.0156 to 0.0673 nm−<sup>1</sup> in rainwater, 0.0142 to 0.0147 nm−<sup>1</sup> in oak stemflow, 0.0139 to 0.0169 nm−<sup>1</sup> in oak throughfall, 0.0145 to 0.0154 nm−<sup>1</sup> in epiphyte-covered cedar stemflow, 0.0157 to 0.0166 nm−<sup>1</sup> in epiphyte-covered cedar throughfall, 0.0145 to 0.0155 nm−<sup>1</sup> in bare cedar stemflow, and 0.0152 to 0.0165 nm−<sup>1</sup> in bare cedar throughfall across the two storms sampled.

SUVA<sup>254</sup> values ranged from 0.5 to 1.3 L mg-C−<sup>1</sup> <sup>m</sup>−<sup>1</sup> in rainwater, 1.9 to 4.2 L mg-C−<sup>1</sup> m−<sup>1</sup> in oak stemflow, 1.6 to 2.8 L mg-C−<sup>1</sup> m−<sup>1</sup> in oak throughfall, 4.0 to 5.1 L mg-C−<sup>1</sup> m−<sup>1</sup> in epiphyte-covered cedar stemflow, 2.3 to 2.8 L mg-C−<sup>1</sup> m−<sup>1</sup> in epiphyte-covered cedar throughfall, 4.0 to 4.1 L mg-C−<sup>1</sup> m−<sup>1</sup> in

bare cedar stemflow, and 2.4 to 3.3 L mg-C−<sup>1</sup> m−<sup>1</sup> in bare cedar throughfall across the two storms sampled.

The full dataset is presented in Table S1. Means and standard deviations for each data and sample type per storm are presented in **Table 1** and summarized in **Figure 3**. Patterns in DOC concentrations and DOM optical properties were similar between storms (**Figure 3**). Rainwater DOM had much lower values of DOC, CDOM, and SUVA, and much steeper spectral slope values than the tree-DOM samples. In general, the stemflow samples exhibited higher DOC, CDOM, and SUVA, and shallower spectral slopes, than their respective throughfall samples (**Figure 3**). The one exception being DOC for the epiphyte-covered cedar during storm A, when both stemflow and throughfall DOC concentrations were similar (**Figure 3A**).

## Molecular Signatures of Tree-Derived Dissolved Organic Matter

Whole water samples mixed 1:1 with methanol yielded mass spectra (**Figure 4**) with sufficient resolution and signal to enable the assignment of 5,852 elemental formulas to tree-DOM. The raw mass spectra for oak and cedar stemflow displayed molecular signatures consistent with those of whole river water DOM run on the same instrument, under the same conditions, during the same month (Kolyma River data in **Figure 4**; Stubbins et al., 2017). Looking at one representative mass to charge (343 m/z; **Figure 4**), DOM in oak stemflow and cedar stemflow has similar molecular diversities (i.e., there are a similar number of peaks). However, the relative abundance of some peaks varies between samples, with some peaks being below detection in one sample and present in the other.

To further explore the diversity of tree-DOM and how it varied among the samples a cluster analysis of the standardized peak intensities of assigned formulas (Ward clustering in JMP <sup>R</sup> ) was performed (Spencer et al., 2014). The distance graph for this cluster analysis revealed a sharp slope break at 4 clusters (**Figure 5**; lower panel). These four clusters were: rainwater, stemflow oak, throughfall oak, cedar, the latter including stemflow and throughfall from both the epiphyte-covered and bare cedar trees (**Figure 5**). Oak DOM, including both stemflow and throughfall, and cedar DOM are clearly separated (distance between clusters, d = 17). Oak stemflow and throughfall are also separated from one another (distance between clusters, d = 12). Although other clusters are formed, the distances between them are smaller (d < 6). For instance, the epiphyte-covered and bare cedar trees form distinct clusters, but with a cluster distance of <2 these samples have limited molecular differences. Based upon the cluster analysis, the molecular properties of three molecularly distinct types of tree-DOM are presented: oak throughfall, oak stemflow and cedar (**Table 2**). Data for the rainwater cluster are not presented.

The oak stemflow cluster comprised 4,765, oak throughfall 3,565, and cedar 3,643 formulas. All samples contained high proportions of CHO-only formulas (65–71%; **Table 2**). The oak stemflow was enriched in nitrogen compared to oak throughfall, and both were enriched in nitrogen compared to cedar DOM. Oak throughfall was enriched in sulfur and depleted in phosphorous compared to the other forms of tree-DOM, with cedar DOM being the most enriched in phosphorous.

The elemental formulas for each tree-DOM type were plotted in van Krevelen space (**Figures 6A–C**). Oak DOM spanned a broader range of van Krevelen space than cedar DOM. All types of tree-DOM contained high intensity elemental formulas in the region bounded by approximately H/C 1.1–1.6 and O/C 0.15 and 0.35. Oak DOM also had similarly high intensity elemental formulas in the approximate region H/C 0.5–1.0 by O/C 0.25–0.7.

All forms of tree-DOM covered a wide area of van Krevelen space and correspondingly included elemental formulas with a diverse range of possible structural properties (**Table 2**). All tree-DOM clusters had similar contributions from aromatic compounds (14–15%), but oak DOM was enriched in condensed aromatics (12–13%) compared to cedar DOM (10%). Highly unsaturated formulas were the most prominent molecular class in each type of tree-DOM and were enriched in cedar DOM (47%) compared to oak DOM (38–40%; **Table 2**). Combined aliphatics, including both unsaturated aliphatics and saturated fatty acids, constituted approximately 25% of peaks in all tree-DOM types. Sugar and peptide contributions were low across tree-DOM types, but were elevated in oak DOM compared to cedar DOM. Oak throughfall DOM had a lower average molecular mass (359 g mol−<sup>1</sup> ) compared to both oak stemflow and cedar DOM (382–383 g mol−<sup>1</sup> ). Average H/C and AImod

values for the clusters were similar. However, O/C decreased from oak stemflow > oak throughfall > cedar.

To reveal the quintessential molecular signatures of each tree-DOM type, the degree to which each elemental formula was enriched in a tree-DOM cluster relative to the mean for all tree-DOM types (i.e., mean intensity for a molecular formula in the whole dataset, excluding the two rain samples) was calculated as:

$$\begin{aligned} & \text{Enricchment Factor} \\ &= \frac{\text{Mean Intensity in Classted Sample}}{\text{Mean Intensity in All Tree DOM samples}} \end{aligned} \tag{3}$$

An elemental formula was then classified as being enriched within a tree-DOM type if it exhibited an enrichment factor >1 and also had relatively low variations in intensity across the clustered samples (standard deviation <50% of mean intensity). The standard deviation term was included in the classification in order to exclude formulas which were not routinely enriched within the samples of a cluster. The results of this classification are presented in **Table 2** and in van Krevelen diagrams (**Figures 6D–F**). Dot size in the van Krevelen diagrams represents the mean intensity of an elemental formula in the mass spectrum, while dot color represents the degree of enrichment ranging from just above 1 (yellow) to 3 or higher (dark blue). Oak stemflow was enriched in a large number (**Table 2**) and wide variety of molecules (**Figure 6D**) compared to oak throughfall and cedar DOM (**Figures 6E,F**). However, many of the elemental formulas that were highly enriched in oak stemflow (enrichment factor >2; darker blues) were present at relatively low intensity (small size of the dots). Oak throughfall was also enriched in a wide variety of DOM types (**Figure 6E**), but had highest enrichment in the low H/C, low O/C region typical of condensed aromatics and within the region bounded by approximately H/C 1.1– 1.6 and O/C 0.15 and 0.35 where the original van Krevelens (**Figures 6A–C**) revealed high abundance within all tree-DOM types. Finally, cedar DOM was enriched in elemental formulas with H/C values from approximately 1.0–1.5 and O/C 0.1–0.45 (**Figure 6F**).

Quintessential cedar DOM formulas (i.e., formulas that were consistently enriched in cedar DOM; **Table 2** right side) were enriched in CHO-only formulas (93%) compared to oak DOM (83–84%). The number of nitrogen containing quintessential formulas decreased in the order oak stemflow > oak throughfall > cedar, while sulfur containing quintessential formulas decreased in the order oak throughfall > cedar > oak stemflow, and phosphorous containing quintessential formulas decreased in the order cedar > oak stemflow > oak throughfall (**Table 2**). The average molecular mass, O/C, and AImod of quintessential oak stemflow formulas was higher than that for cedar and oak throughfall, while quintessential cedar formulas had the highest average H/C and lowest average O/C and AImod. Quintessential oak formulas were enriched in condensed aromatics (17–21%) and contained a small proportion of peptide (0.3%) and sugar (2.0–4.3%) formulas, while the quintessential cedar formulas included zero condensed aromatic, peptide or sugar formulas (**Table 2**). Quintessential oak stemflow formulas included approximately twice the percentage of aromatic formulas (23%) when compared to the other tree DOM types (12–13%). Quintessential cedar formulas were highly enriched in highly unsaturated formulas (71%), which represented approximately half of the quintessential oak stemflow formulas (53%), and about a third (30%) of quintessential oak throughfall formulas. Finally, quintessential oak throughfall formulas were enriched in unsaturated aliphatics (30%) compared to quintessential cedar (16%) and oak stemflow (5%).

Distribution plots further resolved variations in molecular mass, H/C, O/C, and AImod of the quintessential formulas that are enriched were the different tree-DOM types (**Figure 7**). Quintessential oak stemflow formulas covered a broad, relatively evenly distributed range in molecular mass, O/C, H/C and AImod, with the H/C distribution skewed toward lower values (center ∼1.0; **Figure 7C**). Those formulas that were enriched in oak throughfall DOM exhibited pronounced peaks in abundance at low molecular mass (∼260 g mol−<sup>1</sup> ; **Figure 7A**), low O/C (∼0.24; **Figure 7B**), high H/C (∼1.6; **Figure 7C**), and contained three spikes in AImod (0, ∼0.2 and ∼0.8; **Figure 7D**). Quintessential cedar formulas also exhibited marked peaks in molecular mass (∼310 g mol−<sup>1</sup> ; **Figure 7A**), midranges in H/C (∼1.0–1.2; **Figure 7C**) and AImod (∼0.4; **Figure 7D**) and a peak in O/C that reached a maximum at ∼0.36, but exhibited two smaller shoulders at 0.21 and 0.11 (**Figure 7B**).


TABLE 2 | Molecular signatures of tree-derived dissolved organic matter (DOM) within live oak stemflow, live oak throughfall and cedar (stemflow plus throughfall).

*Percentages represent the relative contributions of each molecular class within each type of tree-DOM. Left side: values for all formulas within each type of tree-DOM. Right side: values for formulas that were enriched within each type of tree-DOM.*

## DISCUSSION

#### Concentrations and Optical Signatures of Tree-DOM

The high levels of DOC and CDOM in stemflow and throughfall relative to rainwater samples indicate that rainwater DOM was a very minor component of stemflow and throughfall DOM, with the majority of tree-DOM being entrained during interaction with the tree canopy and stem.

Mean DOC concentrations in throughfall (10–52 mg-C L−<sup>1</sup> ) and stemflow (25–78 mg-C L−<sup>1</sup> ) were both within the range of values reported by previous throughfall (1–100 mg-C L−<sup>1</sup> ) (Michalzik et al., 2001; Neff and Asner, 2001; Le Mellec et al., 2010; Inamdar et al., 2012) and stemflow (5–200 mg-C L−<sup>1</sup> ) (Moore, 2003; Tobón et al., 2004; Levia et al., 2011) studies. The observed range in tree-DOC concentrations is at the higher end or exceeds the mean DOC concentrations in major US rivers (1– 12 mg-C L−<sup>1</sup> ; Spencer et al., 2012), but is consistent with the high DOC values observed in black water rivers draining swamps (e.g., St. Marys = 42 mg-C L−<sup>1</sup> ; Spencer et al., 2012). Given the small size of the current dataset (two storms, six trees), no estimates of DOC fluxes are made.

The optical properties of tree-DOM are broadly consistent with those for CDOM in other aquatic environments. Tree-DOM CDOM spectra exhibit an exponential increase in absorbance with decreasing wavelength (**Figure 2**). The range in mean spectral slope values (S275-295) for stemflow (0.0144 nm−<sup>1</sup> ) and throughfall (0.0157–0.0165 nm−<sup>1</sup> ) are consistent with values for US rivers (0.012–0.023 nm−<sup>1</sup> ) (Spencer et al., 2012). Literature values for stemflow and throughfall spectral slope were not found for comparison. SUVA<sup>254</sup> values for stemflow (means 3.0–5.1 L mg-C−<sup>1</sup> m−<sup>1</sup> ) from our oaks and cedars compare with ranges of 2.5–4.9 L mg-C−<sup>1</sup> m−<sup>1</sup> for American beech (Fagus grandifolia) and 3.7–6.2 L mg-C−<sup>1</sup> m−<sup>1</sup> for yellow poplar (Liriodendron tulipfera) (Levia et al., 2011). These values are all at the higher end or exceeding the range in mean SUVA<sup>254</sup> values reported for US rivers (1.3–4.6 L mg-C−<sup>1</sup> m−<sup>1</sup> ) (Spencer et al., 2012) and are consistent with highly colored, aromatic-rich DOM (Weishaar et al., 2003). The mean SUVA<sup>254</sup> values for throughfall (2.2–2.9 L mg-C−<sup>1</sup> m−<sup>1</sup> ) compare to previous literature values for throughfall of 1.8–4.7 L mg-C−<sup>1</sup> m−<sup>1</sup> (Inamdar et al., 2012) and also indicate a significant contribution of aromatics to throughfall DOM. All tree-DOM SUVA values were higher than for rainwater CDOM and all tree-DOM spectral slopes steeper than for rainwater DOM (**Figures 3C,D**), indicating tree-DOM to be more aromatic than the trace amounts of DOM in rainwater. Hydrological fluxes become enriched with aromatic compounds, including lignin degradation products, as contact time with bark increases (Guggenberger et al., 1994). Therefore, the enrichment of stemflow in highly aromatic, high SUVA<sup>254</sup> DOM compared to throughfall is likely due to the high hydrological connectivity of stemflow with tree bark and other sources of tree-derived organics. The aromatics in tree-DOM are likely dominated by autochthonous, tree-produced aromatics, such as lignin, and their degradation products (Guggenberger et al., 1994) that accumulate and are then washed off the tree surface during rain events. In addition, allochthonous organics delivered to the tree via atmospheric deposition (Guggenberger and Zech, 1994) could include aromatics derived from soils, combustion sources, or distal vegetation.

The quantity, but not the optical quality of tree-DOM exported during both storms varied with epiphyte cover. DOC

concentration and CDOM were consistently lower in both stemflow and throughfall from the bare cedar than from the epiphyte covered cedar and the four oaks, which all had mixed epiphyte cover (**Table 1**; **Figures 1**, **3**). These results suggest that epiphytes were either a direct source of autochthonous, epiphyte organics or an intermediate accumulator of organics derived from the tree, fauna or atmospheric deposition. Research into the fluxes and quality of epiphyte DOM release is scarce. With respect to the dominant epiphytes encountered in our study, no data is available for resurrection ferns; while Spanish moss collected from nearby sites in coastal Georgia leached DOM with significantly lower SUVA<sup>254</sup> values (Van Stan et al., 2015) than for the current tree-DOM samples (**Table 1**). The Spanish moss samples leached in Van Stan et al. (2015) were cleaned of all canopy soil and any decaying or damaged moss. As such, these leachates contained organics derived directly from Spanish moss rather than from the more diverse canopy ecosystem and potential organic sources that the presence of Spanish moss in a tree cultivates. For the limited data collected in our study (two storms, six trees, one of which has no epiphyte cover), the presence of epiphytes increased DOC concentrations, but did not reduce the SUVA<sup>254</sup> values as would be expected if the additional DOC leached directly from healthy Spanish moss. Therefore, the DOM enrichment within stemflow and throughfall exported by epiphyte covered trees is likely due to the increase in hydrological contact time and flow path (Levia and Frost, 2003), the accumulation of organic matter from bark, litter and fauna facilitated by the presence of epiphytes in the canopy ecosystem (Hietz et al., 2002; Woods et al., 2012), and potentially, the increase in canopy surface area for atmospheric deposition (Rodrigo et al., 1999; Woods et al., 2012).

#### Molecular Signatures of Tree-DOM

The molecular signatures of whole water tree-DOM as revealed by negative mode electrospray ionization FT-ICR MS (**Figure 2**) share many features of whole water and extracted DOM from other aquatic environments (Mopper et al., 2007; Singer et al., 2012; Chen et al., 2014; Dittmar and Stubbins, 2014; Cawley et al., 2016). Shared features include a high diversity of elemental formulas distributed broadly in van Krevelen space (**Figures 6A–C**) and spanning a range of molecular classes (**Table 2**). The average molecular mass of tree-DOM (359–382 g mol−<sup>1</sup> ; **Table 2**) was greater than for Kolyma River whole water DOM run on the same mass spectrometer (336 g mol−<sup>1</sup> ) (Spencer et al., 2015), but lower than for Congo River whole water DOM run on a different mass spectrometer (424 g mol−<sup>1</sup> ) (Stubbins et al., 2010). Tree-DOM was also H-poor (H/C 1.22–1.24) and O-rich (O/C 0.39–0.45) compared to Kolyma River DOM (H/C 1.27; O/C 0.39) (Spencer et al., 2015). Average H/C and O/C were not reported for Congo River DOM (Stubbins et al., 2010). Other reports of elemental formulas for non-extracted river water DOM are scarce.

As for other whole water FT-ICR mass spectra for terrigenous DOM from freshwater environments (Stubbins et al., 2010,

2012a; Spencer et al., 2015), tree-DOM was dominated by CHOonly formulas and was rich in highly unsaturated molecules (**Table 2**). As they have a high degree of isomeric freedom, any single highly unsaturated elemental formula may represent a diverse array of structures of correspondingly diverse potential biogeochemical sources and functions (Stubbins et al., 2010). For instance, the possible isomers of any highly unsaturated formula include aromatic ring containing lignin degradation products derived from vascular land plants (Stubbins et al., 2010) and carboxylic-rich alicyclic molecules (Hertkorn et al., 2006) of indeterminate, potentially microbial origin. Due the high SUVA<sup>254</sup> values of tree-DOM and particularly stemflow DOM, it is likely that a significant proportion of the unsaturated formulas within tree-DOM represent vascular plant derived molecules that contain aromatic rings.

The modified aromaticity index classifies formulas as either aromatic or condensed aromatic (Koch and Dittmar, 2006, 2016). Tree-DOM was enriched in aromatic formulas (14–15%; **Table 2**) compared to Congo River whole water analyzed on a different FT-ICR mass spectrometer (9%) (Stubbins et al., 2010), but similar in aromatic content to Kolyma River whole water analyzed on the FT-ICR mass spectrometer used in the current study (13%) (Spencer et al., 2015) consistent with the view that river DOM is derived predominantly from the degradation products of vascular plants (Ertel et al., 1984) and the enrichment of similar compounds within tree-DOM.

Tree-DOM was also enriched in condensed aromatics (10– 12%; **Table 2**) compared to both Congo (1%) (Stubbins et al., 2010) and Kolyma River (6%) (Spencer et al., 2015) samples. Condensed aromatics are known to form during the incomplete combustion of organics (Goldberg, 1985) and when observed in DOM are usually termed dissolved black carbon and ascribed a thermogenic source (Kim et al., 2004; Hockaday et al., 2006; Ziolkowski and Druffel, 2010). The ubiquity of dissolved black carbon in river waters (Dittmar et al., 2012; Jaffé et al., 2013; Stubbins et al., 2015; Wagner et al., 2015) is explained as resulting from the ubiquity of refractory, apparently thermogenic black carbon in soils (Forbes et al., 2006; Guggenberger et al., 2008; Schmidt et al., 2011). Other sources of black carbon to natural waters and landscapes include direct input from local combustion sources and atmospheric deposition from distant combustion sources. Atmospheric deposition has been posited as a source of organics to remote regions of the earth (Stubbins et al., 2012a; Spencer et al., 2014) and organics transported from global and regional sources of combustion (e.g., automobile, industrial, domestic, agricultural, wildfire, and biomass burning), as well as produced locally on the Skidaway Institute of Oceanography campus, could all produce black carbon for deposition to the trees sampled.

Aliphatic formulas (sum of unsaturated aliphatics and saturated fatty acids) were slightly enriched in tree-DOM (23– 25%; **Table 2**) relative to Congo (19%) (Stubbins et al., 2010) and Kolyma River (22%) (Spencer et al., 2015) DOM. Tree-DOM was also enriched in sugar (2.3–3.8%) and peptide (2.2–5.1%) formulas compared to Kolyma River DOM (0.6% sugar; 2.2% peptide) (Spencer et al., 2015). Sugar and peptide formulas were not reported for the Congo River (Stubbins et al., 2010). All of these compound classes likely derive directly from foliar leachates, foliar washoff, and their breakdown products (Guggenberger and Zech, 1994; Michalzik et al., 2001; Kalbitz et al., 2007). However, similar molecular formulas

samples.

are also observed in atmospheric aerosols (Wozniak et al., 2008).

## Quintessential Signatures of Oak Stemflow, Oak Throughfall and Cedar DOM

Tree-DOM molecular signatures were significantly different from rainwater DOM and clustered in three groups: oak stemflow, oak throughfall and cedar, the latter including stemflow and throughfall samples from both the epiphyte covered and bare cedars (**Figure 5**). Distinct differences between the molecular signatures of each class of tree-DOM were evident in van Krevelen plots (**Figures 6A–C**). In order to determine the elemental formulas that distinguished these three classes of tree-DOM from one another, the quintessential formulas associated with each type of tree-DOM were classified as those formulas that are consistently enriched across the samples within a cluster (Equation 3). Plotting the resultant data in van Krevelen space revealed that the above data treatment accentuated differences in the molecular signatures of oak stemflow, oak throughfall, and cedar DOM (**Figures 6D–F**). As noted in the results, dot size in the van Krevelen diagrams (**Figures 6D–F**) represents the mean intensity of an elemental formula in the mass spectrum, while dot color represents the degree of enrichment ranging from just above 1 (yellow) to 3 or higher (dark blue).

Quintessential oak stemflow formulas (**Figure 6D**) occupied much of the van Krevelen space typically populated by riverine DOM (e.g., Spencer et al., 2015 in which data for a Kolyma River whole water sample run on the same mass spectrometer is displayed). The van Krevelen is densely populated, as is also typical of riverine DOM samples. The enrichment of stemflow DOM in this diverse population of formulas indicates that the organics within oak stemflow are of more diverse origin or have undergone more extensive processing than the organics within oak throughfall and cedar DOM. The canopy structure and coarse bark of live oaks makes them excellent habitats for colonization by epiphytes, such as the Spanish moss and resurrection ferns observed on our sampled trees (**Figure 1d**), the accumulation of organic debris, and utilization by fauna. The development of these canopy microhabitats, replete with observable canopy soils within the trees on SkIO campus, may have led to the development of a rich molecular mix of organics for export. While of diverse stoichiometric composition, the quintessential oak stemflow elemental formulas were of higher average molecular mass and AImod (**Table 2**; **Figure 7A**), and enriched in aromatics, compared to quintessential oak throughfall and cedar formulas, suggestive of the greater interaction of stemflow with the oak bark, debris and canopy soils.

Quintessential oak throughfall formulas (**Figure 6E**) were not as evenly distributed in van Krevelen space as the quintessential formulas of oak stemflow suggesting formulas enriched in throughfall derive from more distinct sources and have undergone less processing. Compared to the quintessential formulas associated with oak stemflow and cedar, quintessential oak throughfall formulas were of low average molecular weight (311 g mol−<sup>1</sup> ) and were enriched in unsaturated aliphatics (30%) and sugars (4.3%; **Table 2**), suggesting significant inputs from the direct leaching or washing of foliar surfaces (Guggenberger and Zech, 1994; Michalzik et al., 2001; Kalbitz et al., 2007). Oak throughfall also contained elevated levels of condensed aromatics (21%; **Table 2**; **Figure 7B**) compared to the other tree-DOM classes, suggesting potential wash off of deposited combustion products from leaf surfaces.

By comparison to oak DOM, cedar DOM was enriched in CHO-only and highly unsaturated formulas (**Table 2**) of limited molecular diversity (**Figure 6F**; **Figure 7C**), suggesting cedar DOM is enriched in minimally processed non-descript tree-DOM.

## CONCLUSIONS

The relatively high SUVA<sup>254</sup> values and abundance of aliphatics and aromatic formulas within tree-DOM are consistent with autochthonous (i.e., tree-derived) organics. However, the presence of condensed aromatics within tree-DOM also suggests that some of the DOM exported from trees derives from the atmospheric deposition of allochthonous organics. As electrospray ionization efficiency varies with analyte chemistry, the current dataset cannot be used to robustly quantify the contribution of condensed aromatics or other forms of deposited organics to tree-DOM export. Future work should therefore seek to quantify the contribution of condensed aromatics and other allochthonous forms of DOM to tree-DOM to assess how much of the tree-DOM flux is derived from autochthonous vs. allochthonous sources.

As the crowning headwaters of the terrestrial hydrological cycle, tree canopies are the point of first contact between precipitation and terrestrial ecosystems. The quality of tree-DOM as detailed by absorbance and FT-ICR MS is similar to the terrigenous DOM described in inland waters, but sufficiently distinct that optical and chemical signatures may be of use in tracking tree-DOM into receiving ecosystems including forest floor soils and inland waters. Further study is required to develop an understanding of the fate and ecological functions of tree-DOM within receiving ecosystems. Such knowledge will be essential in assessing how ongoing changes to forest cover distributions will impact both soil and aquatic ecosystems.

## AUTHOR CONTRIBUTIONS

AS wrote the paper with input from other authors. AS and JV designed the study. LS collected the samples, analyzed samples and worked up the DOC and CDOM data sets. AS and TD analyzed the samples by FT-ICR MS and worked up the FT-ICR MS data.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/feart. 2017.00022/full#supplementary-material

## REFERENCES


at the interface of terrestrial and aquatic ecosystems. Ecosystems 6, 301–312. doi: 10.1007/s10021-003-0161-9


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Stubbins, Silva, Dittmar and Van Stan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Carbon Dioxide Emissions along the Lower Amazon River

Henrique O. Sawakuchi 1,2 \*, Vania Neu<sup>3</sup> , Nicholas D. Ward4,5, Maria de Lourdes C. Barros <sup>1</sup> , Aline M. Valerio<sup>6</sup> , William Gagne-Maynard<sup>2</sup> , Alan C. Cunha<sup>7</sup> , Diani F. S. Less <sup>7</sup> , Joel E. M. Diniz <sup>7</sup> , Daimio C. Brito<sup>7</sup> , Alex V. Krusche<sup>1</sup> and Jeffrey E. Richey <sup>2</sup>

<sup>1</sup> Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Piracicaba, Brazil, <sup>2</sup> School of Oceanography, University of Washington, Seattle, WA, USA, <sup>3</sup> Instituto Sócio Ambiental e dos Recursos Hídricos, Universidade Federal Rural da Amazônia, Belém, Brazil, <sup>4</sup> Whitney Laboratory for Marine Bioscience, University of Florida, St. Augustine, FL, USA, <sup>5</sup> Marine Sciences Laboratory, Pacific Northwest National Laboratory, Sequim, WA, USA, <sup>6</sup> Departamento de Sensoriamento Remoto, Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil, <sup>7</sup> Departamento de Meio Ambiente e Desenvolvimento, Universidade Federal do Amapá, Macapá, Brazil

#### *Edited by:*

Marta Álvarez, Instituto Español de Oceanografía, Spain

#### *Reviewed by:*

Hugh Daigle, University of Texas at Austin, USA Katlin Louise Bowman, University of California, Santa Cruz, USA

*\*Correspondence:*

Henrique O. Sawakuchi riqueoliveira@yahoo.com.br

#### *Specialty section:*

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

*Received:* 03 October 2016 *Accepted:* 02 March 2017 *Published:* 21 March 2017

#### *Citation:*

Sawakuchi HO, Neu V, Ward ND, Barros MdLC, Valerio AM, Gagne-Maynard W, Cunha AC, Less DFS, Diniz JEM, Brito DC, Krusche AV and Richey JE (2017) Carbon Dioxide Emissions along the Lower Amazon River. Front. Mar. Sci. 4:76. doi: 10.3389/fmars.2017.00076 A large fraction of the organic carbon derived from land that is transported through inland waters is decomposed along river systems and emitted to the atmosphere as carbon dioxide (CO2). The Amazon River outgasses nearly as much CO<sup>2</sup> as the rainforest sequesters on an annual basis, representing ∼25% of global CO<sup>2</sup> emissions from inland waters. However, current estimates of CO<sup>2</sup> outgassing from the Amazon basin are based on a conservative upscaling of measurements made in the central Amazon, meaning both basin and global scale budgets are likely underestimated. The lower Amazon River, from Óbidos to the river mouth, represents ∼13% of the total drainage basin area, and is not included in current basin-scale estimates. Here, we assessed the concentration and evasion rate of CO<sup>2</sup> along the lower Amazon River corridor and its major tributaries, the Tapajós and Xingu Rivers. Evasive CO<sup>2</sup> fluxes were directly measured using floating chambers and gas transfer coefficients (k600) were calculated for different hydrological seasons. Temporal variations in pCO<sup>2</sup> and CO<sup>2</sup> emissions were similar to previous observations throughout the Amazon (e.g., peak concentrations at high water) and CO<sup>2</sup> outgassing was lower in the clearwater tributaries compared to the mainstem. However, k600-values were higher than previously reported upstream likely due to the generally windier conditions, turbulence caused by tidal forces, and an amplification of these factors in the wider channels with a longer fetch. We estimate that the lower Amazon River mainstem emits 20 Tg C year−<sup>1</sup> within our study boundaries, or as much as 48 Tg C year−<sup>1</sup> if the entire spatial extent to the geographical mouth is considered. Emissions from the Xingu and Tapajós lower tributaries contribute an additional 2.3 Tg C year−<sup>1</sup> . Including these values with updated basin scale estimates and estimates of CO2 outgassing from small streams we estimate that the Amazon running waters outgasses as much as 0.95 Pg C year−<sup>1</sup> , increasing the global emissions from inland waters by 15% for a total of 2.45 Pg C year−<sup>1</sup> . These results highlight the lower reaches of large rivers as a missing gap in basin-scale and global carbon budgets. In the case of the Amazon River, the previously unstudied tidally-influenced reaches contribute to 5% of CO2 emissions from the entire basin.

Keywords: GHG emission, CO<sup>2</sup> emission, Lower Amazon, CO<sup>2</sup> outgassing, river, global CO<sup>2</sup> emission

## INTRODUCTION

Rivers are no longer viewed as passive conduits from land to sea but, rather, play an active role in processing organic carbon derived from land and returning it to the atmosphere as carbon dioxide (CO2) (Cole et al., 2007; Battin et al., 2009). The remaining organic and inorganic carbon that is exported to the coastal ocean is further processed, released to the atmosphere, or stored in marine waters and sediments (Medeiros et al., 2015; Ibanhez et al., 2016). Streams, rivers, and lakes have most recently been estimated to emit 2.1 Pg C year−<sup>1</sup> to the atmosphere (Raymond et al., 2013), increasing from past estimates of 1.4 Pg C year−<sup>1</sup> (Tranvik et al., 2009), and 0.8 Pg C year−<sup>1</sup> (Cole et al., 2007). Although data coverage is more sparse, wetlands, which were not included in estimates by Raymond et al. (2013), emit another ∼2.1 Pg C year−<sup>1</sup> (Aufdenkampe et al., 2011). These combined estimates, along with storage and export terms, imply that roughly 5.7 Pg C year−<sup>1</sup> is transported through inland waters, with nearly 75% of this carbon being returned to the atmosphere (Le Quéré et al., 2015). Tropical regions have been identified as hotspots for aquatic CO<sup>2</sup> outgassing, representing ∼75% of global emissions, yet they are under-represented in global datasets, particularly with respect to direct measurements of fluxes and concentrations, which allows quantification of gas transfer velocity values that are used in regional and global models (Regnier et al., 2013; Wehrli, 2013).

The Amazon River is the largest river system in the world, responsible for 20% of the fresh water discharge to world's oceans and 25% of the emissions of CO<sup>2</sup> from inland waters to the atmosphere, globally (Richey et al., 2002; Raymond et al., 2013). The influence of the Amazon River on primary productivity in the Atlantic Ocean can be seen from space, driving a net uptake of CO<sup>2</sup> in the plume (Subramaniam et al., 2008). The source of dissolved CO<sup>2</sup> in large river systems shifts from headwaters to higher order rivers/streams. In small headwater streams the primary source is subsurface flow from riparian soils (Johnson et al., 2008). The relative contribution from soil respiration decreases compared to in situ production via microbial respiration as stream order increases (Butman and Raymond, 2011). The breakdown of young terrestrially-derived organic carbon (OC) by heterotrophic river microbes is thought to be the primary source of CO<sup>2</sup> in the Amazon River mainstem (Mayorga et al., 2005; Ward et al., 2013, 2016), although plant respiration and OC decomposition in floodplains also contribute to CO<sup>2</sup> supersaturation (Abril et al., 2014).

The majority of geochemical studies in the Amazon River have focused on the central Amazon, which represents about 30% of the 6 million km<sup>2</sup> drainage basin (Hedges et al., 1986, 2000; Moreira-Turcq et al., 2003). For example, the current basinscale CO<sup>2</sup> budget for the Amazon River is based on aereal outgassing rates determined for this corridor, and outgassing rates were conservatively assumed to be 50% less in unstudied regions outside of the central corridor (Richey et al., 2002). The lower reaches of the Amazon River, between the historic gauging station, Óbidos, and ∼800 km downstream to the mouth, have not been included in current basin-scale budgets. This represents ∼13% of the basin's total surface area (in terms of land, not water surfaces) and is characterized by expansive floodplains and flooded forests, which likely provide large inputs of OC and CO<sup>2</sup> to the river. In fact, 75% of the particulate OC load is lost between Óbidos and the mouth largely due to degradation, while dissolved OC concentrations slightly increase due to constant inputs from the watershed and floodplains that balance OC degradation (Seidel et al., 2015; Ward et al., 2015). Tidal effects can be detected more than halfway upstream to Óbidos with flow completely reversing near the mouth. These forces increase water residence time and along with strong winds and wide channels (2–15 km) with a long fetch, create rough water surface conditions that likely promote CO<sup>2</sup> degassing. Including a quantitative evaluation of CO<sup>2</sup> emissions in this unique reach of the river is critical for constraining the basin scale carbon budget, which directly influences global estimates.

This study provides the first detailed evaluation of CO<sup>2</sup> concentrations and fluxes along the lower Amazon River and its major tributaries, the Xingu and Tapajós rivers. Direct measurements of CO<sup>2</sup> outgassing were made with floating domes for each hydrologic period (i.e., low, rising, high, and falling water) from 2014 to 2016 along with measurements of CO<sup>2</sup> concentrations and calculations of gas transfer velocities. Total CO<sup>2</sup> evasion was estimated for three discreet sections of the lower river: (1) the Amazon River main channel from Óbidos to the downstream study boundaries near Macapá, (2) the lower regions of the Tapajós and Xingu tributaries, and (3) the extended region from Macapá to the actual geographic river mouth. These estimates were used to calculate a range of updated basin scale CO<sup>2</sup> outgassing budgets based on previous estimates (Richey et al., 2002; Rasera et al., 2013), which were compared with global budgets.

## METHODS

#### Study Area

A series of four expeditions were performed from 2014 to 2016 along the lower reach of the Amazon River, from Óbidos, the furthest downstream gauging station in the Amazon River mainstem, to the last two well-constrained channels near the river mouth at Macapá, ∼650 km downstream from Óbidos (**Figure 1**-Area 1). Tides drive a ∼3 m semi-diurnal variation in river depth, completely reversing river flow with no salinity intrusion. The river continues to widen and channelize between large islands an additional 150 km downstream of Macapá before being entirely disconnected from land and the riparian zone/floodplains (**Figure 1**-Area 2). The water entering the ocean can remain completely fresh at the surface as much as 60 km offshore from this point (**Figure 1**-Area 3; Molinas et al., 2014).

Between Óbidos and the ocean, an additional ∼20% discharge is added by lowland tributaries, primarily from the Tapajos and Xingu rivers, which are the largest clear water tributaries in the Amazon basin (Sioli, 1985; Mayorga and Aufdenkampe, 2002). The lower Amazon River, from Óbidos to the river mouth, is characterized by an intricate mixture of large channels, clear water tributaries, floodplain lakes and flooded forests, representing ∼13% of the total Amazon River drainage basin.

Measurements of pCO<sup>2</sup> and fluxes were carried out at different sites along the Amazon River main channel—Óbidos; Almeirim, which is halfway to the river mouth; and the north and south channels near Macapá—along with measurements near the outflow of the Tapajós and Xingu rivers (**Table 1**, **Figure 1**).

#### Partial Pressure of CO<sup>2</sup> and Flux Measurements

Measurements of the partial pressure of CO<sup>2</sup> (pCO2), CO<sup>2</sup> fluxes (FCO2), and calculations of gas transfer velocity (k) were made during each hydrological season (low, rising, high, and falling water). For Amazon River mainstem sites measurements were made at three sub-stations distributed equidistantly across the channel profile (e.g., center and left/right margin). Only a single station was sampled in the Tapajós and Xingu tributaries.

pCO<sup>2</sup> was determined using a plexiglas equilibration chamber filled with glass beads to enhance gas transfer interfaced to an Infrared Gas Analyzer (LICOR Instruments, LI-820) (Frankignoulle et al., 2001; Abril et al., 2014). Briefly, a submersible pump delivered approximately 1.5 L of water per minute flowing from the top to the bottom of the chamber, leaving approximately 0.4 L of internal air headspace. The equilibrated headspace was circulated from the top to the bottom of the chamber through a desiccating water trap, filled with Drierite for drying the air before enter in the gas analyzer using a small diaphragm pump (AS-200; Spectrex), at a flow rate of 150 mL min−<sup>1</sup> . Values were recorded once pCO<sup>2</sup> readings remained stable.

Evasive CO<sup>2</sup> fluxes were directly measured from the river surface using a light weight floating chamber made of polypropylene and covered with reflective alumina tape to avoid internal heating (Galfalk et al., 2013). A floating collar made with a Styrofoam rod was attached around the chamber covering as little area as possible and positioned to leave the chamber edges submersed 2.5 cm into the water. The chamber was round with a volume and area of 7,500 ml and 0.071 m<sup>2</sup> , respectively, and was interfaced to a second portable Infrared Gas Analyzer (LICOR Instruments, LI-820) using the same type of air pump and water trap as the equilibration chamber. Flux measurements started only after atmospheric air concentration readings by the analyzer were stable. The chamber was deployed for approximately 5 min and then lifted up to equilibrate with atmospheric air concentrations prior to the next measurement. On average three measurements were carried out for each location while drifting with the boat to avoid creating extra turbulence.

The flux of CO<sup>2</sup> across the air-water interface (FCO2, mol m−<sup>2</sup> s −1 ) can be described by the following equation:

$$F\_{CO\_2} = \left(\frac{d\left(pCO\_2\right)}{dt}\right)\left(\frac{V}{RT\_KA}\right) \tag{1}$$

where d(pCO2)/dt is the slope of the CO<sup>2</sup> accumulation in the chamber (µatm h−<sup>1</sup> ), V is the chamber volume (m<sup>3</sup> ), T<sup>K</sup> is air temperature (in degrees Kelvin, K), A is the surface area of the chamber at the water surface (m<sup>2</sup> ), and R is the gas constant (L atm K−<sup>1</sup> mol−<sup>1</sup> ) (Frankignoulle, 1988). Measurements were TABLE 1 | CO<sup>2</sup> fluxes to atmosphere (FCO2), partial pressure of CO<sup>2</sup> in the water (pCO2), gas transfer velocity (k600) measurements according to site and season (mean ± SD) and measurements of mean depth (z), water velocity (w), discharge (Q) and wind speed (U10).


\*Outliers removed for statistical analysis.

discarded if the r 2 -value from the slope of pCO<sup>2</sup> vs. time was lower than 0.90.

### Gas Transfer Velocity (*k*) Estimation

Despite the difficulty of directly measuring piston velocity (k), it can be estimated by the relation between the diffuse flux and the difference among surface water and air-equilibrium concentrations given by the followering equation (Wanninkhof et al., 2009):

$$F\_{CO\_2} = k \cdot (C\_w - C\_0) \tag{2}$$

where F is flux (mol m−<sup>2</sup> d −1 ), k the piston velocity (m d−<sup>1</sup> ), C<sup>w</sup> is the concentration of CO<sup>2</sup> measured in the water (mol m−<sup>3</sup> ), and C<sup>0</sup> is the CO<sup>2</sup> concentration at the water surface in exchange with the atmosphere, where Cw,<sup>0</sup> is given by the CO<sup>2</sup> partial pressure and solubility (i.e., Cw,0=K<sup>0</sup> × pCO2w,0). Thus, we have:

$$F\_{CO\_2} = k \cdot K\_0 \left( pCO\_{2\_W} - pCO\_{20} \right) \tag{3}$$

where <sup>K</sup><sup>o</sup> (mol m−<sup>3</sup> Pa−<sup>1</sup> ) is the aqueous-phase solubility of CO<sup>2</sup> as a function of temperature, pCO2w and pCO2<sup>0</sup> are the partial pressures (Pa) of CO<sup>2</sup> in water and air inside the chamber, Then, substituting into Equation 1 and integrating partial pressure from time i to f, this equation can be rewritten as:

$$k = \frac{V}{A \cdot \alpha} \ln \left(\frac{pCO\_{2\text{w}} - pCO\_{2i}}{pCO\_{2\text{w}} - pCO\_{2f}}\right) / \left(t\_f - t\_i\right) \tag{4}$$

where V is the chamber volume (cm<sup>3</sup> ), A is the chamber area (cm<sup>2</sup> ), α is the Ostwald solubility coefficient (dimensionless), t is time (h), and the subscripts i and f refer to initial and final times and partial pressures. The Ostwald solubility coefficient can be calculated from K<sup>0</sup> as a function of temperature as described by Wanninkhof et al. (2009), given by K<sup>0</sup> = α(RTw) −1 , where R (m<sup>3</sup> Pa K <sup>−</sup><sup>1</sup> mol−<sup>1</sup> ) is the ideal gas constant and T<sup>w</sup> (K) is the water temperature.

After solving k, flux measurements, water and air concentration of CO<sup>2</sup> using the equation 2, and later normalized into k600-values using the followering equations (Jahne et al., 1987; Wanninkhof, 1992; Alin et al., 2011) derived from Equations 1–2:

$$k\_{600} = \left(k\_T \left(\frac{600}{\text{Sc}\_T}\right)^{-0.5}\right)^{-0.5} \tag{5}$$

where k<sup>T</sup> is the measured k-value at in situ temperature (T), Sc<sup>T</sup> is the Schmidt number calculated as a function of temperature (T):

Sc<sup>T</sup> = 1911.1 − 118.11 T + 3.4527 T <sup>2</sup> − 0.041320 T 3 (6)

### Hydrological and Climatological Characterization

Discharge, water velocity, and river depth were measured across the Amazon River channel sites during all cruises using a Sontek River Surveyor M9 Portable nine-beam 3.0/1.0/0.5 MHz acoustic Doppler Current Profiler (ADCP). Cross-channel ADCP transects were performed three to four times in the upstream sites with no tidal influence and 8–11 times through a complete tidal cycle (10–13 h) in sites with tidal influence in order to assess river velocity over the span of a tidal cycle and accurately calculate the total Amazon River discharge (Ward et al., 2015). Discharge measurements were not conducted in the Xingu River neither during high water season at Tapajós River. For the purpose of qualitative comparisons, we obtained data on average discharge, at the time of sampling or from long-term monthly or weekly averages from the nearest monitoring station(s) to fill these gaps. The hydrological stations searched were Óbidos at the Amazon River, Itiatuba at Tapajós River, and Altamira at the Xingu River. For sites in the Amazon, hydrological data came from the Brazilian national water agency web site (Agência Nacional de Águas, http://www.snirh.gov.br/hidroweb/).

Wind speed was measured during falling and rising water cruises with a weather station (Onset HOBO) installed on the boat or with a handheld weather station (Kestrel 5500). Wind speed was normalized to 10 m height (U10) according to Alin et al. (2011) using the following equation:

$$
\bar{u}\_z = \left(\frac{u\_\*}{\kappa}\right) \ln\left(\frac{z}{z\_0}\right) \tag{7}
$$

where u¯<sup>z</sup> is the is mean wind speed (m s−<sup>1</sup> ) at the height z, u<sup>∗</sup> is friction velocity (m s−<sup>1</sup> ), κ is von Karman's constant (∼0.40), and z<sup>0</sup> is roughness length (10−<sup>5</sup> m). Friction velocity was first calculated by rearranging equation (6) to solve for u<sup>∗</sup> and using the mean wind speed measured at 1.5 m as u¯z. Monthly historical wind data for comparison was obtained from the National Institute of Meteorology web site (Instituto Nacional de Meteorologia, http://www.inmet.gov.br).

Water temperature was measured with a Thermo-Orion 290APlus probe submerged in a continuously overflowing graduated cylinder.

### Annual CO<sup>2</sup> Emissions from the Lower Amazon River

Data for the Tapajós and Xingu rivers were only acquired at one station near their river mouth. In both rivers the approximately last 100 km area is characterized as Rias, which have lake-like sedimentary dynamics (Archer, 2005). Thus, we only included this area of the tributaries for the outgassing budget for the lower Amazon River. We divided the main channel into two zones: (Area 1) our study boundaries from Óbidos to Macapá, which has a surface area of 7,118 km<sup>2</sup> , and (Area 2) the region extending from Macapá to the geographical river mouth, which has an additional surface area of 11,261 km<sup>2</sup> (**Figure 1**). Although it has never been studied, we assume that Area 2 will have similar geochemical characteristics as the region near Macapá considering there are still inputs from land and the Amazon River water discharged to the ocean still remains completely fresh at the surface up to ∼100 km further offshore (**Figure 1**-Area 3; Molinas et al., 2014).

The CO<sup>2</sup> outgassing budget for Area 1 was determined by multiplying the average FCO<sup>2</sup> measured along the study boundaries by the surface area. For Area 2 we used the seasonal average FCO<sup>2</sup> measured across the North and South Macapá stations combined. FCO<sup>2</sup> results from each cruise were applied to a 3-month period for the particular hydrologic period and the sum of these values was used to represent annual estimates.

The surface area of the lower Amazon River main channel and lower regions of the Tapajós and Xingu rivers were estimated using the mesh generation tool of SisBaHiA (Base System for Environmental Hydrodynamic; www.sisbahia.coppe.ufrj.br). The finite element methods and mesh generation techniques used in SisBaHiA is detailed in Ern and Guermond (2004). Basically, the mesh is composed of biquadratic quadrilateral elements with specific area and the sums represent the total area of the studied surface.

#### Statistical Analyses

Statistical evaluations of the FCO<sup>2</sup> in the Lower Amazon River were done through non-parametric analysis due to the lack of normal distribution of the data (Shapiro–Wilk, p < 0.05, rejecting the null hypothesis of normality). Evaluation of the differences between broad hydrological settings such as tidallyinfluenced and tributaries vs. upstream mainstem locations were carried out by the Mann-Whitney test, and differences between sites were assessed by the Kruskal-Wallis test. The relationship among physical characteristics were done using Spearman correlation test. All analyses were performed with R (http://www.r-project.org).

## RESULTS

## *F*CO<sup>2</sup> Fluxes and *p*CO<sup>2</sup>

The average pCO<sup>2</sup> and FCO<sup>2</sup> including all seasons and sites measured in the lower Amazon River and its tributaries was 2914 ± 1768 µatm and 6.31 ± 5.66 µmol m−<sup>2</sup> s −1 , respectively. Tributaries had significantly lower pCO<sup>2</sup> and FCO<sup>2</sup> compared to the mainstem stations (Mann-Whitney, p < 0.05) with values of 1322 ± 1545 µatm and 2.4 ± 2.3 µmol m−<sup>2</sup> s −1 , respectively, in the tributaries and 3218 ± 1656 µatm and 6.9 ± 5.8 µmol m−<sup>2</sup> s −1 , respectively, in the mainstem (**Figure 2**). Both tributaries are considered clear water and are characterized by low suspended sediment loads and high primary productivity (Ward et al., 2015), which results in the fixation of dissolved CO<sup>2</sup> and reduction in pCO<sup>2</sup> and FCO2. For this reason, the tributaries were considered separately from the mainstem for further comparisons.

The outlier data points observed in the Xingu River (**Figure 2**) for both pCO<sup>2</sup> and FCO<sup>2</sup> occurred during the high water season (**Table 1**) likely because of the delivery of different source of water coming from a floodplain area that discharges just upstream of the sampling station during the high water season. For example, pCO<sup>2</sup> values as high as 10,000 ppm were observed at the confluence of the Xingu River and the Jaraçu River, which connects the Amazon and Xingu rivers through an extensive floodplain network (Ward et al., 2016). The intrusion of high suspended sediment water from a small channel was clearly observed during the high water sampling. Thus, the sampled water was a mixture of Xingu River water and Amazon River water fed through floodplains. Since this data point does not represent pure Xingu River water, the value was not considered in further comparisons.

There was no significant difference in pCO<sup>2</sup> and FCO<sup>2</sup> between the two tributaries (Mann-Whitney, p > 0.05), and pCO<sup>2</sup> was also not statistically different between the mainstem stations. Nevertheless, FCO<sup>2</sup> was significantly different between stations in the mainstem (Kruskal-Wallis, p < 0.05), with higher fluxes measured further upstream (**Figure 2**). Tidal and nontidally influenced sites in the mainstem and tributaries were compared in order to evaluate the effects of tides play on pCO<sup>2</sup> and FCO2. Tidally-influenced stations in the mainstem had slightly lower pCO<sup>2</sup> and FCO<sup>2</sup> than the non-tidal stations (Mann-Whitney, p < 0.05). No difference in pCO<sup>2</sup> or FCO<sup>2</sup>

was observed between the Tapajos (non-tidal) and Xingu (tidal) rivers.

Seasonal variation was recognized for both pCO<sup>2</sup> and FCO<sup>2</sup> in the mainstem (Kruskal-Wallis, p < 0.001) with the highest values observed during the high water season and lowest values during the lower water season (**Figure 3**). pCO<sup>2</sup> was substantially higher in the Tapajós River during the high water season (**Table 1**). Although high pCO<sup>2</sup> and FCO<sup>2</sup> during high water in the Xingu River was attributed to floodplains fed by Amazon River source water, this was not the case in the Tapajós River (i.e., only pure Tapajós River water was present at the sampling station). However, the tributaries were only sampled at one location in the center of the channel, which limited our ability to make statistical inferences regarding seasonal differences.

#### Evaluation of *k*<sup>600</sup>

The average k<sup>600</sup> estimated for all stations was 33.60 ± 15.72 cm h −1 . There were no statistically significant differences between the mainstem and tributaries or non-tidal and tidal stations (**Figure 4**; Mann-Whitney, p > 0.05). The average k<sup>600</sup> for the mainstem sites was 33.71 ± 15.63 cm h−<sup>1</sup> compared to 32.58 ± 19.10 cm h−<sup>1</sup> for the tributaries. Considering tidal vs. nontidal stations, average <sup>k</sup>600-values were 33.80 <sup>±</sup> 16.97 cm h−<sup>1</sup> and 33.53 ± 15.50 cm h−<sup>1</sup> , respectively.

The primary control on k<sup>600</sup> appeared to be seasonality considering the lack of spatial differences (**Figure 5**). We

observed higher values of k<sup>600</sup> during the low water season in the mainstem (Kruskal-Wallis, p < 0.05), and during the rising water period in the tributaries (**Figure 3**). For the mainstem this pattern is in agreement with the historical monthly average of wind speed (**Figure 6**).

## Environmental Characterization and Correlations with *F*CO2, *p*CO2, and *k*<sup>600</sup>

The mean annual discharge (Q) at the mouth of the Amazon River to the ocean (i.e., the sum discharge measured near the mouth across the Macapá South and North channels, which integrate the discharge from all non-measured tributaries upstream) was 238,997 m<sup>3</sup> s −1 , ranging from 156,858 to 344,680 m<sup>3</sup> s <sup>−</sup><sup>1</sup> during low and high water periods, respectively (**Figure 7**). Water speed (w) for the mainstem sites ranged from 45 to 147 cm s−<sup>1</sup> .

Wind speed (U10) was measured in the river during the falling and rising water seasons and averaged 4.00 ± 1.95 m s −1 , ranging from 1.21 to 10.65 m s−<sup>1</sup> . Mean values measured at each station in each season are shown on **Table 1**. To better assess the annual variability of wind speed in the lower Amazon River we evaluated the historical monthly average using data from 2000 to 2016 for stations located in cities along the lower Amazon River monitored by the Brazilian Institute of Meteorology (INMET). Higher wind speeds and higher seasonal variation were observed in the stations further downstream from Óbidos.

A Spearman correlation test was performed considering all stations to evaluate the relationship between pCO2, FCO2, and k<sup>600</sup> with hydrological parameters and wind. A correlation matrix was generated for inter-comparisons of all these parameters among each other (**Table 2**). The strongest positive correlation was observed between pCO<sup>2</sup> and FCO<sup>2</sup> (r = 0.8, p < 0.001). However, pCO<sup>2</sup> was also positively correlated with all three measured hydrological parameters (**Table 2**). FCO<sup>2</sup> was correlated with depth (z) and k600, while discharge (Q) was correlated with water speed, which in turn was correlated with depth. Unbinned wind speed did not present any direct correlation with any parameter considered in this study. However, average binned <sup>k</sup><sup>600</sup> for <sup>U</sup><sup>10</sup> bins of 0.5 m s−<sup>1</sup> presented a stronger positive correlation (**Figure 8**; Spearman, r = 0.7, p < 0.05).

#### Upscaling CO<sup>2</sup> Emissions from the Lower Amazon River

The total flux of CO<sup>2</sup> from the main channel of the lower Amazon River was calculated for two zones (Figure 1) for each 3-month hydrologic period and annually (**Table 3**). The most conservative estimates (Area 1) only included the boundaries of this study, from Óbidos to Macapá, which had an average wetted surface area of 7,118 km<sup>2</sup> . The flux of CO<sup>2</sup> from Area 1 was 20 Tg C year−<sup>1</sup> . Area 2 extends from Macapá to the region of the mouth where the connection to land terminates, which had an additional surface area of 11,261 km<sup>2</sup> . Area 2 had a total CO<sup>2</sup> flux of 28 Tg C year−<sup>1</sup> , based on an extrapolation of average values measured across the north and south Macapá channels (**Table 3**). The sum of fluxes for these two zones, or the total emissions from Óbidos to the actual river mouth was 48 Tg C year−<sup>1</sup> . The emissions from the lower Tapajós and Xingu rivers were 1.40 and 0.86 Tg C year−<sup>1</sup> , respectively. Including these two tributaries to the budget would add more 2.26 Tg C year−<sup>1</sup> .

#### DISCUSSION

In this study we measured pCO<sup>2</sup> and CO<sup>2</sup> fluxes from the water to atmosphere along with discharge, water current velocity and wind speed in the Lower Amazon River channel, from the last gauging station at Óbidos to the mouth. The two major clear water tributaries of the Amazon River had lower pCO<sup>2</sup> and FCO<sup>2</sup> values than the Amazon River mainstem, which is consistent with other studies (Alin et al., 2011; Rasera et al., 2013). This was expected based on the lower suspended sediment loads, which enables high rates of primary productivity as indicated by enhanced levels of Chlorophyll-a (Sioli, 1985; Mayorga and Aufdenkampe, 2002; Ward et al., 2015).

Previous pCO<sup>2</sup> measurements for sites within our study boundaries ranged from 1600 to 6037 µatm in the mainstem and from 70 to 1070 µatm in the tributaries, agreeing well with our observations (Alin et al., 2011; Borges et al., 2015). Direct FCO<sup>2</sup> measurements in the lower Amazon River mainstem and Tapajós River have only been reported by Alin et al. (2011) and ranged from 1.87 to 10.62 µmol m−<sup>2</sup> s −1 in the mainstem and from 0.04

to 6.36 µmol m−<sup>2</sup> s −1 in the Tapajós River, which is also similar to our measurements. The seasonal trend observed in pCO<sup>2</sup> in the lower Amazon River follows the hydrologic cycle as observed by Richey et al. (2002), with maximum CO<sup>2</sup> concentrations during high water and minimal concentrations during the lower water season. FCO<sup>2</sup> measurements followed the same trend since it was strongly correlated with pCO<sup>2</sup> (**Table 2**).

Despite the similar ranges of pCO<sup>2</sup> and FCO<sup>2</sup> our range of k<sup>600</sup> was considerably higher than Alin et al. (2011). This can mostly be attributed to the higher wind speeds recorded in the river during our measurements compared with Alin et al. (2011). The furthest downstream station sampled by Alin et al. (2011) was near Santarém, which is roughly 500 km upstream of Macapá. We observed a downstream increase in wind speeds (**Figure 6**), which should lead to higher k<sup>600</sup> values considering the typical correlation between wind speed and k<sup>600</sup> in large rivers (Borges et al., 2004b; Alin et al., 2011; Rasera et al., 2013). Although we did not find any correlation between direct k<sup>600</sup> calculations and the hydrological parameters or wind, the average binned <sup>k</sup><sup>600</sup> for <sup>U</sup><sup>10</sup> bins of 0.5 m s−<sup>1</sup> presents a stronger positive correlation here and in the aforementioned studies (**Figure 8**). Channel width and the fetch length (i.e., the distance

traveled by wind or waves across open water) also increases downstream, which amplifies the effects of wind, waves, and currents on surface water texture and turbulence. For example, the Amazon River is characterized by large sweeping curves and relatively narrow channels upstream of Santarém, which limits wave formation, whereas the main channel remains fairly straight for ∼250 km between Santarém and Almeirim, and remains straight again after a slight turn to the northwest for 2,250 km between Almeirim and Macapá.

The gas exchange coefficient and its variability within a system is among the most important factors controlling CO<sup>2</sup> emissions from different parts of a large basin (Raymond and Cole, 2001; Borges et al., 2004a,b; Alin et al., 2011; Striegl et al., 2012). In shallower streams and rivers where turbulence is high due to bottom friction, k<sup>600</sup> can be expressed as a function of the water flow speed and depth, where shallower and faster streams tend to have higher k600-values than slower streams (Raymond and Cole, 2001). In the Yukon basin, tributaries had higher k-values for CO<sup>2</sup> than the mainstem, and in the mainstem k increased

TABLE 2 | Spearman correlation matrix showing the r-values in the top right side of the diagonal and adjusted p-values in the bottom left side in italic.


Bold numbers indicate significant correlations. Discharge (Q), wind speed (U10), water velocity (w) and mean depth (z).

downstream (Striegl et al., 2012). In the Amazon, a similar trend was observed where k<sup>600</sup> was higher in rivers with narrower channels (<100 m wide) (Rasera et al., 2008; Alin et al., 2011).

In larger rivers and estuaries the main driving factors controlling k<sup>600</sup> is wind and water currents, which in turn are



Estimates for tributaries only include the lower reaches measured in this study.

attributed to a balance between several factors (e.g., discharge, tidal range and fetch) (Alin et al., 2011). In the Scheldt estuary water current is an important factor controlling k<sup>600</sup> but its significance decreases with increasing wind speed (Borges et al., 2004b). In the Amazon and Mekong rivers, water currents are generally higher than in estuaries but with a more limited fetch and lower wind speeds (Alin et al., 2011). However, as previously mentioned, fetch dramatically increases along the lower Amazon River (**Figure 1**). The factors controlling k in large river basins is a mixture of all those described above. Wind are generally higher than those in headwater streams and current velocities are typically faster than in estuaries (Beaulieu et al., 2012). Additionally, the effect of wind on the water turbulence can be related to the water body orientation, shape, size, and direction of wind and water current. When wind and water currents are directionally opposed they can interact synergistically to produce unusually high k-values for any given wind speed (Zappa et al., 2007; Beaulieu et al., 2012). Prevailing winds in the lower Amazon are from the NE, which is the opposite direction of river outflow from Almeirim and Macapá (during an outgoing tide when net discharge is positive).

In large rivers and estuaries, the simple parametrization of k as a function of wind speed tends to be site specific due to local climatological, geomorphological, and hydrological characteristics, implying substantial errors in flux estimates using generic wind based functions from different systems (Borges et al., 2004a; Zappa et al., 2007). Rasera et al. (2013) used more detailed estimates of k600, taking into account spatial variability including the difference between k<sup>600</sup> for small (<100 m wide) and large (>100 m) rivers in the Amazon River, which nearly doubled estimates of basin-scale CO<sup>2</sup> emission from the Amazon basin estimated by Richey et al. (2002).

The lower Amazon River contains many of the features that would lead to very high gas transfer velocities and overall emissions. It is characterized by extremely wide channel(s) that flow in a convergent direction with the Intertropical Convergence Zone (ITCZ) creating a large fetch where the stronger winds can substantially increase turbulence, resulting in higher k<sup>600</sup> as observed in this study. Not only are k-values high, but the lower river also has a very high amount of wetted surface area for CO<sup>2</sup> to escape from, particularly in wide channels near the mouth.


Here we considered two different areas for upscaling annual CO<sup>2</sup> emissions from the lower Amazon River. Our most conservative estimate, including the area from Óbidos to Macapá (Area 1), was 20 Tg C year−<sup>1</sup> . Area 2, which extends to the geographic terminus of the river, is a relatively short distance compared to Area 1, but covers 58% more surface area than the Óbidos to Macapá reach (**Table 3**). Applying the average FCO<sup>2</sup> observed near Macapá, we estimate Area 2 to emit 28 Tg C year−<sup>1</sup> , which combined with the upstream section totalize 48 Tg C year−<sup>1</sup> . Emissions from the Xingu and Tapajós tributaries contribute an additional 2.3 Tg C year−<sup>1</sup> , resulting in a total flux of 51 Tg C year−<sup>1</sup> from the lower Amazon River basin. This estimate for the lower Amazon River, alone, is roughly 53% in magnitude compared to CO<sup>2</sup> emissions from all rivers in the conterminous United States (97 Tg C year−<sup>1</sup> ; Archer, 2005) and ∼12% of past estimates of basin-scale emissions from the Amazon (0.47 Pg C year−<sup>1</sup> ; Butman and Raymond, 2011) and ∼ 12% of past estimates of basin-scale emissions from the Amazon (0.47 Pg C year−<sup>1</sup> ; Richey et al., 2002).

Adding our estimations of the fluxes estimated for Area 1 and the sum of Area 1 and 2 to estimations by Butman and Raymond (2011) increases basin-scale CO<sup>2</sup> outgassing to 0.49 and 0.52 Pg C year−<sup>1</sup> , respectively. A recent re-evaluation of basin-wide outgassing estimates upstream of Óbidos was done using a combination of direct flux measurements and more detailed k-values calculations along with observations by Alin et al. (2011) for tributaries and streams. It was estimated that basin scale CO<sup>2</sup> outgassing upstream from Óbidos was roughly 0.8 Pg C year−<sup>1</sup> (Borges et al., 2004a). First order streams add an additional 0.1 Pg C year−<sup>1</sup> to basin scale CO<sup>2</sup> fluxes in the Amazon basin (Battin et al., 2009). Adding our estimates for the lower river to these values results in basin-wide budgets of 0.92 Pg C year−<sup>1</sup> and 0.95 Pg C year−<sup>1</sup> for Area 1 and the sum of Area 1 and 2, respectively (**Table 4**). Replacing these new estimates for the Amazon in the global CO<sup>2</sup> budget by Borges et al. (2004b) increases the global budget by as much as 18% for a total of 2.48 Pg C year−<sup>1</sup> .

In the case of this updated global budget, the Amazon River represents 38% of global CO<sup>2</sup> emissions. However, the contribution of other tropical rivers to the global budget are likely also underestimated considering that most tropical rivers are even less well-characterized than the Amazon, particularly in the lower reaches, where we've demonstrated that emissions can be high relative to the rest of the basin. Furthermore, we have not included the entirety of the Amazon in our newest budgets. For example, the Tapajós and Xingu rivers were not included due to their large spatial expanse and minimal data coverage. The lower portion of these tributaries, alone (**Figure 1**), emit roughly 2.3 Tg C year−<sup>1</sup> , and these estimates do not encompass their entire surface area nor potentially elevated CO<sup>2</sup> concentrations closer to their headwaters.

Another factor that can lead to an underestimation of basinwide budgets is not including Amazon River water that travels further offshore from Area 2 and along the coastline. For example, the Amazon River can remain unmixed with the ocean as far as 60 km offshore from Area 2 (Molinas et al., 2014). Abril et al. (2014) estimated that only 18% of the CO<sup>2</sup> from a point source would be degassed in a stretch of approximately 150 km downstream in the Amazon River taking into account a k-value of 15 cm h−<sup>1</sup> and water current of 150 cm s−<sup>1</sup> . Thus, it is reasonable to assume that the mouth of the Amazon is the last point source of CO<sup>2</sup> to the Amazon plume, sustaining significant emissions for a significant distance offshore. This region, along with near-shore coastal waters, is not included in any studies of CO<sup>2</sup> cycling in the Amazon River plume in the Atlantic Ocean due to a lack of sampling and terrestrial contamination of remote sensing products by adjacency effect near-shore (Cooley et al., 2007; Subramaniam et al., 2008). We estimate that Area 3 may emit up to an additional 31 Tg C year−<sup>1</sup> , but note that this is a simple calculation based on measurements at Macapá. Further exploration of this offshore area is essential for constraining the total basin-wide CO<sup>2</sup> flux. Although it is too early to confidently incorporate this offshore freshwater region (Area 3) into basinwide budgets, this rough estimation highlights that expansive regions of offshore freshwater plumes may be an important missing gap in aquatic carbon budgets.

### CONCLUDING REMARKS

Numerous studies in the Amazon River have demonstrated its importance on global carbon budgets (Richey et al., 2002; Johnson et al., 2008; Rasera et al., 2013; Scofield et al., 2016). However, a missing gap in the continuum that connects the river network to the ocean has been neglected. Here we show that the lower reaches of the Amazon River are an active area in terms of freshwater CO<sup>2</sup> emissions from the Amazon, and perhaps the world, although, lower river reaches have yet to be adequately studied in other comparable tropical systems. A mixture of vast floodplains, tidally-flooded forests, and the surrounding watershed supplies the lower river with both CO<sup>2</sup> (Abril et al., 2014) and organic carbon, which is broken down extensively by heterotrophic bacteria (Ward et al., 2013, 2016). The downstream expansion of channel width and the alignment of a long fetch opposing the prevailing winds, which also increase toward the mouth, leads to high CO<sup>2</sup> emissions attributed to higher k-values in the lower river.

The enormous surface area of the lower Amazon River (18,379 km<sup>2</sup> ) is slightly less than half of the area of rivers and stream in the conterminous United States (Butman and Raymond, 2011), and similarly emits roughly half as much CO<sup>2</sup> to the atmosphere. This area alone releases an amount of CO<sup>2</sup> in the same order of magnitude than the uptake by the Amazon River plume in the Atlantic Ocean (Kortzinger, 2003; Cooley et al., 2007; Subramaniam et al., 2008). We argue that the region where freshwater extends offshore should be included in basin-scale budgets considering that studies in the plume do not cover this area, however, there is no available data in this region to date. There is a pressing need to perform measurements of pCO<sup>2</sup> and FCO<sup>2</sup> along lower rivers and near-shore plume waters, which are currently large missing gaps in our coverage of regional and global scale carbon budgets.

#### AUTHOR CONTRIBUTIONS

JR, AK, HS, and NW responsible for the conception and design of the work. HS, VN, AV, WG, DL, and JD executed in situ measurements. HS, VN developed the data calculation and interpretation. MB conceived the areal estimations. JR, AK, HS, and NW organized overall project logistics. AC, DL, JD, and

#### REFERENCES


DB organized local logistics. All authors critically revised the manuscript and approved the final submission. All authors also agreed to be accountable for all aspects of the work related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

#### FUNDING

This study was supported by FAPESP Grants 12/51187-0, 2014/21564-2, and 2015/09187-1, NSF DEB Grant #1256724, and NSF IGERT grant DGE-125848.

#### ACKNOWLEDGMENTS

We thank Alexandra Montebello, for assistance in the laboratory at CENA. Gilvan Portela de Oliveira and Geison Carlos Xisto da Silva for logistical support in Macapá, Cica and the crew of the B/M Mirage for contributions made during the river cruises. Open access publication fees were supported by the Gordon and Betty Moore Foundation.


carbon dioxide outgassing from Amazonian rivers. Nature 436, 538–541. doi: 10.1038/nature03880


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Sawakuchi, Neu, Ward, Barros, Valerio, Gagne-Maynard, Cunha, Less, Diniz, Brito, Krusche and Richey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Evaluation of Primary Production in the Lower Amazon River Based on a Dissolved Oxygen Stable Isotopic Mass Balance

William C. Gagne-Maynard<sup>1</sup> , Nicholas D. Ward1, 2, 3 \*, Richard G. Keil <sup>1</sup> , Henrique O. Sawakuchi 1, 4, Alan C. Da Cunha<sup>5</sup> , Vania Neu<sup>6</sup> , Daimio C. Brito<sup>5</sup> , Diani F. Da Silva Less <sup>5</sup> , Joel E. M. Diniz <sup>5</sup> , Aline De Matos Valerio<sup>7</sup> , Milton Kampel <sup>7</sup> , Alex V. Krusche<sup>4</sup> and Jeffrey E. Richey <sup>1</sup>

#### Edited by:

Hongbin Liu, The Hong Kong University of Science and Technology, Hong Kong

#### Reviewed by:

Benoit Thibodeau, University of Hong Kong, Hong Kong Bingzhang Chen, Japan Agency for Marine-Earth Science and Technology, Japan

> \*Correspondence: Nicholas D. Ward nicholas.ward@pnnl.gov

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 02 November 2016 Accepted: 24 January 2017 Published: 07 February 2017

#### Citation:

Gagne-Maynard WC, Ward ND, Keil RG, Sawakuchi HO, Da Cunha AC, Neu V, Brito DC, Da Silva Less DF, Diniz JEM, De Matos Valerio A, Kampel M, Krusche AV and Richey JE (2017) Evaluation of Primary Production in the Lower Amazon River Based on a Dissolved Oxygen Stable Isotopic Mass Balance. Front. Mar. Sci. 4:26. doi: 10.3389/fmars.2017.00026 <sup>1</sup> School of Oceanography, University of Washington, Seattle, WA, USA, <sup>2</sup> Marine Sciences Laboratory, Pacific Northwest National Laboratory, Sequim, WA, USA, <sup>3</sup> Whitney Laboratory for Marine Bioscience, University of Florida, St. Augustine, FL, USA, <sup>4</sup> Centro de Energia Nuclear na Agricultura, Universidade de São Paulo, Piracicaba, Brazil, <sup>5</sup> Departamento de Meio Ambiente e Desenvolvimento, Universidade Federal do Amapá, Macapá, Brazil, <sup>6</sup> Instituto Sócio Ambiental e dos Recursos Hídricos, Universidade Federal Rural da Amazonia, Belem, Brazil, <sup>7</sup> Departamento de Sensoriamento Remoto, Instituto Nacional de Pesquisas Espaciais, São José dos Campos, Brazil

The Amazon River outgasses nearly an equivalent amount of CO<sup>2</sup> as the rainforest sequesters on an annual basis due to microbial decomposition of terrigenous and aquatic organic matter. Most research performed in the Amazon has been focused on unraveling the mechanisms driving CO<sup>2</sup> production since the recognition of a persistent state of CO<sup>2</sup> supersaturation. However, although the river system is clearly net heterotrophic, the interplay between primary production and respiration is an essential aspect to understanding the overall metabolism of the ecosystem and potential transfer of energy up trophic levels. For example, an efficient ecosystem is capable of both decomposing high amounts of organic matter at lower trophic levels, driving CO<sup>2</sup> emissions, and accumulating energy/biomass in higher trophic levels, stimulating fisheries production. Early studies found minimal evidence for primary production in the Amazon River mainstem and it has since been assumed that photosynthesis is strongly limited by low light penetration attributed to the high sediment load. Here, we test this assumption by measuring the stable isotopic composition of O<sup>2</sup> (δ <sup>18</sup>O-O2) and O<sup>2</sup> saturation levels in the lower Amazon River from Óbidos to the river mouth and its major tributaries, the Xingu and Tapajós rivers, during high and low water periods. An oxygen mass balance model was developed to estimate the input of photosynthetic oxygen in the discrete reach from Óbidos to Almeirim, midway to the river mouth. Based on the oxygen mass balance we estimate that primary production occurred at a rate of 0.39 ± 0.24 g O m<sup>3</sup> d <sup>−</sup><sup>1</sup> at high water and 1.02 ± 0.55 g O m<sup>3</sup> d <sup>−</sup><sup>1</sup> at low water. This translates to 41 ± 24% of the rate of O<sup>2</sup> drawdown via respiration during high water and 67 ± 33% during low water. These primary production rates are 2–7 times higher than past estimates for the Amazon

**66**

River mainstem. It is possible that at high water much of this productivity signal is the result of legacy advection from floodplains, whereas limited floodplain connectivity during low water implies that most of this signal is the result of in situ primary production in the Amazon River mainstem.

Keywords: photosynthesis, tropical rivers, respiration, oxygen, mass balance

#### INTRODUCTION

A considerable body of work over the last several decades has demonstrated that the world's inland waters are generally supersaturated with carbon dioxide (CO2), resulting in a large flux of CO<sup>2</sup> from rivers and lakes to the atmosphere (Cole et al., 2007; Battin et al., 2009). This state of supersaturation is driven by the balance between CO<sup>2</sup> outputs such as primary production, degassing, and export and CO<sup>2</sup> inputs from soils/sediments, the riparian zone, physical and biological decomposition of organic matter (OM) derived from both the terrestrial biosphere (Mayorga et al., 2005; Ward et al., 2013) and aquatic production (Kritzberg et al., 2004; del Giorgio and Pace, 2008), and aquatic plant respiration in floodplains (Abril et al., 2014). Rivers outgas an order of magnitude more CO<sup>2</sup> than lakes on a global scale (Raymond et al., 2013). This is commonly attributed to higher rates of primary production relative to respiration in still waters compared to flowing rivers (Tranvik et al., 2009).

Although it is clear that rivers provide a positive flux of CO<sup>2</sup> to the atmosphere in most cases, quantifying the relative amount of carbon/energy that is produced vs. consumed in an ecosystem is central to understanding its overall productivity and the potential movement of energy up trophic levels (Odum, 1971). For example, rapid rates of carbon remineralization do not preclude productive fisheries. Lower and higher trophic levels depend on both autochthonous and terrestrial carbon/energy sources to thrive, and the balance in food sources to freshwater ecosystems is tightly linked to hydrology (Hoffman et al., 2008). Likewise, recent evidence has shown that the presence of fresh algal biomass and exudates can stimulate the breakdown of less reactive OM such as terrestrially-derived molecules, implying that systems receiving large inputs of both allochthonous and autochthonous OM are likely highly efficient at recycling carbon (Guenet et al., 2014; Bianchi et al., 2015; Ward et al., 2016).

In large turbid river systems such as the Amazon River it has long been assumed that primary production is minimal relative to heterotrophic respiration due to high sediment loads and low light penetration relative to river depth (Fisher, 1979; Devol et al., 1987; Richey et al., 1990; Hedges et al., 2000). In fact, early measurements of primary production based on <sup>14</sup>C uptake indicated rates of primary production that are about an order of magnitude lower than respiration in the Amazon River mainstem, while clearwater tributaries in the basin's lowlands exhibited significantly higher rates of primary production (Wissmar et al., 1981; Benner et al., 1995). However, several studies have more recently found evidence for a higher relative contribution of primary production (and subsequent breakdown of algal OM) to ecosystem metabolism in the Amazon River based on the stable isotopic signature of dissolved oxygen (δ <sup>18</sup>O-O2) and the presence of algal biomarkers in the mainstem (Quay et al., 1995; Mortillaro et al., 2011; Ellis et al., 2012). Similar observations of high rates of primary production and algal abundance have also been made in the mainstem of the Congo River, a similar large tropical system (Descy et al., 2017).

These emerging results suggest that there is a large gap between the conventional understanding of production in turbid tropical rivers, and what is actually occurring in the environment. Gross primary production and respiration rates are critical components in understanding the production, transformation and export of OM and other biogeochemically important nutrients in aquatic ecosystems. The balance between these two is a key measure of the net metabolic state of the ecosystem. As rivers worldwide are saturated with respect to atmospheric CO<sup>2</sup> and outgas this CO<sup>2</sup> at a massive scale, understanding the exact dynamics of photosynthesis and respiration remain key to understanding the role of river systems in global biogeochemical cycles and have further implications on ecosystem productivity as previously described (Cole and Caraco, 2001; Butman and Raymond, 2011).

The Amazon River is the largest source of freshwater to the world's oceans and outgases 0.5 Pg C year−<sup>1</sup> , which is roughly 25% of recent global estimates (Richey et al., 2002; Raymond et al., 2013). The Amazon River ecosystem also supports the world's most diverse fishery with over 2500 unique species and is among the primary sources of protein for the region (Junk et al., 2007). The massive contribution of the Amazon basin to global biogeochemical cycling and regional economics makes it an ideal setting for evaluating large-scale processes. The lower Amazon River, from Óbidos to the river mouth, is an understudied component of the Amazon basin, but is a region where the interplay between primary production and heterotrophic respiration likely rapidly evolves due to a gradient in environmental conditions (Ward et al., 2016). Most historic measurements of the Amazon River have only been made as far downstream as Óbidos, roughly 800 km from the mouth of the river (Hedges and Clark, 1986; Quay et al., 1992; Moreira-Turcq et al., 2013). The lower Amazon watershed represents ∼13% of the total drainage basin and is composed of a mixture of wide channels, flooded forests, clear water tributaries and floodplain lakes (Ward et al., 2015). The lower reach of the river is also subject to tidal effects, increasing the residence time of the water and increasing connectivity between flooded forests and the mainstem.

Previous studies have used δ <sup>18</sup>O-O<sup>2</sup> values to estimate gas exchange (Jamieson et al., 2012) and estimate net metabolism (i.e., the ratio of respiration to photosynthesis) in steady-state systems (Quay et al., 1995; Bocaniov et al., 2012) as well as non-steady-state systems (Tobias et al., 2007; Holtgrieve et al., 2010). Past estimates of primary production in the Amazon River are based on a steady-state assumption (i.e., no diel variability in δ <sup>18</sup>O-O2) (Quay et al., 1995; Ellis et al., 2012); however, if primary production is actually occurring at appreciable rates in the mainstem of the river this steady-state assumption should not hold true.

This study seeks to examine the occurrence and magnitude of primary production within the lower Amazon River relative to respiration, from the historic downstream gaging station, Óbidos, to the river mouth. Measurements of the concentration and stable isotopic composition of dissolved oxygen made along the lower river and throughout a diel cycle were used to construct an oxygen mass balance for a discreet reach of the system in an effort to determine the relative contribution of primary production and respiration to ecosystem metabolism.

#### MATERIALS AND METHODS

Measurements of the concentration and stable isotopic composition of dissolved oxygen were made along the lower Amazon River and coupled with measurements of microbial respiration rates (Ward et al., unpublished) and gas transfer velocities (Sawakuchi et al., unpublished) made during the same study period to construct an oxygen mass balance for the lower river. The stable isotopic composition of water (δ <sup>18</sup>O-H2O) was also measured for calculations associated with δ <sup>18</sup>O-O2.

#### Study Area

Two expeditions were performed along the lower Amazon River during high water (May 2014) and low water (November 2015). The upstream boundary was Óbidos, the historic downstream gaging station, which is roughly 800 km from the river mouth (**Figure 1**). The downstream boundary was the northern and southern channels near the city of Macapá (NMCP and SMCP, respectively), which is ∼650 km downstream of Óbidos. Sampling was also performed in the Amazon River mainstem midway to the mouth near the city of Almeirim and near the mouth of the two primary clearwater tributaries downstream of Óbidos, the Xingu and Tapajós rivers. The Tapajós and Xingu rivers add ∼10% to the total discharge of the Amazon River

FIGURE 1 | Samples were collected in the Amazon River mainstem at Óbidos, Almeirim, and the north and south channels near the city of Macapá. Samples were also collected in the Lagoa Grande de Curuai (high water only, inaccessible at low water), the Tapajós River, and the Xingu River. Adapted from Ward et al. (2015).

(Sioli, 1985; Molleri et al., 2010). The lower Amazon River is characterized by a mixture of large channels, clear water tributaries, floodplain lakes and flooded forests, representing ∼13% of the total Amazon River drainage basin (Ward et al., 2015). River discharge throughout the study period was reported by Ward et al. (2016) and is also shown in **Table 1**.

#### Sample Collection

Water was collected at three equidistant cross-channel sites at each of the five stations on the mainstem (**Figure 1**) using a Shurflo submersible pump with a 297 µm mesh screen at 50% river depth and from the surface (Ward et al., 2016). Sampling was only performed at one central station in the Tapajós and Xingu rivers at surface and 50% depth. Only one cross channel station was sampled in these tributaries due to their large channel width and relatively low depth, which results in little variability in chemical parameters across the channel. A graduated cylinder was continuously overflowed without bubbles using the submersible pump. Dissolved O<sup>2</sup> concentrations were measured using a YSI ProOdo optical probe submerged in the overflowing cylinder and temperature was similarly measured using a Thermo Orion 4-star meter. Continuous measurements of dissolved O<sup>2</sup> concentrations and temperature were also made while the ship was underway using a YSI Exo 2 sonde placed in an overflowing bucket with water pumped from the ship's intake.

Samples were collected for analysis of the stable isotopic composition of water (δ <sup>18</sup>O-H2O) in 15–50 mL centrifuge tubes and do not require preservation before analysis. Samples were collected for analysis of the stable isotopic composition of dissolved O<sup>2</sup> in 12 mL exetainers (Barth et al., 2004). Four replicate samples were collected at 50% depth and the surface at all cross channel sites. To assess diel variability in O<sup>2</sup> saturation and δ <sup>18</sup>O-O2, surface water samples were collected on a 3 hr interval for 15 h (evening to morning) at the center of the channel near Óbidos during late high water/early falling water (July, 2015).

Prior to sample collection, exetainers were rinsed with MilliQ water and combusted at 500◦C. Fifty microliters of saturated mercuric chloride was added to each exetainer as a preservative. The exetainers were then dried and flushed with helium gas and capped in a pure helium glove box to remove any trace of atmospheric O2. Samples were collected by immersing the exetainers in the overflowing graduated cylinder. The vials were opened, filled completely, and then capped without removing them from the cylinder to prevent bubbles. To ensure the halt of all biological activity, an additional 50 µL of saturated mercuric

TABLE 1 | River discharge, O2 saturation, the stable isotopic composition of dissolved O2 , and the stable isotopic composition of H2O in the lower Amazon River.


chloride solution was injected through the septa. Septa were then coated with a thin layer of vacuum grease. Samples were analyzed within 2 months at the University of Washington Oceanography Stable Isotope Lab.

#### Analysis of δ <sup>18</sup>O-O<sup>2</sup> and δ <sup>18</sup>O-H2O

For δ <sup>18</sup>O-O<sup>2</sup> samples pure helium was pumped into the exetainer until half of the water was displaced 1 day prior to analysis. The vials were injected with 50 µL of 50% H3PO<sup>4</sup> to convert all carbonate species to CO2. Isotopic ratios of headspace gases were determined by simultaneously measuring masses 32, 34, and 40 (i.e., <sup>16</sup>O-16O, <sup>18</sup>O-16O, and <sup>40</sup>Ar) on a Finnegan Delta XL mass spectrometer (Thermo Electron). Dissolved oxygen concentrations were calculated based on the O2:Ar ratio (masses 32 and 40) and the estimated dissolved Ar concentration as a function of water temperature (Weiss, 1970). These values were compared to measurements taken with the YSI ProOdo and YSI Exo 2 sonde described above and were roughly equivalent (i.e., within ∼5%). The stable isotopic composition of water (18O-H2O) was measured on 15 mL whole water samples using a Picarro L2130-I Analyzer.

#### Mass Balance Calculations

A dissolved oxygen mass balance was calculated for the Amazon River mainstem reach between Óbidos and midway to the mouth near the city of Almeirim (**Figure 2**). Two mass balance models were developed with the same set of inputs/outputs—one model uses only bulk parameters and the other includes the stable isotopic composition of O<sup>2</sup> and associated fractionation factors.

Inputs of O<sup>2</sup> included the measured/calculated flux of O<sup>2</sup> through Óbidos, measured/calculated O<sup>2</sup> fluxes from the Tapajós River (**Figure 1**), calculated O<sup>2</sup> gas exchange with the atmosphere (concentrations in the river were consistently below atmospheric saturation), estimated floodplain inputs, and primary production, which was the unknown parameter that was solved for. O<sup>2</sup> outputs included microbial respiration rates measured by Ward et al. (unpublished) and the measured/calculated flux of O<sup>2</sup> through Almeirim. In brief, respiration rates were measured in the dark in rotating

incubation chambers interfaced to an YSI Exo 2 sonde equipped with an optical dissolved O<sup>2</sup> probe. Respiration rates were calculated based on the linear drawdown of O<sup>2</sup> during a roughly 24 hr period in duplicate 2.85 L chambers. Respiration rates include uncertainty associated with differences between replicates and averaged sites (±1 SD) and an additional 3% uncertainty due to the lack of inclusion of benthic respiration rates (more details on this in both the results and discussion sections).

This particular section of the river was chosen because the river maintains one well-constrained channel, for the most part, and river flow is not reversed by tides in this reach (although discharge is dampened semi-diurnally at Almeirim). This mass balance was not performed from Almeirim to Macapá because the river becomes extensively channelized and semi-diurnal tidal variation greatly complicates estimations of floodplain connectivity. The uncertainty associated with each mass balance parameter was propagated to our final calculation of primary production.

#### Calculation of O<sup>2</sup> Gas Exchange

The molecular diffusivity of O<sup>2</sup> across the air-water interface is described by the gas transfer velocity parameter (k). Gas transfer velocities can be calculated for other gases (and temperatures) based on the ratio of Schmidt numbers (Sc) (Jähne et al., 1987):

$$k\_{\mathbb{S}\mathfrak{c}1} \Big/ k\_{\mathbb{S}\mathfrak{c}2} = \sqrt{\mathfrak{C}^{\mathbf{c}\_1} \Big/\_{\mathbb{S}\mathfrak{c}\_2}}$$

where kSc<sup>1</sup> and kSc<sup>2</sup> are the gas transfer velocities of a reference gas with a unknown and known value, respectively, and Sc<sup>1</sup> and Sc<sup>2</sup> are the respective Schmidt numbers for each gas. To determine the gas transfer velocity of O2, we converted calculated kCH4 values that were determined simultaneously (Sawakuchi, unpublished). We calculated Sc for O<sup>2</sup> and CH<sup>4</sup> at 28◦C (measured mean temperature of the water) following Wanninkhof (1992) and used these values and the calculated kCH4 to determine kO2.

The flux of a gas across the air-water interface was calculated as follows:

$$F = k(\mathcal{C}\_{\le} - \mathcal{C}\_{eq})$$

where F is flux (mol m−<sup>2</sup> d −1 ), k is the gas transfer velocity (m d−<sup>1</sup> ), C<sup>w</sup> is the concentration of gas measured in the water (mol m−<sup>3</sup> ) and Ceq is the concentration of a water sample at equilibrium with the atmosphere at in situ temperature. Water temperatures ranged from 27.5 to 28.5◦C over the course of our study. As such, we used a constant value of 28◦C for all necessary equations. The equilibrium concentration of oxygen at 28◦C was calculated using Henry's Law. Fluxes were calculated using varied k values calculated as described above, the calculated equilibration concentration and the average concentration of dissolved O<sup>2</sup> for this reach of the river during high and low water periods. Uncertainty the gas transfer was calculated as the combination of analytical error and the standard deviation between observations at Óbidos and Almeirim (± 1 SD).

#### River Discharge

The water mass balance for this reach can be described by the following equation:

$$Q\_{OBD} + Q\_{TAP} + Q\_{blood} = Q\_{ALM}$$

where QOBD is the discharge through Óbidos (m<sup>3</sup> s −1 ), QTAP is discharge from the Tapajós River tributary, Qflood is discharge from floodplains, and QALM is discharge through Almeirim. Discharge was measured across the Amazon River main channel at Óbidos and Almeirim and across the mouth of the Tapajós River using a Sontek River Surveyor M9 Portable nine-beam 3.0 MHz/1.0 MHz/0.5 MHz acoustic Doppler Current Profiler (Ward et al., 2015). Discharge from floodplains was estimated by subtracting the measured discharge at Óbidos and the Tapajós River from the measured discharge at Almeirim. It should be noted that this does not necessarily represent the total amount of water that passed through floodplain complexes, but, rather, additional discharge added along this reach of the river from water stored in floodplains.

#### O<sup>2</sup> Mass Balance

In order to calculate the non-measured/estimated oxygen inputs (i.e., primary production), two mass balance equations were used based on bulk O<sup>2</sup> measurements (Equation 4) and both bulk and stable isotopic O<sup>2</sup> measurements (Equation 5):

$$O\_{2ALM} = \frac{\begin{bmatrix} Q\_{OBD} \ O\_{2,OBD} + Q\_{TAP} O\_{2,TAP} + \\ Q\_{flood} O\_{2,flood} + AF - VR + VP \end{bmatrix}}{Q\_{ALM}}$$

where A is the area of the river, F is the flux of O<sup>2</sup> from the atmosphere to the river, V is the volume of the river, R is the average measured respiration rate, and P is the average rate of primary production, which was solved for. O2,ALM, O2,OBD, O2,TAP, and, O2,flood, are the measured concentrations of dissolved O<sup>2</sup> at Almeirim, Óbidos, the Tapajós River, and the Lagoa Grande de Curuai floodplain lake, respectively. Uncertainty for the flux terms to/from Óbidos, the Tapajós River, and Almeirim are based on the analytical error of the O<sup>2</sup> probe, variability in O<sup>2</sup> concentrations with depth and across the channel, and discharge measurements (±1 SD).

Using equation 4 alone can give an estimate of inputs of O<sup>2</sup> via primary production between Óbidos and Almeirim, but this can be improved by adding a mass balance for the stable isotopic composition of O<sup>2</sup> as follows:

$$\begin{aligned} ^{18:16}O\_{2,ALM} &= [Q\_{OBD}O\_{2,OBD}]^{18:16}O\_{2,OBD} \\ &+ Q\_{TAPO}O\_{2,TAP} ^{18:16}O\_{2,TAP} \\ &+ Q\_{flood}O\_{2,flood} ^{18:16}O\_{2,flood} \\ &+ AF^{18:16}O\_{\xi\alpha}\alpha\_{\xi} - V \times R^{18:16}O\_2 \times \alpha\_R \\ &+ VP^{18:16}O\_{\eta\alpha}\alpha\_P \big] / \ [Q\_{OBD}O\_{2,OBD} \\ &+ Q\_{TAPO}O\_{2,TAP} + Q\_{flood}O\_{2,flood} + AF - VR + VP \big] \end{aligned}$$

where <sup>18</sup>:16O<sup>2</sup> is the ratio of <sup>18</sup>O to <sup>16</sup>O calculated from <sup>δ</sup> <sup>18</sup>O-O<sup>2</sup> values for each respective term relative to the Vienna Standard Mean Ocean Water (VSMOW) standard, <sup>18</sup>:16O<sup>g</sup> and <sup>18</sup>:16O<sup>w</sup> are the stable isotopic compositions of atmospheric oxygen and river water, respectively. αR, αP, and α<sup>g</sup> are the fractionation factors of respiration, photosynthesis, and gas exchange respectively. A value of 0.982 was used for αR, a value of 0.997 was used for α<sup>g</sup> , and a value of 1.000 was used for α<sup>P</sup> considering O<sup>2</sup> produced via primary production reflects the stable isotopic composition of the water (Quay et al., 1995). The above equations were solved for P, or the rate of primary production within this reach. Uncertainty from each model parameter (±1 SD) was propagated through to the final result. All calculations were performed based on the average temperature of 28◦C.

#### RESULTS

## Dissolved O<sup>2</sup> Saturation

Dissolved oxygen was under-saturated relative to the atmosphere at all sampling locations at both high and low water with an annual average value of 79.3 ± 18.1% across the study boundaries. The highest and least variable O<sup>2</sup> saturation levels were found in the Tapajós and Xingu clearwater rivers, with seasonal average values of 86.7 ± 1.0 and 96.0 ± 1.8% at high and low water, respectively (**Table 1**). The lowest O<sup>2</sup> saturation levels were observed during high water in the Amazon River mainstem, specifically at Óbidos. Across all mainstem sites including the mouth, average O<sup>2</sup> saturation was 54.3 ± 6.0% at high water compared to 88.5 ± 6.5% at low water (**Table 1**).

O<sup>2</sup> saturation in the mainstem exhibited a downstream increase from Óbidos to the river mouth with an annual average value of 62.8 ± 23.2% at Óbidos, 71.0 ± 25.4% at Almeirim, and 76.0 ± 19.6%. The downstream increase was most evident during low water, with values increasing from 79.2 ± 1.9% at Óbidos to 93.0 ± 0.3% across the mouth. During the high water period O<sup>2</sup> saturation increased from 46.4 ± 4.8% at Óbidos to 59.0 ± 1.3% across the mouth (**Table 1**).

The partial pressure of CO<sup>2</sup> was previously measured along the same study boundaries (Sawakuchi et al., under review; Ward et al., 2016). There was a negative correlation between the molar concentration of dissolved O<sup>2</sup> and CO<sup>2</sup> across the study boundaries (**Figure 3**). Spatial variability of O<sup>2</sup> and CO<sup>2</sup> in the entire study region, including both the Amazon River mainstem and tributary sites, was higher during the high water period. For example, CO<sup>2</sup> concentrations ranged from 86.9 to 200.7 µmol L−<sup>1</sup> and 36.7 to 61.5 µmol L−<sup>1</sup> during high and low water, respectively (**Figure 3**). O<sup>2</sup> concentrations varied from 90.0 to 150.8 µmol L−<sup>1</sup> and 160.3 to 197.2 µmol L−<sup>1</sup> at high and low water, respectively, across our study boundaries. The slope of the relationship between O<sup>2</sup> and CO<sup>2</sup> concentrations was 2.2 times higher during low water.

#### Dissolved O<sup>2</sup> Stable Isotopic Composition

The oxygen stable isotopic composition of water (i.e., δ <sup>18</sup>O-H2O) is influenced by isotopic fractionation during each phase of the hydrologic cycle. Annually-averaged δ <sup>18</sup>O-H2O was −4.3 ± 1.1‰ across the entire study boundary. δ <sup>18</sup>O-H2O was the most negative in the Amazon River mainstem at high water

with an average value of −5.4 ± 2.1‰ compared to −3.6 ± 0.4‰ at low water (**Table 1**). δ <sup>18</sup>O-H2O was more positive in the clearwater tributaries (average = −3.6 ± 1.2‰) compared to the Amazon River mainstem (average = −4.5 ± 1.0‰). There was also a downstream increase in δ <sup>18</sup>O-H2O from Óbidos to the river mouth. For example, at high water an average δ <sup>18</sup>O-H2O value of −5.7 ± 0.1‰ was observed at Óbidos compared to −5.2 ± 0.1‰ across the river mouth. The difference between the mainstem and tributaries is likely due to differences in the isotopic composition of rainfall, since much of the water flowing through the mainstem is derived from far upstream.

The annual average stable isotopic composition of dissolved oxygen (i.e., δ <sup>18</sup>O-O2) was 23.8 ± 1.4‰ across the entire study boundary. The highest δ <sup>18</sup>O-O<sup>2</sup> values were observed in the Amazon River mainstem at high water with an average value of 25.0 ± 1.1‰ compared to 23.9 ± 0.7‰ at low water. The clearwater tributaries had consistently lower and less seasonally variable δ <sup>18</sup>O-O<sup>2</sup> values, varying from 22.9 ± 1.5‰ at high water to 22.2 ± 1.7‰ at low water (**Table 1**). The Xingu River had the lowest δ <sup>18</sup>O-O<sup>2</sup> values across all sites (annual average = 21.5 ± 0.6‰). δ <sup>18</sup>O-O<sup>2</sup> and oxygen saturation levels were negatively correlated during both sampling seasons (**Figure 4**). Higher δ <sup>18</sup>O-O<sup>2</sup> values were found at sites with lower oxygen saturation levels and vice versa. The regression slope between δ <sup>18</sup>O-O<sup>2</sup> values and oxygen saturation, while following the same trend, was steeper for samples collected at low water compared to water.

It is typically assumed that d18O-O<sup>2</sup> should remain at steady state in the Amazon River mainstem considering the low levels of light penetration and limited primary production (Quay et al., 1995; Ellis et al., 2012). To test this assumption we measured O<sup>2</sup> saturation and d18O-O<sup>2</sup> throughout a day/night cycle in the center of the channel at Óbidos. While the overall

oxygen saturation levels did not vary significantly relative to the signal noise over a 15 hr period, our results indicate that d <sup>18</sup>O-O<sup>2</sup> does follow a diel cycle even in the Amazon River mainstem (**Figure 5**). The minimum d18O-O<sup>2</sup> was observed at 19:00, roughly 1 h after sunset (24.7 ± 0.6‰) and the maximum value was observed at 06:00 just prior to sunset (26.4 ± 0.3‰), followed by a decrease to 25.3 ± 0.7‰ at 08:00 (**Figure 5**). The difference in d18O-O<sup>2</sup> values observed between midnight and 06:00 and the rest of the diel cycle were statistically significant based on an unpaired t-test within a 95% confidence interval (p = 0.02). It should be noted that a complete 24 h cycle was not able to be captured due to logistical constraints of the field campaign.

FIGURE 5 | Diel O2 saturation (red) and δ <sup>18</sup>O-O<sup>2</sup> (blue) measurements made in the center of the channel at Obidos. The shaded portion of the chart represents hours between sunset and sunrise.

#### Oxygen Mass Balance Model

In the past, δ <sup>18</sup>O-O<sup>2</sup> values have been used to calculate the ratio of photosynthesis to respiration (P:R) under a steady state assumption for the Amazon River (Quay et al., 1995; Ellis et al., 2012). However, considering diel variability of d18O-O<sup>2</sup> was observed, a steady state assumption is not accurate. Thus, in order to determine the rates of photosynthetic oxygen input into the mainstem of the lower Amazon River, we used a mass balance model that takes into account the various sources and sinks of O<sup>2</sup> within the system (**Figure 2**). Because the stable isotopic composition of dissolved O<sup>2</sup> is also known, we were able to calculate two separate O<sup>2</sup> budgets to model the system based on the following: (1) bulk O<sup>2</sup> concentrations only (i.e., Equation 4) and (2) the concentration and stable isotopic composition of O<sup>2</sup> (i.e., Equation 5). Including d18O-O<sup>2</sup> observations and literature-derived fractionation factors for each process expressed in Equation 5 resulted in a slight increase in each model parameter compared to the bulk O<sup>2</sup> mass balance, however, the resulting calculated P:R ratios were essentially equivalent (**Table 2**). For this reason, we report the average value for each parameter from these two different models below.

Although O<sup>2</sup> concentrations were considerably lower in the Amazon River mainstem during high water (**Table 1**), the flux of O<sup>2</sup> through Óbidos and Almeirim were greater due to higher river discharge rates. The average flux of O<sup>2</sup> through Óbidos at high water, determined using both models, was 901 ± 87 kg O s−<sup>1</sup> compared to 739 ± 80 kg O s−<sup>1</sup> at low water (**Table 2**). The flux of O<sup>2</sup> through Almeirim was consistently greater than at Óbidos, with values of 1228 ± 112 kg O s−<sup>1</sup> at high water and 853 ± 83 kg O s−<sup>1</sup> . This difference was greatest during high water with a 36% increase in O<sup>2</sup> flux from Óbidos to Almeirim compared to a 15% increase during the low water period.

Aside from primary production, gas exchange was the dominant source of O<sup>2</sup> to this reach of the river, resulting in an

TABLE 2 | Dissolved O2 fluxes through Óbidos, Almeirim, floodplains, and the Tapajós River and from respiration, gas exchange, and primary production using mass balance models derived from (1) bulk O2 concentrations and (2) O2 stable isotopic compositions.


Positive values indicate inputs and negative values indicate outputs.

inward flux of 722 ± 169 kg O s−<sup>1</sup> at high water and 565 ± 130 kg O s−<sup>1</sup> at low water. The Tapajós River provided an additional 81 ± 10 kg O s−<sup>1</sup> to the Amazon River mainstem during high water and 25 ± 3 kg O s−<sup>1</sup> during low water, which represents about 20–25% of the difference in fluxes between Óbidos and Almeirim during both seasons. We estimate that the flux of O<sup>2</sup> from adjacent floodplains to the mainstem was roughly 247 ± 75 kg O s−<sup>1</sup> during high water (**Table 2**). At low water, we considered floodplain discharge to be zero considering that river discharge at Almeirim was lower than the sum discharge from Óbidos and the Tapajós River, indicative of water entering and being stored in the floodplains. Further, the Lagoa Grande de Curuai was inaccessible during this sampling period, illustrating the limited floodplain connectivity to the mainstem during this time.

Gagne-Maynard et al. Amazon River Oxygen Balance

Aside from the flux through Almeirim, respiration was the primary O<sup>2</sup> output. The average water column respiration rate measured across the Amazon River mainstem sampling sites was 0.95 ± 0.10 g O m<sup>3</sup> d −1 at high water and 1.52 ± 0.35 g O m<sup>3</sup> d −1 at low water (Ward et al., unpublished). Benthic respiration rates were not measured, however, literature data is available from sites further upstream. Average benthic respiration rates of 1.46 g O m<sup>2</sup> d <sup>−</sup><sup>1</sup> have been reported for the central Amazon River (Devol et al., 1987). Considering that the average river depth is ∼40 m between Óbidos and Almeirim (**Table 1**), benthic respiration rates represent only 3% of depth-integrated water column respiration rates. Since benthic respiration was not measured in our study region, we have included this 3% error to the uncertainty in our total respiration estimate rather than adding a separate model parameter similar to calculations made by Devol et al. (1987). Water column respiration rates were multiplied by the volume of water between Óbidos and Almeirim (1.09 × 10<sup>11</sup> m<sup>3</sup> at high water and 8.17 × 10<sup>10</sup> m<sup>3</sup> at low water), resulting in a consumption of 1228 ± 112 kg O s−<sup>1</sup> at high water and 1447 ± 368 kg O s−<sup>1</sup> at low water (**Table 2**).

With all O<sup>2</sup> inputs and outputs other than primary production either measured or estimated we then solved for primary production as the difference between the mass balance outputs and inputs. Primary production was estimated to add 492 ± 291 kg O s−<sup>1</sup> to the Amazon River mainstem between Óbidos and Almeirim at high water and 971 ± 407 kg O s−<sup>1</sup> . This corresponds to a P:R ratio of 0.41 ± 0.24 at high water, 0.67 ± 0.33 at low water, and 0.54 ± 0.41 annually-averaged. When multiplied by the measured respiration rates, this corresponds to a primary production rate of 0.39 ± 0.24 g O m<sup>3</sup> d −1 at high water, 1.02 ± 0.55 g O m<sup>3</sup> d −1 at low water, and 0.70 ± 0.58 g O m<sup>3</sup> d −1 annually-averaged.

## DISCUSSION

#### Dissolved Oxygen Balance

The stable isotopic composition of dissolved O<sup>2</sup> (i.e., δ <sup>18</sup>O-O2) is primarily regulated by three processes: air-water gas exchange, respiration, and photosynthesis. In a system purely affected by gas-exchange, the δ <sup>18</sup>O-O<sup>2</sup> would be 24.2‰ (relative to VSMOW) due to the dissolution of atmospheric O<sup>2</sup> (23.5‰) and the equilibrium fractionation of 0.7‰ that occurs during gas dissolution (Kroopnick, 1975; Benson and Krause, 1984). Photosynthesis produces O<sup>2</sup> with a δ <sup>18</sup>O of the surrounding water because there is little fractionation observed during photosynthesis (Guy et al., 1993). Respiration selectively fractionates oxygen by consuming O<sup>2</sup> that is more depleted than the average, which enriches the remaining pool of O2. Thus, photosynthesis is expected to deplete the δ <sup>18</sup>O-O<sup>2</sup> of a system while respiration will enrich the δ <sup>18</sup>O-O2.

Bulk O<sup>2</sup> levels were consistently undersaturated relative to the atmosphere across the entire study boundary. Likewise, δ <sup>18</sup>O-O<sup>2</sup> values were consistently more enriched in the Amazon River mainstem relative to values expected due to gas transfer with the atmosphere alone (i.e., 24.2‰; **Table 1**), indicating a predominance of respiration relative to photosynthesis. These factors both reflect the net heterotrophic state previously observed in both the Amazon River (Mayorga et al., 2005) and river systems worldwide (Cole et al., 2007), which results in a net flux of CO<sup>2</sup> to the atmosphere (Richey et al., 2002; Alin et al., 2011; Raymond et al., 2013).

However, these bulk and stable isotopic O<sup>2</sup> observations do not preclude the input of photosynthetically-produced oxygen into the mainstem. In fact, our observations of diel variability in δ <sup>18</sup>O-O<sup>2</sup> values at Óbidos clearly indicate a shift in the balance between primary production and respiration in the mainstem throughout a day/night cycle. Our observations were consistent with daytime production depleting δ <sup>18</sup>O-O<sup>2</sup> and nighttime respiration enriching the signal, implying that O<sup>2</sup> does not remain at a steady-state as previously assumed (Quay et al., 1995; Wassenaar, 2012).

The more depleted δ <sup>18</sup>O-O<sup>2</sup> values observed in the Amazon River mainstem during low water indicate a relatively higher amount of primary production relative to respiration during low water. This makes sense considering that suspended sediment concentrations are generally higher during high water in the Amazon River mainstem, limiting primary production (Hedges and Clark, 1986; Moreira-Turcq et al., 2003, 2013).

Respiration rates also vary seasonally, influencing the ratio of production to respiration. The highest respiration rates observed here and in the literature occur during the low water period (Benner et al., 1995; Ellis et al., 2012; Ward et al., 2013), which impliesthat photosynthesis rates would need to be proportionally higher to achieve a shift in P:R ratios during the low water period. In fact, our mass balance model predicted this difference. We estimate that primary production was 2.0 times higher in the Óbidos to Almeirim reach during low water based on our mass balance models (**Table 2**). Average measured respiration rates were 1.6 times greater in the mainstem during low water, but when integrated across the entire volume of water in the Óbidos to Almeirim reach, the total respiration flux was only 1.2 times greater at low water compared to high water. These differences result in the observed shift in the ratio of P:R from 0.41 ± 0.24 at high water to 0.67 ± 0.33 at low water.

O<sup>2</sup> levels were closer to atmospheric saturation in the Tapajós and Xingu rivers (**Table 1**), and δ <sup>18</sup>O-O<sup>2</sup> values were below 24.2‰ (**Table 2**), indicating a predominance of primary production relative to respiration in the less turbid clearwater tributaries. Sediment levels are significantly lower in these tributaries compared to the Amazon River mainstem, and high rates of primary production are reflected in elevated chlorophyll a concentrations relative to the Amazon River (Moreira-Turcq et al., 2013; Ward et al., 2015, 2016). Interestingly, respiration rates have been observed to be higher in clearwater tributaries than in the Amazon River mainstem throughout the Amazon basin (Benner et al., 1995; Ellis et al., 2012), yet these tributaries are overall less net heterotrophic. The elevated respiration rates observed in clearwater tributaries have been attributed to both the breakdown of algal organic matter (Benner et al., 1995; Ellis et al., 2012) and the enhanced breakdown of terrestrially-derived organic matter due to priming effects, i.e., the enhanced breakdown of a recalcitrant substrate due to the presence of a labile substrate (Bianchi et al., 2015; Ward et al., 2016).

Previous studies have used δ <sup>18</sup>O-O<sup>2</sup> values to estimate gas exchange(Jamieson et al., 2012) and estimate net metabolism (e.g., the P:R ratio) in steady-state systems(Quay et al., 1995; Bocaniov et al., 2012) as well as non-steady-state systems (Tobias et al., 2007; Holtgrieve et al., 2010; Hotchkiss and Hall, 2014). For the Amazon River mainstem it has been estimated that the ratio of P:R is roughly 0.25 based on δ <sup>18</sup>O-O<sup>2</sup> values and a steady state assumption (Quay et al., 1995), which is 2.2 times lower than our annual average P:R estimate. Primary production rates have also been estimated based on <sup>14</sup>C-NaHCO<sup>3</sup> uptake experiments; the average rate of primary production was estimated to be 0.10 ± 0.06 g O m<sup>3</sup> d −1 in the Amazon River mainstem and 1.45 ± 0.75 g O m<sup>3</sup> d −1 in the Tapajós and Xingu rivers (Wissmar et al., 1981). This is roughly 6.7 times lower than our mass balance estimates and 2.9 lower than estimates based on multiplying our measured rates of respiration by P:R ratios determined by Quay et al. (1995). We argue that this difference is due to the fact that bottle experiments cannot adequately capture the complex ecosystem dynamics occurring along the entire river reach, whereas a mass balance integrates these signals. Our mass balance results are in agreement with observations of a significant abundance of phytoplankton biomarkers in the mainstem of the Amazon (Mortillaro et al., 2011). Likewise, primary production has recently been found to be important in the main channel of the Congo River (Descy et al., 2017).

#### Limitations and Future Considerations

One factor that remains difficult to constrain is whether this input of isotopically depleted O<sup>2</sup> is conclusively from in situ production in the mainstem itself or is the legacy of advection from lateral floodplain lakes. These floodplain lakes have high rates of productivity and there is evidence of their legacy biomass in the Amazon mainstem (de Moraes Novo et al., 2006; Abril et al., 2014). This input of floodplain-derived oxygen is reflected in the ratio of CO<sup>2</sup> to O<sup>2</sup> measured at high and low water (**Figure 3**). At high water, the ratio is higher, likely due to the input of root respiration-derived CO<sup>2</sup> from the floodplains. At high water, these flooded macrophytes draw down oxygen from the atmosphere to their roots, thereby producing CO<sup>2</sup> without the concomitant drawdown of O<sup>2</sup> from the water column. Thus, it is possible that the isotopically depleted O<sup>2</sup> signals observed at high water is actually the legacy of floodplain production and not mainstem production. However, at low water, there is little evidence of macrophyte root respiration as the river loses connectivity with the floodplains. Thus, it is likely the production signal observed at low water is primarily from in situ production within the mainstem rather than floodplains.

Another factor that we did not consider is the importance of benthic respiration from the riverbed. However, past estimates suggest that benthic respiration in the OM-poor sandy sediments only accounts for ∼3% of our measured water column respiration rates (Devol et al., 1987). Thus we did not include a benthic respiration term in our mass balance, but rather added 3% error to our total respiration term. In some estuarine settings benthic respiration has been shown to be the dominant factor controlling dissolved oxygen concentrations and stable isotopic compositions (Lehmann et al., 2009), however this does not appear to be the case in the fast-flowing, deep river setting where reactive organic matter is highly abundant in the water column but relatively depleted in sediments (Richey et al., 1990; Hedges et al., 2000; Moreira-Turcq et al., 2013).

We did not attempt this same mass balance calculation for the river reach between Almeirim and Macapá due to the complex nature of this region. The river evolves into a series of sub-channels in the lower reaches limiting the applicability of a simplistic model, whereas the reach between Óbidos and Almeirim represents a single, well-constrained channel for the most part. Further, semi-diurnal flow reversals due to tides make it even more difficult to quantify connectivity and import/export from floodplain networks near the river mouth. Future efforts will be applied to unraveling the complex tidally-influenced reach of the river. It should also be noted the floodplain discharge used in this mass balance does not represent the total amount of water that passed through floodplain systems, but, rather, excess water entering the river from floodplains between Óbidos and Almeirim. It is difficult to accurately constrain the amount of water that passes through the Lagoa Grande de Curuai, for example, but this will be the focus of future hydrodynamic modeling efforts to further constrain our mass balance estimates.

## CONCLUSIONS

Early investigations of biogeochemical cycling in the Amazon River found minimal rates of primary production in the turbid mainstem relative to the productive clearwater tributaries (Fisher, 1979; Wissmar et al., 1981). Here, we have shown that there is evidence for primary production within the Amazon River mainstem based on an O<sup>2</sup> mass balance, which exceeds past estimates based on a steady-state stable isotopic model by 2.0 times (Quay et al., 1995) and is 6.7 times higher than estimates from radiocarbon uptake experiments (Wissmar et al., 1981). The Amazon River is a major source of CO<sup>2</sup> to the atmosphere, but understanding the interplay between photosynthesis and respiration is critical for understanding the fundamental mechanisms driving these fluxes and the overall productivity of the ecosystem. Likewise, in situ primary production has been shown to be an important factor in elevating both bulk respiration rates (Ellis et al., 2012) and the breakdown of terrestrially-derived organic matter (Ward et al., 2016). Although this result does not undermine the idea of the Amazon River as a net heterotrophic system, it does indicate that more work remains to understand the dynamic interplay between production and respiration in large turbid river systems.

#### AUTHOR CONTRIBUTIONS

WG performed the collection and analysis of δ18O-O<sup>2</sup> samples and developed the O<sup>2</sup> mass balance. VN, NW, DD, and JD. performed measurements of dissolved O<sup>2</sup> concentrations. VN and HS. performed estimations of gas transfer velocities used in the mass balance. NW and WG. performed measurements of respiration used in the O<sup>2</sup> mass balance. JR, AD, and DB performed measurements of river discharge. The field sampling scheme was developed and led by JR, AK, RK, NW, and HS. Field and sampling logistics were undertaken by all authors. All authors contributed to the preparation of the manuscript and approved its final submission.

#### REFERENCES


#### FUNDING

This study was supported by FAPESP Grant # 08/58089-9 and NSF DEB Grant # 1256724. WGM was funded by an NSF IGERT grant DGE-1258485.

#### ACKNOWLEDGMENTS

We thank the crew of the B/M Mirage for contributions made during the river cruises, Dr. Gordon Holtgrieve for his advice on oxygen isotopes and the Quay Lab at UW for assistance in isotopic analysis.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Gagne-Maynard, Ward, Keil, Sawakuchi, Da Cunha, Neu, Brito, Da Silva Less, Diniz, De Matos Valerio, Kampel, Krusche and Richey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Fate of Carbon in Sediments of the Xingu and Tapajós Clearwater Rivers, Eastern Amazon

Dailson J. Bertassoli Jr. <sup>1</sup> \*, André O. Sawakuchi <sup>1</sup> , Henrique O. Sawakuchi <sup>2</sup> , Fabiano N. Pupim<sup>1</sup> , Gelvam A. Hartmann<sup>3</sup> , Michael M. McGlue<sup>4</sup> , Cristiano M. Chiessi <sup>5</sup> , Matthias Zabel <sup>6</sup> , Enno Schefuß<sup>6</sup> , Tatiana S. Pereira<sup>7</sup> , Rudney A. Santos <sup>1</sup> , Samantha B. Faustino<sup>8</sup> , Paulo E. Oliveira<sup>1</sup> and Denise C. Bicudo<sup>8</sup>

1 Institute of Geosciences, University of São Paulo, São Paulo, Brazil, <sup>2</sup> Environmental Analysis and Geoprocessing Laboratory, Center for Nuclear Energy in Agriculture, University of São Paulo, Piracicaba, Brazil, <sup>3</sup> Instituto de Geociências, Universidade Estadual de Campinas, Campinas, Brazil, <sup>4</sup> Department of Earth and Environmental Sciences, University of Kentucky, Lexington, KY, USA, <sup>5</sup> School of Arts, Sciences and Humanities, University of São Paulo, São Paulo, Brazil, <sup>6</sup> MARUM-Center for Marine Environmental Sciences, University of Bremen, Bremen, Germany, <sup>7</sup> Federal University of Pará, Campus de Altamira, Altamira, Brazil, <sup>8</sup> Department of Ecology, Instituto de Botânica, São Paulo, Brazil

#### Edited by:

Carol Robinson, University of East Anglia, UK

#### Reviewed by:

Michael Gonsior, University of Maryland Center For Environmental Sciences, USA Armstrong-Altrin John S., National Autonomous University of Mexico, Mexico

\*Correspondence:

Dailson J. Bertassoli Jr. dailson.bertassoli@gmail.com

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 07 November 2016 Accepted: 07 February 2017 Published: 22 February 2017

#### Citation:

Bertassoli DJ Jr., Sawakuchi AO, Sawakuchi HO, Pupim FN, Hartmann GA, McGlue MM, Chiessi CM, Zabel M, Schefuß E, Pereira TS, Santos RA, Faustino SB and Oliveira PE and Bicudo DC (2017) The Fate of Carbon in Sediments of the Xingu and Tapajós Clearwater Rivers, Eastern Amazon. Front. Mar. Sci. 4:44. doi: 10.3389/fmars.2017.00044 The Xingu and Tapajós rivers in the eastern Amazon are the largest clearwater systems of the Amazon basin. Both rivers have "fluvial rias" (i.e., lake-like channels) in their downstream reaches as they are naturally impounded by the Amazon mainstem. Fluvial rias are widespread in the Amazon landscape and most of the sedimentary load from the major clearwater and blackwater rivers is deposited in these channels. So far, little is known about the role of Amazon rias as a trap and reactor for organic sediments. In this study, we used organic and inorganic geochemistry, magnetic susceptibility, diatom, and pollen analyses in sediments (suspended, riverbed, and downcore) of the Xingu and Tapajós rias to investigate the effects of hydrologic variations on the carbon budget in these clearwater rivers over the Holocene. Ages of sediment deposition (∼100 to 5,500 years) were constrained by optically stimulated luminescence and radiocarbon. Major elements geochemistry and concentration of total organic carbon (TOC) indicate that seasonal hydrologic variations exert a strong influence on riverine productivity and on the input and preservation of organic matter in sediments. Stable carbon isotope data (δ <sup>13</sup>C from −31.04 to −27.49‰) and pollen analysis indicate that most of the carbon buried in rias is derived from forests. In the Xingu River, diatom analysis in bottom sediments revealed 65 infrageneric taxa that are mostly well-adapted to slack oligotrophic and acidic waters. TOC values in sediment cores are similar to values measured in riverbed sediments and indicate suitable conditions for organic matter preservation in sediments of the Xingu and Tapajós rias at least since the mid-Holocene, with carbon burial rates varying from about 84 g m−<sup>2</sup> yr−<sup>1</sup> to 169 g m−<sup>2</sup> yr−<sup>1</sup> . However, redox-sensitive elements in sediment core indicate alternation between anoxic/dysoxic and oxic conditions in the water-sediment interface that may be linked to abrupt changes in precipitation. The variation between anoxic/dysoxic and oxic conditions in the water-sediment interface controls organic matter mineralization and methanogenesis. Thus, such changes promoted by hydrological variations significantly affect the capacity of Amazon rias to act either as sources or sinks of carbon.

Keywords: Amazon, Carbon Cycle, Xingu River, Tapajós River, greenhouse gases

## INTRODUCTION

Recent research has undermined the view of rivers as passive conveyors of organic matter, and demonstrates their role in the transport, modulation, and deposition of carbon (Cole et al., 2007; Tranvik et al., 2009; Aufdenkampe et al., 2011; Ward et al., 2015). Inland waters emit at least 0.8 Pg of carbon per year (Cole et al., 2007; Tranvik et al., 2009). Approximately one fourth of it comes from rivers (Cole et al., 2007). The Amazonian rivers represent about 15% of the terrestrial gross primary production and 25% of the global rivers emissions of CO<sup>2</sup> (Field, 1998; Richey et al., 2002). Most studies that have considered organic matter remineralization and carbon storage in the region focused on the Amazon mainstem and its floodplains (Richey et al., 2002; Bouchez et al., 2010, 2012; Abril et al., 2014; Ward et al., 2015). However, the lack of geochemical data to constrain the role of major clearwater tributaries of the Amazon River as a source or sink of organic sediments limited the understanding of the carbon cycling dynamics across the terrestrial and aquatic environments of the Amazon basin.

The Xingu and Tapajós rivers represent the largest Amazon clearwater rivers, with mean annual discharges of 9,700 and 13,500 m<sup>3</sup> s −1 (Latrubesse et al., 2005), respectively. They drain crystalline and sedimentary rocks of the Central Brazil shield and are characterized by relatively low concentrations of suspended sediments, slightly alkaline waters (Sioli, 1984), and high fluxes of methane to the atmosphere (Sawakuchi et al., 2014). Both rivers are naturally impounded by the Amazon mainstem, which give rise to broad downstream lake-like channels known as "fluvial rias" (Gourou, 1949; Archer, 2005). The Xingu and Tapajós rivers have strong seasonal variations, with mean monthly water discharges varying from approximately 1,000 to 20,000 m−<sup>3</sup> s −1 in the Xingu River and 4,000 to 30,000 m−<sup>3</sup> s −1 in the Tapajós River (ANA, 2016). The peak discharges of these clearwater rivers occurs from February to May. The Amazon River, however, has a mean monthly water discharge that varies from approximately 105,000 m−<sup>3</sup> s −1 to 235,000 m−<sup>3</sup> s −1 , with peak discharge from April to July (ANA, 2016). The lag between the peak discharge of the Amazon River and its clearwater tributaries is responsible for strong backwater effects in the Xingu and Tapajós rivers (Meade et al., 1991). The sudden drop in water flow velocity and channel widening related to the hydraulic blockage and backwater effect by the Amazon mainstem prevent most of the Xingu and Tapajós suspended and bedload sediments from entering the Amazon River. These conditions turn rias into massive sinks for organic and inorganic sediments and natural reactors for carbon derived from the headwaters. Fluvial rias in the downstream section of the Xingu and Tapajós rivers reach up to 15 km width and around 150 km length. Hundreds of rias and paleo-rias occupy the Amazonian landscape and connect all clearwater and blackwater tributaries to the Amazon River, playing an important role for the transport and deposition of fine-grained sediments in the Amazon fluvial system. Thus, the accumulation of fine-grained sediments in rias has a strong effect on carbon processing and transport to the Amazon River and the Atlantic Ocean. To date, studies addressing Amazonian fluvial rias focus mainly on climatic and sediment deposition changes, without consensus about their genesis (Sioli, 1984; Keim et al., 1999; Vital and Stattegger, 2000; Bertani et al., 2014). The extant Amazon rias appear to have been formed after the Last Glacial Maximum (LGM, 23–19 ka) (Archer, 2005; Irion et al., 2011). The prevailing hypothesis suggests that the incised valleys formed during the low sea level phase of the LGM were flooded and accumulated sediments during the Holocene transgression (Archer, 2005; Irion et al., 2006, 2011). Sediment cores retrieved from the Tapajós Ria contain fine-grained sediments with 2.2–3.5% of total organic carbon and sedimentation rates reaching up to 6.5 mm yr−<sup>1</sup> (Irion et al., 2006). This reinforces the potential of rias as location of sediment deposition and carbon processing in the Amazon fluvial system.

Rias show sedimentary dynamics similar to that of lakes and reservoirs, which bury more than 200 Tg of organic carbon annually (Dean and Gorham, 1998). Despite the widespread distribution in the Amazon and dominance of organic-rich sediments (Irion et al., 2011), rias are poorly studied with respect to their role as sink or source of carbon in the Amazon basin. According to Sawakuchi et al. (2014), methane fluxes in Amazon rivers account for 22–28% of the global river emissions of methane and the Amazon rias are hotspots of methane production. Rias also serve as analogs for carbon processing in artificial reservoirs like the recently build reservoirs of the Belo Monte hydropower plant in the Xingu River and future reservoirs planned for the Tapajós River (Winemiller et al., 2016). Clearwater Amazon rivers are the main target in the energy expansion plans of the Brazilian government (EPE/MME, 2007), which regards hydropower as inexpensive and clean energy, however, without consideration of impacts of river impoundment on Amazon ecosystems (Fearnside, 2014; Winemiller et al., 2016) and emission of greenhouse gases (Barros et al., 2011; de Faria et al., 2015). To assess the combined effect of carbon fluxes in Amazon rivers, it is imperative to fully understand carbon burial and emission rates in all main Amazon sedimentary environments. In this paper, we focus on carbon dynamics in the rias of the Xingu and Tapajós rivers. Geochemical, magnetic susceptibility, pollen, and diatom analyses combined with luminescence and radiocarbon dating are used to constrain the sources and fate of carbon in sediments of the Xingu and Tapajós rivers on millennial timescales. We integrated our data and results from previous literature to evaluate the origin and content of organic matter and to understand major controls affecting carbon burial in the Xingu and Tapajós rias.

## MATERIALS AND METHODS

#### Sediment Sampling

Sediment cores and samples of riverbed and suspended sediments collected from Xingu, Tapajós and Amazon rivers are presented in **Figure 1** and in Supplementary Material (Table S1). Suspended sediment samples were taken both during the dry and wet seasons from Xingu, Tapajós, and Amazon rivers in order to fully encompass variations in sediment composition related to hydrological changes. Suspended and riverbed sediment samples from the Amazon mainstem were

FIGURE 1 | Study area, location of sediment samples (suspended, riverbed and cores) and bathymetric profiles. In this work, underwater sediment cores were named "XC" and sediment cores from islands were named "EMB." Land-use map from Almeida et al. (2016).

collected upstream and downstream the Tapajós and Xingu rivers mouths (**Figure 1**) in order to evaluate the effect of input of sediments from the Xingu and Tapajós rivers on the Amazon River. Suspended sediment samples were collected through water pumped at approximately 60% of the water depth in the channel thalweg. For each sample, four liters of water were filtrated using cellulose acetate membranes (pore size of 0.45 µm). Riverbed sediments were collected using a Van Veen grab sampler. A total of 23 suspended sediment samples and 28 riverbed sediment samples were used in this study.

Sediment cores of the Xingu Ria were collected from deeper portions of the lake-like channel covered by muddy sediments (named "XC") and from islands in the upstream section of the Xingu Ria (named "EMB"). The locations were selected based on water depth profiles and riverbed sediment sampling (further details in Sawakuchi et al., 2015). Underwater sediment cores from the Xingu Ria channel were retrieved by divers using PVC tubes of up to 6 m in length and percussion method (Sawakuchi et al., 2015). Three underwater sediment cores, characterized by dark gray to brown muds with thicknesses of 300 cm (XC02), 370 cm (XC03), and 470 cm (XC05), respectively, were retrieved from the Xingu Ria. Unconsolidated sediments from islands of the Xingu Ria head were collected using a piston-corer attached to a manual auger. Four sediment cores composed of fine to medium sands and muddy organic sediments were obtained from the islands: EMB13-07 (555 cm), EMB13-13 (430 cm), EMB13-17 (250 cm), and EMB13-21 (345 cm). Samples for luminescence dating were collected using opaque plastic tubes.

Bathymetric profiles were recorded along the upstream part of the Xingu Ria using sonar and GPS equipment. Samples collected with the Van Veen grab sampler were used to characterize riverbed sediments along bathymetric profiles. Water depth profiles and texture of riverbed sediments of the Xingu Ria are shown in **Figure 2**.

## Organic and Inorganic Geochemistry of Sediments

Total organic carbon (TOC) content and its stable carbon isotope ratios (δ <sup>13</sup>Corg) were measured in 28 riverbed sediments samples at MARUM, University of Bremen. Sediments from the core XC05 were sampled at 2 cm intervals for total carbon coulometric and x-ray fluorescence (XRF) analysis at the University of Kentucky.

Major elements (P, Ca, Fe, Mn, Al, Ti, K) concentrations were determined with an Agilent 720 inductively coupled plasma optical emission spectrometer (ICP-OES) for 23 samples of suspended sediments and with a Bruker Tracer IV SD energy dispersive x-ray fluorescence (XRF) spectrometer for 13 riverbed sediments samples and 234 samples of the sediment core XC05. ICP-OES analysis of suspended sediments and XRF analysis of riverbed sediments were performed at MARUM, University of Bremen. For ICP-OES analysis, digestion of suspended material was performed with a MLS 1200 MEGA microwave system. For this purpose, 7 ml HNO<sup>3</sup> (65%), 0.5 ml HF (40%), 0.5 ml HCl (30%), and 0.5 ml MilliQ were added to about 50 mg sample material (filter plus suspended material) previously placed into Teflon liners. All acids were of supra-pure quality. For XRF analysis, sediments were prepared by freeze drying, grinding by hand with an agate mortar and pestle, and sieving to ∼125 µm prior to packing into inert sample boats sealed with an ultra-thin mylar window. Individual XRF scans for each sample were made through the mylar window over 90 s, in order to maximize signal-to-noise. Major elements (atomic numbers 11 through 30) were collected under a ∼9 torr vacuum, by setting the XRF filter to 15 keV and 35 µA. Trace elements (atomic numbers 20 through 51), were collected at 40 keV and 15 µA without a vacuum. Calibration of unknowns utilized Bruker proprietary software (S1CALPROCESS with TR2.cfz and MA1.cfz) and a reference catalog of mudrock chemistry described in Rowe et al. (2012). Internal consistency of the XRF was verified by routine scanning of the SARM41 shale standard (Ring, 1989). The chief focus of our analysis was the major rock-forming (Al, Ca, K, Ti, P) and redoxsensitive (Mn, Fe) elements. Inorganic chemistry data provide insights on mineralogy as well as depositional environments and hydrodynamic processes, especially when paired with organic carbon data (Sageman et al., 2003; Algeo and Rowe, 2012; Scott and Lyons, 2012).

Total organic carbon (TOC) was determined using a LECO CS 200 CS-Analyzing System and an UIC Coulometrics Inc. Carbonate Coulometer (Engleman et al., 1985). The LECO method determines total carbon content through combustion of a powdered sample (Jarvie, 1991). Values of TOC were obtained by subtracting total inorganic carbon determined by the coulometer from total carbon. This method is routinely employed in paleoenvironmental analysis due to its accuracy and small sample size requirements (Jackson and Roof, 1992; Schulte et al., 2000; e.g., Böning et al., 2005). The precision of the analysis, based on analysis of internal standards and replicates, was typically better than ± 0.15%. The δ <sup>13</sup>Corg signature of decarbonated samples was analyzed on a Finnigan MAT Delta plus coupled to a CE elemental analyzer. δ <sup>13</sup>Corg is reported using the delta notation relative to the Vienna Pee Dee Belemnite (VPDB). The uncertainty was less than ± 0.1‰.

Results were analyzed using R software (R Core Team, 2015). Major changes in hydrology and paleoenvironmental conditions were determined using constrained hierarchical cluster analysis of geochemistry and magnetic susceptibility data by the method of incremental sum of squares (CONISS, Grimm, 1987).

### Magnetic Susceptibility

Magnetic susceptibility measurements were performed at the Paleomagnetic Laboratory (USPMag) of the Institute of Astronomy, Geophysics and Atmospheric Sciences of the University of São Paulo (IAG/USP). Magnetic susceptibility is used here as a complimentary method to identify significant variations in paleoenvironmental conditions or sediment composition in core XC05. A total of 194 paleomagnetic specimens were collected from the sediment core using cubic plastic boxes (8 cm<sup>3</sup> ) placed side-by-side throughout the core. Low-field magnetic susceptibility of individual specimen was measured using a Kappabridge MFK1-FA system (AGICO Ltd). Two different frequencies (976 and 15616 Hz) were used in a 200 Am−<sup>1</sup> field at room temperature (Dearing et al., 1996). All data were mass normalized due to the irregular sample mass inside cubic boxes.

#### Pollen and Diatom Analyses

Riverbed samples were subsampled for pollen analyses (Table S1). Samples were processed for extraction of pollen grains using the protocol established by Colinvaux et al. (1999). The analyses were performed in the Institute of Geosciences of the University of São Paulo (IGc/USP) under a microscope with magnification of 600x. A minimum of 300 terrestrial pollen grains were counted per sample. Aquatic taxa and spores are not included in the sum of pollen assemblage. The pollen grains and spores were identified using the pollen reference collection of the University of São Paulo and other published reference materials (Erdtman, 1952; Roubik and Moreno, 1991; Colinvaux et al., 1999). The raw pollen count data were entered into the Tilia software (Grimm, 2011), in order to calculate the taxa percentages and concentrations.

Riverbed and sediment core XC05 were subsampled for diatom analyses. Diatom analysis followed standard procedures discussed by Battarbee et al. (2001). Slides were mounted with Naphrax <sup>R</sup> as medium. Optical observations and counts were conducted at a magnification of 1000x with a Zeiss Axioskop 2 plus microscope. At least 400 valves were counted per slide. Taxonomy and nomenclature followed published literature sources (e.g., Round et al., 1990; Krammer, 2000; Rumrich et al., 2000; Metzeltin and Lange-Bertalot, 2007) and the on-line catalog of valid names (California Academy of Sciences, 2011).

## CH<sup>4</sup> and CO<sup>2</sup> in Sediment Pore Waters

Sediment core XC05 was sub-sampled in field on the same day of collection for the assessment of dissolved methane and carbon dioxide concentrations in pore waters. Aliquots of 5 ml of wet sediment were collected with an open-ended 5 ml polypropylene syringe after carefully drilling holes at 10 cm intervals. Immediately after collection, the aliquots were injected into a 120 ml vial containing 10 ml of NaOH (5%), capped with butyl rubber stoppers and sealed with aluminum crimps. Control standards of ambient air were collected during sampling for corrections. In the Center for Nuclear Energy in Agriculture of the University of São Paulo (CENA/USP), samples were shaken for headspace equilibration and 60 ml of gas sample was retrieved from vials headspace using a 60 ml syringe and analyzed by Cavity Ring Down Spectroscopy using a Picarro G2201-i device. Samples were slowly injected following the pump flow, taking care not to change the internal pressure, which can be monitored by the analyzer. The 60 ml analyzes took approximately 3 min to pass through the analyzer giving a steady measurement for roughly 40 s that was checked with three standard gas mixtures for concentration and carbon isotope composition of CH4. Reproducibility between replicates presented a variation of the standard deviation >0.5% for concentration and >0.9% for δ <sup>13</sup>CCH4. Readings of concentration (ppm) and δ <sup>13</sup>CCH4 (‰) by the analyzer were slightly different than the standard reference values. Thus, sample results were corrected according to a standard calibration curve.

## Optically Stimulated Luminescence and Radiocarbon Datings

Optically stimulated luminescence (OSL) dating was carried out in fine silt or sand quartz aliquots from cores XC02, XC03, and XC05 in the Luminescence and Gamma Spectrometry Laboratory of the IGc/USP. Quartz aliquots for luminescence measurements were prepared under subdued red light. Wet sieving and settling procedures were used to isolate the 4– 11 or 180–250 µm grain sizes. The target grain size fractions were submitted to oxygen peroxide (H2O2) and hydrochloric (HCl 10%) treatments to remove organic matter and carbonates, respectively. Heavy liquid separation of quartz sand grains was performed using lithium metatungstate solution at densities of 2.75 and 2.62 g cm−<sup>3</sup> to remove heavy minerals and feldspar grains, respectively. Concentrates of quartz sand grains were etched with hydrofluoric acid (HF 38%) for 40 min in order to remove the outer rind of quartz grains damaged by alpha particles and remnant feldspar grains. Luminescence measurements were performed in two automated Risø OSL/TL reader model DA-20 equipped with blue (470 nm) and infrared (870 nm) LEDs for light stimulation, Hoya U-340 filter for light detection in the ultraviolet band and built-in beta radiation sources (90Sr/90Y) delivering dose rates of 0.088 and 0.135 Gy s−<sup>1</sup> (for aluminum discs). Equivalent doses were determined through the singlealiquot regenerative dose protocol (Murray and Wintle, 2000). For fine silt, a mean a-value of 0.04 was considered for calculation of alpha dose rate in quartz. Tests with infrared stimulation indicated absence of feldspar in the fine silt concentrates. A dose recovery test was performed to set up the luminescence measurements protocol to the studied samples batch. Samples equivalent doses were calculated through the Central Age Model (Galbraith et al., 1999). The concentrations of K, <sup>238</sup>U and <sup>232</sup>Th for dose rate calculations were determined using high resolution gamma spectrometry with a high purity germanium detector (relative efficiency of 55% and energy resolution of 2.1 KeV) encased in an ultralow background shield. Samples were packed in sealed plastic containers and stored for at least 28 days for radon equilibration before gamma spectrometry. Radiation dose rates were calculated using conversion factors outlined by Guérin et al. (2011). Radiation dose rates were corrected for water saturation (water weight/dry sample weight). Cosmic dose rates were calculated through samples latitude, longitude, elevation and burial depth, according to Prescott and Hutton (1994).

Accelerator mass spectrometry (AMS) radiocarbon (14C) dating was performed on leaves and charcoal fragments retrieved from the sediment cores. Samples were treated in the Radiocarbon Laboratory of the Illinois State Geological Survey and submitted to the Keck Carbon Cycle AMS Laboratory of the University of California-Irvine for AMS <sup>14</sup>C analysis. The <sup>14</sup>C dates were calibrated using the program Calib 7.0 and the calibration curve IntCal 13 (Stuiver and Reimer, 1993; Reimer et al., 2013).

#### RESULTS

#### Geochemistry of Suspended and Riverbed Sediments

Geochemical results are summarized in **Tables 1, 2** (discretized dataset in Table S1). Strong variations in suspended sediment composition were identified between wet and dry seasons in the Xingu, Tapajós, and Amazon rivers (**Table 1**, **Figure 3**). Seasonal changes in suspended sediments of the Xingu and Tapajós rivers include the increase in the relative amount of P in sediments of the dry season, which is directly related to primary productivity (Engstrom and Wright, 1984; Dean and Gorham, 1998). The P/Ti ratio is an effective proxy to evaluate the sedimentation of phosphorous regardless of terrigenous input and it avoids dilution effects caused by changes in sedimentation rate (Latimer and Filippelli, 2002; Filippelli et al., 2003). Average P/Ti ratios in the Xingu River vary from 0.66 in the wet season to 4.20 in the dry season. Seasonal changes in P concentration have lower amplitudes in the Tapajós River, with the P/Ti ratio shifting from 0.75 to 0.94 from the wet to the dry seasons, respectively. The Fe/K ratio is a useful indicator of mineralogical stability, since the most stable rock-forming minerals have lower Fe/K ratios (K-feldspar, muscovite, and quartz) and unstable minerals tend to be richer in iron content (Herron, 1988; Armstrong-Altrin and Machain-Castillo, 2016). Fe/K ratios increase during the wet season both in the Xingu and Tapajós rivers, with relatively little variation in the Amazon River (**Table 3**). In the Xingu River, Ca concentrations increase from 0.09 to 0.35 mg l−<sup>1</sup> and P concentration from 0.03 to 0.71 mg l−<sup>1</sup> between the wet and TABLE 2 | Summary of major elements concentrations, TOC and δ13Corg of bulk organic matter in bottom sediments of the Xingu, Tapajós, and Amazon rivers.


Data of individual samples are presented in Table S1 of the Supplementary Material.

dry season. Fe concentrations shift from 1.11 to 2.56 mg l−<sup>1</sup> from the wet to the dry season. The inverse pattern is observed in the Amazon River, with Ca, P, and Fe concentrations decreasing in the dry season (**Table 1**, **Figure 3**). In the Tapajós River, these shifts between wet and dry seasons have lower amplitude, with Ca concentration varying from 0.07 to 0.13 mg l−<sup>1</sup> , P varying from 0.03 to 0.05 mg l−<sup>1</sup> and Fe varying from 0.99 to 0.76 mg l−<sup>1</sup> .

#### Pollen and Diatoms in Riverbed and Core Sediments

In total, 45 pollen taxa were identified in riverbed sediments of the Xingu Ria (Figures S1–S4). The vegetation represented by pollen retrieved from riverbed sediments has great diversity and high percentages of forest elements. Arboreal taxa encountered in the samples include Acalypha, Alchornea, Annacardiaceae, Apocynaceae, Arecaceae, Attalea, Bignoniaceae, Cecropia, Combretum, Dalbergia, Didymopanax, Euterpe, Fabacea, Genipa, Machaerium, Mauritia, Mauritiella, Matayba, Melastomataceae, Mimosa, Myrtaceae, Palmae, Psidium, Psychotria, Sapium, Spondias, Sterculiaceae, Stigmaphyllon, Talisia, and Zygia. Terrestrial herbaceous elements are represented by Alternanthera, Asteraceae, Bambusa, Begonia, Caesaria, Lamiaceae, Phyllanthus, Pilea, Poaceae, Polygonaceae


Data of individual samples are presented in Table S1 of the Supplementary Material.

FIGURE 3 | Scatterplots comparing P, Ti and K concentrations in suspended sediments collected in the Xingu, Tapajós and Amazon rivers during the wet and dry seasons.

TABLE 3 | Variation of elemental ratios in suspended sediments of the Xingu, Tapajós, and Amazon rivers during the wet and dry seasons.


and Symmeria. Sagitaria, Cyperaceae, Alismataceae represent taxa of aquatic herbs. Spores identified in the analysis include Polypodiaceae, trilete and Monolete. The palynologic data indicate that riverbed sediments of the Xingu Ria record the input of sediments from areas where the predominant vegetation type is forest (Figure S5).

Regarding the diatoms, a total of 65 infrageneric taxa were identified in riverbed sediments of the Xingu Ria. Eunotia and Gomphonema were the most represented genera regarding species number (24.6% of total species), followed by genus Aulacoseira (9.2%) (Figure S6). Eunotia species are mostly acidophilic, oligotrophic, and typified by a dominant periphytic habitat (Round et al., 1990; van Dam et al., 1994; Moro and Fürstenberger, 1997; Hamilton and Siver, 2010). Due to the secretion of mucilage by the apical pore fields, Gomphonema is a common genus in periphytic algal communities, being well represented in richness and density (Tremarin et al., 2009). Aulacoseira is a common planktonic genus inhabiting lacustrine and running freshwaters, developing in various trophic conditions (Denys et al., 2003; Zalat and Vildary, 2007). Ecological information at species level is required to use this genus as indicator of water quality (Bicudo et al., 2016).

Diatoms presented well-preserved frustules throughout the core. Assemblages from the core XC05 base showed little variation among genera, and besides the planktonic genus Aulacoseira, comprised benthic and periphytic genera that require surfaces for developing (e.g., Eunotia, Pinnularia and Surirella), suggesting a water column under acidic and oligotrophic conditions with high light penetration. Genera Surirella, Diploneis, Aulacoseira, Placoneis, Gomphonema, Encyonema, and Eunotia were observed throughout the core, suggesting a relatively uniform slack water environment.

#### Chronology, Magnetic and Geochemistry Data of Sediment Cores

Sediments of the three studied underwater cores (XC02, XC03, and XC05) are composed of dark gray to brown organic-rich muddy sediments. As highlighted in **Figure 2**, these sediments are similar to the modern surface sediments that accumulate in


TABLE 4 | Radionuclides concentrations, radiation dose rates, equivalent doses and luminescence ages of sediment samples retrieved from cores XC02, XC03, and XC05.

most of the rias. Samples from the top and bottom of sediment cores XC02 (10–293 cm), XC03 (10–363 cm) and XC05 (45– 440 cm) were collected for OSL dating. The luminescence ages of cores XC02, XC03, and XC05 ranged from 118 ± 81 to 4,086 ± 159 years (**Table 4**), suggesting that the Xingu Ria is a slack water environment acting as trap of fine-grained sediments at least since the middle Holocene. XC02 sediments showed ages from 2,503 ± 462 to 3,295 ± 542 years. XC03 and XC05 sediments cover periods from 118 ± 81 to 1,251 ± 211 years, and from 3,025 ± 185 to 4,086 ± 159 years, respectively. Sediment depth profiles described in the islands of the head of the Xingu Ria (EMB13-07, EMB13-13 and EMB13-17) show dark gray organic-rich sediments interlayered with sands and light gray floodplain sediments (**Figure 4**). Radiocarbon ages obtained in the dark gray muds, which likely represent buried ria sediments, varied from 141–267 to 5,479–5,580 cal years BP (**Table 5**, **Figure 4**).

Luminescence and radiocarbon ages allowed the calculation of sedimentation rates for the organic-rich sediments of the Xingu Ria. In the Xingu Ria, sedimentation rates calculated through luminescence ages of core XC05 varied between 0.27 cm yr−<sup>1</sup> and 0.70 cm yr−<sup>1</sup> (average of 0.37 cm yr−<sup>1</sup> ). The rate of deposition of core XC05 increases from about 0.28 cm yr−<sup>1</sup> during the period of ∼4,100 to ∼3,300 years ago to an average of 0.60 cm yr−<sup>1</sup> during the period of ∼3,300 to ∼3,000 years ago. Cores XC02 and XC03 presented average sedimentation rates of 0.36 cm yr−<sup>1</sup> and 0.31 cm yr−<sup>1</sup> , respectively. Sedimentation rates for organicrich sediments underneath floodplain sediments of the Xingu Ria head were 0.24 cm yr−<sup>1</sup> , 0.78 cm yr−<sup>1</sup> , 0.05 cm yr−<sup>1</sup> in profiles EMB13-07, EMB13-13 and EMB13-17, respectively (**Figure 4**).

Bertassoli Jr. et al. Carbon in Amazonian Clearwater Rivers

TABLE 5 | Radiocarbon ages of sediment cores EMB13-07, EMB13-13, EMB13-17, and EMB13-21.


In the Tapajós Ria, sedimentation rates were calculated through radiocarbon ages presented in Irion et al. (2006). Rates of sediment deposition in the Tapajós Ria vary from 0.24 cm yr−<sup>1</sup> to 0.65 cm yr−<sup>1</sup> (average of 0.37 cm yr−<sup>1</sup> ).

Magnetic susceptibility measurements were performed on the XC05 core. Low-field magnetic susceptibility varies from 3.4 × 10−<sup>8</sup> to 1.9 × 10−<sup>7</sup> m3kg−<sup>1</sup> (**Figure 5**). Strong magnetic susceptibility variations (peaks) may indicate variations in sedimentation rate or increased magnetic mineral concentrations. The dendogram of geochemical data demonstrates abrupt shifts in the compositional similarity of sediments deposited at 3,690–3,650, 3,345–3,335, and 3,235– 3,240 years ago. These shifts fit well with peaks in magnetic susceptibility and P, Fe, Mn, and TIC concentration as well as with the decrease in TOC and lithophilic elements (**Figure 5**).

Ti concentration reveals significant positive correlations with Al (r = 0.81, n = 198) and K (r = 0.75, n = 198). These elements were used as proxies for the relative contribution of terrigenous sediment input from the Xingu River to the downstream Xingu Ria lake (Engstrom and Wright, 1984; Boës et al., 2011). Results show significant inverse correlation among P concentration and lithophilic elements represented by Ti (r = −0.31, n = 198), Al (r = −0.32, n = 198) and K (r = −0.46, n = 198). P concentration also shows significant positive correlations with Fe (r = 0.57, 198), TIC (r = 0.51, n = 198) and Mn (r = 0.51, n = 198).

Total organic carbon (TOC) concentration varies between 1.77 and 3.35% (average of 2.45%) in sediment core XC05. The measured concentrations are similar to values in samples from bottom sediments of the Xingu Ria (Häggi et al., 2016). Values of δ <sup>13</sup>Corg range from −31.04 to −27.49‰, and indicate that organic matter from bottom sediments of the Xingu Ria is mainly derived from arboreal C3 vegetation (Table S1). This is supported by palynological data, with pollen assemblages dominated by forest taxa (Figure S1).

CH<sup>4</sup> concentrations in pore waters reach values up to 25% (volume) in XC05 sediment core (**Figure 5**). The values of δ <sup>13</sup>CCH4 in pore waters of XC05 core range from −71 to −57‰ PDB. Higher (more positive) values of δ <sup>13</sup>CCH4 often agree with peaks in magnetic susceptibility, Fe concentration and P/Ti (**Figure 5**). This suggests that CH<sup>4</sup> oxidation in pore waters might increase during drier periods.

#### DISCUSSION

Despite the relatively low suspended load in clearwater rivers, the composition of suspended sediments in the Xingu and Tapajós rivers are organic-rich in comparison with white water rivers, showing a marked variation between the dry and wet seasons. Major variations in sediment composition occur due to seasonal changes of hydrology, driving the input of organic and inorganic compounds and regulating primary productivity. The seasonal changes in river discharge controls the relative concentrations of lithophilic elements (Ti, Al, K) compared to elements influenced by redox conditions in the water column and riverbed (P, Fe, Mn). While the uncertainty of organic carbon sources and its stability through time limits its use as proxy for lake productivity or paleo-redox conditions, phosphorous concentration is a useful indicator of paleo-productivity (Engstrom and Wright, 1984; Dean and Gorham, 1998; Boyle, 2001). In order to compensate for fluctuations in allochthonous (terrigenous) inputs of sediments to the ria lake, the P to Ti ratio was chosen as productivity indicator (Latimer and Filippelli, 2002; Filippelli et al., 2003). Average P/Ti values vary from 4.20 to 0.66 in suspended sediments of the Xingu River during the dry and wet seasons, respectively. The higher values during the dry season are linked to higher primary productivity favored by low water turbidity.

There are significant positive correlations between Fe and Ca (p = 0.71, n = 198), Fe and P (p = 0.57, n = 198) and Fe and TIC (p = 0.97, n = 198) concentrations in core XC05, which point to biogeochemical controls on Fe and TIC deposition in sediments. The high concentration of Fe relative to K in suspended sediments transported during the dry season suggests that assimilation of Fe in sediments is favored by the increase in organic productivity and oxygenated bottom waters (**Table 3**). Major decreases in TOC concentrations are correlated to peaks of P/Ti, TIC, Fe and Mn, suggesting lower preservation of organic matter during drier periods (**Figure 5**). The lack of correlation between TOC and TIC in core XC05 indicates that the shallow bathymetry during drier low-water periods likely decreased calcite solubility, enhanced evaporation and led to the elevated concentration of the carbonate ions (Kelts and Talbot, 1990). Mn and Fe in sediments are remobilized under reducing conditions and precipitated in the presence of oxygenated waters (Davison, 1993). Lower values of Fe/Mn ratios, proposed by Mackereth (1966) as a redox proxy, are correlated with peaks of P/Ti, suggesting abrupt shifts in redox conditions during sediment deposition in the Xingu Ria. Peaks of P/Ti indicate dominant wetter conditions punctuated by dry events at 3,690–3,650, 3,345–3,335, and 3,235–3,240 years ago (**Figure 5**). Nutrient-rich and oxygenated waters favor the accumulation of P, Fe and Mn in sediments deposited under drier and low-water level periods. In contrast, TOC is negatively correlated with P, Fe and Mn concentrations due to the lower organic matter preservation under oxic conditions prevailing during drier periods. During the

low water season, the shallower water column would be better mixed and the oxic water in contact with the bottom sediment would lead to a faster degradation of the deposited organic matter. Thus, considering the redox and productivity proxies already described, the variation in TOC concentration through time is mainly derived from changes in the water-sediment interface from anoxic/dysoxic (higher TOC) to oxic (lower TOC) conditions that may be related to hydrologic changes in the area drained by the Xingu River.

Fine-grained sediments accumulated in Amazon rias have relatively high concentration of organic carbon (average TOC of 2.45%) compared to fine-grained sediments deposited in floodplains under influence of white water rivers such as the Amazon River (0.4–1.4%; Moreira-Turcq et al., 2004; Aniceto et al., 2014; Moreira et al., 2014). Thus, rias formed in large clearwater rivers such as the Xingu and Tapajós rivers act as hotspots for carbon storage within rivers in the Amazon Basin. These areas are present at least since the middle Holocene, considering the ages obtained for the sediment cores from the Xingu (>4,000 years, this study) and Tapajós rivers (>9,600 years, Irion et al., 2006). Stable carbon isotope data (δ <sup>13</sup>Corg from −31.04 to −27.49‰) indicate a dominant contribution of C3 plants and possibly phytoplankton for the organic particles accumulated in the ria bottom. Palynological data document a large contribution of C3 arboreal plants as major sources of carbon in sediments of the Xingu and Tapajós rias. The labile organic matter derived from phytoplankton may be rapidly consumed and turned into CO<sup>2</sup> in the oxic surface sediment layer, while the more recalcitrant terrestrial organic matter accumulates. Also, once buried in the anoxic sediment, the

autochthonous and labile organic matter may be used to fuel the CH<sup>4</sup> production within sediments. This mechanism is in accord with the high CH<sup>4</sup> concentrations in pore waters and released to the atmosphere obtained by this work and by Sawakuchi et al. (2014). Despite the large carbon storage, the Xingu and Tapajós rias show the highest fluxes of CH<sup>4</sup> to the atmosphere among the major Amazon tributaries (Sawakuchi et al., 2014), suggesting that sedimentary conditions suitable for organic matter preservation favor anoxic conditions in the sediments, where CH<sup>4</sup> can be produced and released to the atmosphere. This is supported by the high concentrations (up to 24%) of CH<sup>4</sup> in sediment pore waters of the Xingu Ria, with δ <sup>13</sup>CCH4 varying from −70.4 to −56.9‰, suggesting low CH<sup>4</sup> oxidation rates. These results indicate that rias in clearwater Amazon rivers are significant organic carbon sinks and sources of CH<sup>4</sup> to the atmosphere. They should be considered as distinct environments in the Amazon in order to refine modern carbon balance assessments.

The pollen assemblages show a high diversity of arboreal taxa, pointing to the terra firme and igapó forests as main sources of organic matter within sediments of the Xingu River catchment. Despite the increasing deforestation in the Xingu catchment during the last several decades (Barona et al., 2010), organic carbon stored in riverbed sediments is mainly derived from forest. This could be related to a delay in transfer of the deforestation signal to river sediments, or a dominant input of organic matter from the igapó flooding forest tract along the river channel. The diatom assemblages in riverbed sediments and the sediment core point to a slack water environment in the Xingu Ria since the middle Holocene. Considering that the organic matter pool in sediments of the Xingu and Tapajós rias is mainly terrigenous and derived from forested areas, the input of carbon to the ria sediments is favored during high precipitation episodes while carbon preservation is controlled by duration and frequency of dry periods. Low sedimentation rates combined with well oxygenated waters due to higher primary productivity in the water column favor organic matter degradation in riverbed sediments during dry periods. Despite high rates of organic matter degradation during low-water periods, the higher primary productivity favors the uptake of CO<sup>2</sup> and carbon recycling in the water column and riverbed. During high-water phases, the flux of CO<sup>2</sup> to the atmosphere increases due to lower primary productivity, while higher sedimentation rates favor organic matter preservation and carbon uptake in sediments. High primary production is the main indicator of CO<sup>2</sup> undersaturation observed in large clearwater rivers in the Amazon during the low-water phase (Rasera et al., 2013).

The sediment cores retrieved in the Xingu Ria have ages varying from 3,025 ± 185 to 4,086 ± 159 years (XC05), from 118 ± 81 to 1,251 ± 211 years (XC03) and from 2,503 ± 462 to 3,295 ± 542 years (XC02). The cores were collected in the deepest part of the river profile and core tops with ages between 3,025 ± 185 (XC05) and 2,503 ± 462 (XC02) years indicate erosional processes and sediment reworking within the ria. The lack of recent deposits indicates that erosion due to channel migration may remobilize sediments in certain portions of the ria. The average sedimentation rate obtained in this study for the Xingu Ria is 0.35 cm yr−<sup>1</sup> since the middle Holocene (∼4,000 years). On millennial timescales, sediment deposition rates at different locations are relatively stable. A similar average sedimentation rate of 0.37 cm yr−<sup>1</sup> was also observed for the Tapajós Ria over the last ∼11,000 years Irion et al. (2006). Sedimentation rates from 0.35 to 0.37 cm yr−<sup>1</sup> are relatively high compared to sedimentation rates observed in low latitude lakes, which are typically around 0.05 to 0.25 cm yr−<sup>1</sup> (Cross et al., 2000; Chu et al., 2002; Hodell et al., 2005; Caballero et al., 2006; Conroy et al., 2008; McGlue et al., 2011). Our data show that sedimentation rates varied through time in core XC05. The temporal and spatial variations in sedimentation rates would result from hydrologic changes and sediment reworking within the Ria. The migration of underwater channels and wave action would be the major autogenic processes responsible for sediment remobilization and spatial changes in sedimentation rates. Temporal variations would also be related to precipitation changes affecting sediment supply and accommodation space within the ria. Additional data are required to calculate sedimentation rates with higher temporal resolution and allow a comparison with paleoprecipitation records. This is necessary to constrain the major controls on sedimentation rates and the role of precipitation changes for carbon burial. The TOC concentration in core XC05 varies from 1.77 to 3.35% (average of 2.45%), similar to TOC values measured in bottom sediments collected in the Xingu Ria (average of 2.36%) (Häggi et al., 2016). Irion et al. (2006) showed TOC values with low variation in the Tapajós Ria, varying from 2.2 to 2.9% in sediments younger than 7,500 years. The high sedimentation rates would reduce the effectiveness of degradation processes, implying that the Xingu and Tapajós rias are potentially massive carbon sinks on a millennial timescale.

We estimated rates of organic carbon burial in the Xingu and Tapajós rias based on a dry bulk density of 1.3 g cm−<sup>3</sup> for the bottom ria sediments, average sedimentation rates of 0.35 cm yr−<sup>1</sup> and 0.37 cm yr−<sup>1</sup> (Irion et al., 2006) for the Xingu and Tapajós rias, respectively, and the minimum and maximum TOC concentrations obtained in this work and by Irion et al. (2006). In the Xingu Ria, carbon burial ranges from 84 g m−<sup>2</sup> yr−<sup>1</sup> to 159 g m−<sup>2</sup> yr−<sup>1</sup> (average of 116 g m−<sup>2</sup> yr−<sup>1</sup> ) and from 106 to 169 g m−<sup>2</sup> yr−<sup>1</sup> (average of 121 g m−<sup>2</sup> y −1 ) in the Tapajós Ria. Assuming that 75% (conservative estimate) of the ria surface is covered by fine-grained organicrich sediments (**Figure 2**), the Xingu Ria buries from 0.08 Tg C yr−<sup>1</sup> to 0.16 Tg C yr−<sup>1</sup> (average of 0.11 Tg C yr−<sup>1</sup> ) and the Tapajós Ria buries from 0.19 Tg C yr−<sup>1</sup> to 0.30 Tg C yr−<sup>1</sup> (average of 0.21 Tg C yr−<sup>1</sup> ). Conceding that the Xingu and Tapajós Rias were totally flooded at least since the middle Holocene (∼5,000 years), around 560 Tg C and 1050 Tg C of carbon are stored in sediments of the Xingu and Tapajós rias, respectively.

A rough attempt to estimate the carbon budget of the Xingu Ria is possible considering the average accumulation rate of carbon in sediments of the Xingu Ria of 0.11 Tg C yr−<sup>1</sup> (this study) and the carbon evasion estimates of 8 Tg C yr−<sup>1</sup> as CO<sup>2</sup> (Sawakuchi et al., this volume) and 0.053 Tg C yr−<sup>1</sup> as CH<sup>4</sup> (Sawakuchi et al., 2014) fluxes to the atmosphere. These estimates indicate that despite the significant carbon pool and storage, the Xingu Ria is a net source of carbon to the atmosphere. Furthermore, the Xingu Ria can serve as an analog for sediment accumulation and carbon processing in the reservoirs of the Belo Monte hydropower plant. The multidecadal carbon sedimentation pattern estimated for the Xingu Ria allows projections about sediment accumulation and carbon processing in the Belo Monte reservoirs.

#### CONCLUSIONS

The lowermost sectors of the Tapajós and Xingu rivers have lakelike (fluvial ria) sedimentary dynamics due to the impoundment by the Amazon mainstream during the Holocene. This lakelike physiography hinders the transfer of sediments to the Amazon River and induces the accumulation of fine-grained organic-rich sediments derived from upstream areas. Thus, the downstream reaches of these major clearwater rivers are sinks for particulate organic matter chiefly derived from arboreal forest (terra firme and igapó). Wetter periods favor the supply and preservation of forest-derived organic matter in sediments while drier periods increase the organic productivity in the water column and favor degradation of organic matter in riverbed sediments. Microbial consumption of organic matter in the oxic surface layers of the sediment contributes to the anoxic conditions that prevail within sediments, with methanogenesis being an important path for biodegradation of particulate organic matter. During dry and low-water level periods, the flux of CH<sup>4</sup> to the atmosphere increases while the CO<sup>2</sup> flux decreases due to the higher organic productivity within the water column. Thus, seasonal variations play an important role for carbon dynamics in clearwater rivers. Despite the significant carbon storage that occurs in the Xingu and Tapajós rias at least since the middle Holocene, our carbon balance estimates indicate that Amazon rias would act as source of carbon to the atmosphere. A future drier climate as projected for eastern Amazonia would reduce the carbon sink potential of clearwater rivers and increase the methane flux to the atmosphere.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

DJB, AS, and HS were responsible for design of the study and preparation of the manuscript. AS, HS, GH, DJB, and FP organized overall project logistics. DJB and HS executed the sampling and analysis of CH<sup>4</sup> in sediment core and performed calculations of carbon balance. FP and AS executed luminescence dating. GH performed magnetic susceptibility analysis. MM, FP, and DJB performed sampling and geochemical analysis of the sediment core. CC, MZ, and ES executed geochemical analysis of suspended and riverbed sediments. TP collaborated with sediment and greenhouse gas sampling and supported local logistics. RS and PO executed palynology analysis. SF and DCB performed diatom analysis. All authors contributed to data interpretation and critically revised the manuscript for final submission.

### FUNDING

This study was funded by FAPESP grants #2011/06609-1, #2014/23334-4, #2016/02656-1 and CAPES grant #AUXPE 2043/2014.

### ACKNOWLEDGMENTS

We thank Luciana Nogueira and Thays Desiree Mineli for the support with luminescence dating procedures. Bailee Hodelka, Joseph Lucas, and Jason Backus assisted with XRF and coulometry measurements. We also thank the reviewers for constructive criticisms and valuable comments. Open access publication fees were supported by the Gordon and Betty Moore Foundation Marine Microbial Initiative.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2017.00044/full#supplementary-material

Palaeoclimatol. Palaeoecol. 415, 165–174. doi: 10.1016/j.palaeo.2014. 08.018


(8 pages) in the Abstract-Volume (CD-Rom) of the International Symposium on "Hydrological and Geochemical Processes in Large Scale River Basins" (Manaus), 1–5.


mudrock chemostratigraphy and geochemistry. Chem. Geol. 324, 122–131. doi: 10.1016/j.chemgeo.2011.12.023


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Bertassoli, Sawakuchi, Sawakuchi, Pupim, Hartmann, McGlue, Chiessi, Zabel, Schefuß, Pereira, Santos, Faustino, Oliveira and Bicudo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Flux of Dissolved and Particulate Low-Temperature Pyrogenic Carbon from Two High-Latitude Rivers across the Spring Freshet Hydrograph

#### Allison N. Myers-Pigg<sup>1</sup> \* † , Patrick Louchouarn1, 2 and Roman Teisserenc<sup>3</sup>

#### Edited by:

Nicholas David Ward, University of Florida, USA

#### Reviewed by:

Rudolf Jaffe, Florida International University, USA Alysha Coppola, University of Zurich, Switzerland

> \*Correspondence: Allison Myers-Pigg anmp@tamu.edu

#### † Present Address:

Allison N. Myers-Pigg, Memorial University of Newfoundland, Department of Earth Sciences, St. John's, NL, Canada

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 07 October 2016 Accepted: 31 January 2017 Published: 15 February 2017

#### Citation:

Myers-Pigg AN, Louchouarn P and Teisserenc R (2017) Flux of Dissolved and Particulate Low-Temperature Pyrogenic Carbon from Two High-Latitude Rivers across the Spring Freshet Hydrograph. Front. Mar. Sci. 4:38. doi: 10.3389/fmars.2017.00038 <sup>1</sup> Department of Oceanography, Texas A&M University, College Station, TX, USA, <sup>2</sup> Department of Marine Sciences, Texas A&M University at Galveston, Galveston, TX, USA, <sup>3</sup> EcoLab, Université de Toulouse, CNRS, Institute National Polytechnique de Toulouse, Université Paul Sabatier Toulouse, France

A number of recent studies have documented that pyrogenic carbon (PyC) is an integral and significant proportion of DOM in worldwide rivers. This material originates from all fractions of the PyC continuum, from highly condensed PyC to more functionalized components that retain some structural identity of fuel molecules. Understanding the transfer of PyC to river systems is paramount for Arctic regions, given the projected increase in frequency and intensity of forest fires within these ecosystems. However, the environmental distribution and concentration of soluble and particulate PyC, parameters that govern the overall fate of PyC in aquatic systems, has so far been unstudied. Here, we analyze the concentration and phase distribution of the anhydrosugar biomarker levoglucosan, as a proxy for low-temperature PyC, in two high-latitude river systems: a small sub-Arctic Canadian river, the Great Whale River in northern Québec, and the largest Arctic River, the Yenisei River in north-central Siberia. Low-temperature PyC, as estimated by levoglucosan concentrations, is exported predominantly in the dissolved phase. Peak export of low-temperature PyC occurs during the spring freshet period in both rivers. Seasonal variability of dissolved and particulate PyC export in each river elucidated that the export of PyC in the particulate and dissolved phases were temporally decoupled throughout the peak discharge events. While the present work confirms that levoglucosan is exported in particulate phase at a high enough level to enter sedimentary deposits and record historical wildfire signatures, as the phase distribution varies between rivers and during different flow regimes, spatial and temporal differences may affect the usage of levoglucosan as a PyC proxy in depositional settings.

Keywords: pyrogenic carbon, levoglucosan, freshet, Arctic Rivers, dissolved pyrogenic carbon, particulate pyrogenic carbon

## INTRODUCTION

Vegetation fires affect carbon cycling in all major earth systems. These fires release 1.6–2.8 petagrams of carbon (Pg C) to the atmosphere worldwide (Santín et al., 2016), mostly as gaseous CO2. However, inefficient combustion conditions in the environment regularly leave a proportion of the fire-affected carbon pool as carbonaceous residues, called pyrogenic carbon (PyC), which varies in molecular structure and environmental reactivity. The fate of pyrogenic carbon (PyC) in the environment, and the subsequent impact PyC has on carbon cycling, are a result of both its transport and degradation potential. As PyC is comprised of a continuum of materials, of which their formation and resulting chemical structures are temperature dependent (Masiello, 2004), the nature of PyC varies dramatically along the PyC continuum. The PyC continuum is marked by sequential changes in the proportion of chemical markers of combustion by-products, and increased aromatic condensation level (decreasing the ratio of Hydrogen to Carbon; H/C ratios) along temperature gradients (0–1,000◦C; Kuo et al., 2008a,b, 2011a; Keiluweit et al., 2010; **Figure 1**). At low-temperatures (150–300◦C), macropolymer dehydration and fragmentation leads to the release of free monomers and oligomers rich in −OH and −COOH functionalities (Kuo et al., 2008a,b, 2011a; Keiluweit et al., 2010; Harvey et al., 2012; Norwood et al., 2013). At these temperatures, the formation of anhydrosugars (levoglucosan and its isomers, mannosan, and galactosan) is prevalent (Kuo et al., 2008a; **Figure 1**). At intermediate temperatures (350–500◦C), the rapid disappearance of recognizable plant lignocellulosic macromolecules, and the rapid drop in levoglucosan yields around 300–400◦C (Kuo et al., 2008a, 2011a), occur synchronously with a steep increase in the proportion of amorphous aromatic structures (Keiluweit et al., 2010; Schneider et al., 2011; Harvey et al., 2012) as noted by the rapid decrease in H/C ratios and the rise in benzene polycarboxylic acids (BPCA), selective molecular markers of aromatic clusters (Ziolkowski and Druffel, 2010; Schneider et al., 2011; **Figure 1**). At the highest range of the temperature continuum (>500◦C), PyC is characterized by increasing graphene-like structures in highly condensed

Frontiers in Marine Science | www.frontiersin.org

et al. (2008a).

turbostratic aromatics, marked by the increase in condensation levels of molecular markers such as BPCAs (Keiluweit et al., 2010; Schneider et al., 2011). This PyC-derived turbostratic carbon is relatively unordered (Nguyen et al., 2010) and is therefore structurally different than graphite found in natural environments, which is derived from high temperature and pressure conditions (such as within metamorphic rock) and is likely more ordered and highly crystalline (Brandes et al., 2008). High-temperature PyC can also be formed from the condensation of gas-phase intermediates (i.e., soot carbon; Schmidt and Noack, 2000; Hammes et al., 2007). As the lability of organic matter decreases along the PyC continuum (Ascough et al., 2011), the reactivity potential of PyC in the environment decreases with higher combustion temperatures (Masiello, 2004).

Wildfires are an important source of PyC to Arctic environments (Preston and Schmidt, 2006; Hansen et al., 2013; Santín et al., 2015). However, both the overall export and mechanisms for export of PyC within these systems is still an area of active research. Soil regimes in these systems vary widely; for example, the Yenisei River watershed is underlain by every type of permafrost regime, causing the soil active layer depth to vary dramatically within high latitude watersheds (Kawahigashi et al., 2004). It has been suggested that subsoil transport is a primary mechanism for PyC export from terrestrial to aquatic systems (Güereña et al., 2015), which may be regulated by soil-PyC associations. Czimczik and Masiello (2007) suggest that the ability of soil minerals to sorb PyC may be a main controller of PyC storage within soils, similar to bulk soil organic matter (SOM). When PyC is incorporated into soils, it may be associated with soil minerals (Hockaday et al., 2007; Knicker, 2011). PyCmineral interactions observed on the surface of charcoals may also influence associations of PyC with minerals in the dissolved phase within soils (Hockaday et al., 2007), as soil mineral-DOM interactions affect the quantity of DOM exported from soil layers (Kaiser and Guggenberger, 2000; Kawahigashi et al., 2006). The distribution of PyC in Arctic soil profiles indicates vertical movement of PyC between soil layers (Guggenberger et al., 2008), pointing toward transport of Py-DOC within soil profiles. In many high-latitude systems, the highest PyC content within soils is associated with soil mineral layers (Rodionov et al., 2006; Guggenberger et al., 2008). In permafrost regions, DOM export is controlled by active layer thickness and extent of seasonally thawed mineral layers above permafrost regimes (Kawahigashi et al., 2004). These relationships within soils may affect both the phase and timing of PyC export to aquatic systems. Additionally, these processes may govern the turnover time of soil PyC, which is generally <100 years in boreal ecosystems (Hammes et al., 2008). This is much less than the millennial time-scales traditionally thought for PyC turnover in soils, and accounts for all loss processes (decomposition, leaching, erosion). Therefore, the translocation of PyC from soils into riverine systems is likely an important process for PyC cycling.

Export of high-temperature dissolved PyC (dissolved PyC will be henceforth referred to as Py-DOC; high temperature Py-DOC is also known as dissolved black carbon, or DBC) to aquatic systems does not appear to be controlled by fire history (Ding et al., 2013; Wagner et al., 2015), implying that high-temperature PyC needs to be functionalized before export, i.e., the solubility of high-temperature PyC is likely a function of charcoal age and exposure to "weathering" processes (Abiven et al., 2011). This idea is supported by the continual mobilization and export of high-temperature Py-DOC in deforested ecosystems long after slash-and-burn agriculture has stopped (Dittmar et al., 2012). However, as low-temperature Py-DOC also appears to be noncorrelated to recent fire history (Myers-Pigg et al., 2015), the processes controlling the export of low- and high-temperature Py-DOC may be coupled. The relationship between Py-DOC from throughout the continuum and bulk DOC in Arctic river systems (Jaffe et al., 2013; Myers-Pigg et al., 2015; Stubbins et al., 2015) indicates that the mechanisms governing their export are similar. Therefore, it is likely that PyC export is controlled by similar processes as bulk organic matter. However, the temporal variability in PyC mobilization, as well as the variability of PyC export from the entire PyC continuum needs to be further explored before final conclusions on overall PyC export characteristics can be drawn. The export from terrestrial to aquatic systems is important to quantify in order to understand the impacts of PyC on carbon cycling (Masiello and Louchouarn, 2013).

Recent studies have documented the transfer of PyC to river systems in the dissolved and particulate phases, originating from all portions of the PyC continuum, from highly resistant PyC to more soluble, labile components (Dittmar et al., 2012; Jaffe et al., 2013; Myers-Pigg et al., 2015; Wagner et al., 2015; Wang et al., 2016). Understanding the transfer of PyC to aquatic systems is especially pertinent for Arctic regions, given the projected increase in frequency and intensity of boreal forest fires (Kasischke et al., 1995; Running, 2006; Soja et al., 2007) and a larger proportion of PyC created in these environments (Santín et al., 2015). The phase in which these materials are mobilized throughout the environment has implications on its availability for in-situ degradation/transformation, and ability for PyC to be stored in long-term carbon pools (Czimczik and Masiello, 2007; Santín et al., 2016). For example, PyC in the dissolved phase (Py-DOC) can be quickly remineralized (Stubbins et al., 2012; Ward et al., 2014; Myers-Pigg et al., 2015), while PyC in the particulate phase (Py-POC) may be buried in intermediate carbon pools along transport paths (such as river sediments) or on continental margins (Hunsinger et al., 2008; Cotrufo et al., 2016). Given the high potential for biological degradation of levoglucosan and other low-temperature PyC on environmentally relevant time scales (Norwood et al., 2013), one might expect that low-temperature PyC produced in the environment is quickly degraded in-situ during transport from terrestrial systems. However, a few studies have shown that biomarkers of lowtemperature biomass combustion by-products (levoglucosan and isomers) can be found in sedimentary and soil records 100's– 1,000's of years old and help trace fire events induced by climate cyclicity and/or anthropogenic activities (Elias et al., 2001; Hunsinger et al., 2008; Kuo et al., 2011b; Kirchgeorg et al., 2014; Gao et al., 2016). Chemical functionality of low-temperature PyC should favor mobilization and transport in the aqueous phase and therefore, preservation of low-temperature PyC and deposition into sedimentary records may be a function of source and/or associations of low-temperature PyC with minerals or high-temperature PyC components.

Here, we present the first study to our knowledge that analyzes the phase distribution of low-temperature PyC, using levoglucosan as a biomarker proxy, in two high-latitude systems: a small sub-Arctic Canadian River, the Great Whale River, in northern Québec, and the largest Arctic River, the Yenisei River, in north-central Siberia throughout various flow regimes, to discover the evolution and relative importance of phase distribution on the export of low-temperature PyC in these high-latitude systems.

### MATERIALS AND METHODS

For each river, we collected particulate and dissolved organic matter across the spring freshet period. Approximately 100 L of river water was pre-filtered for gross particulates (64 µm) in the field and transported to a field laboratory station in either in the village of Whapmagoostui-Kuujjuarapik for the Great Whale River (2012), or Igarka for the Yenisei River (2014), and immediately filtered using tangential flow filtration (TFF, 0.45 µm filter size) to separate the particulate and dissolved/colloidal phases. The particulate portion, 64–0.45 µm, was concentrated to around 1 litter and immediately frozen. The dissolved portion, <0.45 µm, was isolated using a portable reverse osmosis system (Serkiz and Perdue, 1990), and the concentrate was also frozen. Samples were then transported to the Laboratoire d'Écologie Fonctionnelle et Environnement (EcoLab) laboratory in Toulouse, France, for further treatment. The samples were all freeze-dried at EcoLab, France, and chemical analyses were performed at Texas A&M University at Galveston, USA.

Percent OC (% OC) was determined on vapor-phase acidified (to remove carbonates) freeze-dried subsamples on a Costech Elemental Analyzer. Freeze-dried subsamples were then analyzed for fire-derived biomarkers, monomeric lignin phenols (syringyl, cinnamyl, and vanillyl phenols) and anhydrosugars (levoglucosan, mannosan, galactosan), following methods outlined in Louchouarn et al. (2009), Shakya et al. (2011), and Myers-Pigg et al. (2015). Briefly, biomarkers were extracted from freeze-dried materials, pre-spiked with deuterated (d-7) levoglucosan (NIST SRM 2267), using 9:1 dichloromethane:methanol on an accelerated solvent extractor (ASE 200; Dionex) at 1500 PSI and 100◦C. Sample were then concentrated in a RapidVap (LabConco) under Argon gas at 50◦C, dried to completion in a CentriVap centrifugal concentrator (LabConco) at 50◦C, and re-suspended in pyridine. Extracts were then derivatized for 15 min using O-bis (trimethylsilyl) trifluoroacetamide containing 1% trimethylchlorosilane (9:1 BSTFA:TMCS) at 75◦C, and analyzed on a Varian triple quadrupole 480-300 GC-MS system using a fused silica column (J&W DB-5MS, 30 m × 0.25 mm i.d., 0.25 µm film thickness; Agilent Technologies). Each sample was injected splitless using helium as a carrier gas. Monomeric lignin phenol and anhydrosugar analyses were performed independently under single ion monitoring (SIM) mode, and using specific ions for compound determination. Sample detection limits, recoveries, and precision were determined through concurrently running blanks, standard reference materials (SRM 1649b), and replicates. All quality control parameters were within the ranges presented in Louchouarn et al. (2009).

Ratios of monomeric lignin phenols (syringyl/vanillyl: S/V) and anhydrosugars (levoglucosan to its isomer mannosan: L/M) were calculated to determine sources of PyC in the dissolved and particulate phases (**Table 1;** Kuo et al., 2008a; Myers-Pigg et al., 2015).

Low-temperature Py-DOC concentrations (**Table 1**) were calculated according to the methodology outlined in Myers-Pigg et al. (2015), using measured levoglucosan concentrations and a conversion factor based on the ratio of levoglucosan to lowtemperature Py-DOC potentially present at the sampling site. Briefly, transit times were calculated between sample locations and average fire locations in each watershed to account for degradation during transit. The experimentally-derived ratio of levoglucosan to Py-DOC (Norwood et al., 2013), corresponding to each calculated transit time, was used to convert measured levoglucosan concentrations into an estimate of low-temperature Py-DOC in the rivers (Myers-Pigg et al., 2015). Fire locations were assumed to be concentrated at ∼52–53◦N for the Yenisei River (Myers-Pigg et al., 2015; Ponomarev et al., 2016); the range of fire locations from 1980 to 2014 using fire data from the Canadian Forest Service (2016) were used for the Great Whale River (Figure S1). Error of fire location for each sampling point was propagated for the Py-DOC and Py-POC calculations (**Figure 3**).

To estimate the concentrations of low-temperature Py-POC in each river, mass-weighted distribution coefficients (Kds; **Table 2**) were calculated by the following equation:

$$\mathbf{K}\_d = \frac{[l \text{ in POM mg kg}^{-1}]}{[l \text{ in DOM mg L}^{-1}]} \tag{1}$$

where the concentrations of levoglucosan (l) is measured in the isolated particulate and dissolved matter from each river (**Figure 2**; **Table 2**). Low-temperature Py-POC concentrations were then calculated using K<sup>d</sup> values for levoglucosan calculated in Equation 1, substituting Py-DOC concentrations for l in

TABLE 1 | Average levoglucosan concentrations, low-temperature PyC concentrations, suspended particulate matter (SPM), and source signature ratios from free lignin phenols and anhydrosugars (syringyl/vanillyl: S/V and levoglucosan/mannosan: L/M) for the Yenisei River and Great Whale River.




Systems are ordered by watershed/basin size. Note the similarities of LogK<sup>d</sup> values of high and low-T PyC biomarkers in rivers, suggesting the export of the entire PyC continuum is linked. Data for partition coefficients is from Yunker et al. (1994, 2002); Ziolkowski and Druffel (2010); Coppola et al. (2014); Wagner et al. (2015) and this study (Table 1), and calculated using Equation (1).

DOM in Equation 1 and solving for the concentration of Py-POC in mg kg−<sup>1</sup> . These concentrations were then scaled to mg L <sup>−</sup><sup>1</sup> using total suspended solids (TSS; suspended particulate material, SPM) measured in mg L−<sup>1</sup> (**Table 1**).

Fluxes of Py-DOC and Py-POC were calculated by scaling Py-DOC and Py-POC concentrations to discharge in each river (**Figure 3**). The percentage of flux in Py-DOC vs. in Py-POC was calculated by dividing the Py-POC flux calculated at a specific time point by the sum of Py-DOC and Py-POC fluxes at the same time point, multiplied by 100%(**Table 3**).

Low-temperature PyC phase distribution in these systems were also compared to high-temperature PyC phase distribution in other systems (**Table 2**), using concentrations of low- and high-temperature PyC biomarkers to calculate distribution coefficients as in Equation 1. These distribution coefficients are a reflection of the environmental distribution of PyC and, therefore, reflect all factors influencing this phase distribution, including source, association with minerals, and particle physical properties (such as surface area, porosity, number of reactive sites, etc.).

## RESULTS

#### Arctic River

Levoglucosan concentrations in the dissolved and particulate phase increase with increasing discharge (**Figure 2**). Particulate organic carbon (POC) in the Yenisei River was 0.08 ± 0.02 kg OC kg−<sup>1</sup> throughout the sampling period. Monomeric lignin phenol ratios (syringyl/vanillyl: S/V) are 0.44 ± 0.21 in the Yenisei River (**Table 1**), and the ratios of levoglucosan to its isomer mannosan (L/M) are 4.2 ± 1.2 (**Table 1**).

The timing of peak export of low-temperature Py-DOC and Py-POC were decoupled in the Yenisei River throughout the freshet period (**Figure 3**). Low-temperature Py-DOC peaks during the rising limb of the hydrograph, while low-temperature Py-POC peaks during the recession limb for the Yenisei River (**Figure 3**). LogK<sup>d</sup> values, the measure of environmental phase distribution, for levoglucosan were nearly one order of magnitude lower during peak flow in the Yenisei River

TABLE 3 | Average yearly flux estimates for Py-DOC and Py-POC, and average % PyC in the particulate phase for the Great Whale River and the Yenisei River.


(Figure S2) than during base flow, indicating a higher proportion of total PyC in the dissolved phase during high flow regimes. LogK<sup>d</sup> values of levoglucosan are within the same range as estimated logK<sup>d</sup> values for higher temperature PyC markers (BPCA) in rivers (**Table 2** and references therein).

#### Sub-Arctic River

POC in the Great Whale River was 0.17 ± 0.08 kg OC kg−<sup>1</sup> throughout the sampling period. Monomeric lignin phenol ratios S/V are 0.29 ± 0.19 and L/M ratios are 3.2 ± 3.7 (**Table 1**). In the Great Whale River, the flux of levoglucosan and low-temperature Py-DOC increased with increasing discharge, while the flux of low-temperature Py-POC decreased with increasing discharge (**Figures 2**, **3**). LogK<sup>d</sup> values decreased with increasing discharge (Figure S2).

## DISCUSSION

#### Characterization of Organic Matter and Pyrogenic Carbon

Free monomeric lignin phenol ratios in both phases (syringyl/vanillyl: S/V), coupled with levoglucosan and its isomer, mannosan (L/M), are within the range of values reported for gymnosperm source inputs in charcoals, aerosols, and soils, suggesting that low-temperature PyC in the Yenisei River and the Great Whale River in both phases are influenced predominantly by gymnosperm fire fuels (**Table 1**, source ratios from Kuo et al., 2011a). These ratios are consistent with previously reported source reconstructions for Py-DOC in Arctic Rivers (Myers-Pigg et al., 2015).

Absolute concentrations of levoglucosan in the dissolved phase are on the lower end of the range previously reported in river DOM. For the Yenisei River, concentrations in this study are 0.06 ± 0.04 µg L−<sup>1</sup> vs. 0.8 ± 0.5 µg L−<sup>1</sup> from 2004 to 2006 (Myers-Pigg et al., 2015). Differences in the observed concentrations may be dependent on interannual variability in the mechanisms that govern export of PyC, such as hydrological flow regimes. However, further research must be done to ascertain the varying importance of the mechanisms that control PyC export to and through aquatic systems. Concentrations of levoglucosan in the particulate phase of the studied rivers are also low (**Figure 2**; **Table 2**, 0.73 ± 1.1 µg gdw−<sup>1</sup> ) compared to concentrations of levoglucosan in surface sediments from a river plume (1.3–6.9 µg gdw−<sup>1</sup> ) of a small, mountainous stream (Hunsinger et al., 2008). It should be noted that erosion, resuspension, and deposition of PyC along transport is particularly high in mountainous systems (Cotrufo et al., 2016). This may affect the total amount of PyC transported in the particulate phase to estuarine systems compared to the relatively larger river systems in this study, which also have relatively smaller proportions of total carbon exported in the particulate phase.

### Differences in Phase Distribution throughout the Hydrograph

In watersheds that contain permafrost coverage (such as the watersheds in this study), the permafrost can act as a boundary layer, causing rapid translocation of water to river systems during spring thaw events, which influences SOM transport to rivers. The depth of the soil active layer, in which organic matter has the ability to exchange with the surface terrestrial ecosystem, though generally quite shallow in these regions, tends to vary based on location within a basin. In a subwatershed of the Yenisei River, storage of PyC is largest in areas with permafrost and shallow active layers, whereas mineral soils with thick active layer or lacking permafrost stored less PyC, suggesting increased mobility of PyC from the latter (Guggenberger et al., 2008). The same study showed that most of the export of PyC occurs during the period of snowmelt with dominance of surface flow pointing to a strong relationship between PyC concentrations and hydrological regime. This may help to explain the observed lack of relationship between fire frequency and riverine export of dissolved black carbon (DBC as determined by BPCAs; Ding et al., 2013, 2014), and the strong reported relationships between DBC and hydrological cycling (Jaffe et al., 2013), particularly in Arctic systems (Stubbins et al., 2015).

Here, low-temperature Py-DOC peaks during the rising limb of the hydrograph in the Yenisei River (**Figure 3A**). This early peak of low-temperature Py-DOC is not mirrored in bulk DOC concentrations, which increase linearly with discharge (R <sup>2</sup> = 0.82, p < 0.001; graph not shown). This apparently "early" release of Py-DOC has not been previously observed, as Py-DOC from both high and low-temperature portions of the continuum have been previously correlated to discharge and DOC in these and other systems (Jaffe et al., 2013; Myers-Pigg et al., 2015; Stubbins et al., 2015). However, here we present the first dataset with targeted high resolution sampling during spring peak flow. Hence, the decoupling of Py-DOC and DOC during the rising limb may be a phenomenon not previously captured. During the rest of the hydrograph (low flow, peak flow, and the falling limb) low-temperature Py-DOC is correlated to bulk DOC (R <sup>2</sup> = 0.63, p < 0.001), as has been previously observed. Therefore, further high-resolution sampling is needed to understand Py-DOC dynamics during the early part of the spring freshet period in other years to determine if the periodic decoupling of Py-DOC and DOC is systematic or only an isolated case. This decoupling of low-temperature Py-DOC and DOC in the rising limb of the hydrograph may be linked to the mobilization of a different source of Py-DOC during this period than during other flow regimes.

Although very little research has been done on the transport of particulate PyC from river systems, it has been suggested that transport in this phase may be of importance in recently fire-affected ecosystems (Wagner et al., 2015), where increased surface run off (Moody et al., 2013), soil erosion (Shakesby and Doerr, 2006), and increases in TSS (Ryan et al., 2011) have been observed. In the study regions, particulate carbon export is often a small proportion of the total carbon export; POC export accounts for ∼15% of the total OC flux from the pan-Arctic per year (Holmes et al., 2012; McClelland et al., 2016). Around 23% of the total POC export occurs during the spring freshet period in both the Great Whale River and Yenisei River (Hudon et al., 1996; McClelland et al., 2016), making high-resolution sampling important to capture temporal variability in the Py-POC flux. As previous work estimates that Py-POC export in Arctic Rivers is between 3 and 9% of total POC fluxes (Elmquist et al., 2008), transport of Py-POC may be an important portion of the total POC in Arctic Rivers. Here, we find that low-temperature Py-POC comprises 4.5 ± 2.8% of the low-temperature PyC flux from the Yenisei River (**Figure 3**; **Table 3**). These relatively lower levels of export of low-temperature Py-POC in our systems has implications for the usage of levoglucosan to track fire activity in depositional records, as low-temperature PyC determined from levoglucosan is exported predominantly in the dissolved phase. In the Yenisei River, low-temperature Py-POC export peaks during the falling limb of the freshet hydrograph (**Figure 3C**), which may reflect varying sources of PyC between phases or variations in export mechanisms between the two phases. For example, the composition and bioavailability of OM exported from soils in high-latitude systems is a function of active layer depth and mineral soil association (Kawahigashi et al., 2006). Terrestrial organic carbon (as determined by n-alkanoic acids) in the particulate phase has been found to associate with two distinct pools of soil carbon released during the spring freshet period; a young pool dominated by humics and an old pool with a mineral soil component (Vonk et al., 2010). This observed difference in mobilization of terrestrial OC may also influence the mobilization of PyC in high-latitude systems, which in turn, affects availability for transformations during transport, as a large portion of labile materials, if associated with a mineral layer, may be protected from biodegradation during transport (Hedges et al., 1997; Kaiser and Guggenberger, 2000; Vonk et al., 2010). During the freshet period, low-temperature PyC is proportionally exported more in the dissolved phase than during low-flow (**Figure 3**; Figure S2). This may be due to hydrophilic DOM, such as low-temperature PyC (chemical functionality in **Figure 1**), bypassing association with soils within the active layer, and directly entering the aquatic system (Kawahigashi et al., 2006) during high-flow regimes. Because high-flow regimes release the largest proportion of yearly DOM and POM during a relatively short time period (Amon et al., 2012; Holmes et al., 2012; McClelland et al., 2016), these flow regimes are similarly expected to contribute substantially to the overall flux of PyC from highlatitude rivers. Additionally, since PyC is predominantly released in the dissolved phase, this may allow highly functionalized and hydrophilic portions of PyC to be more accessible to in-situ microbial degradation and lead to increased riverine respiration rates during this period (Masiello and Louchouarn, 2013).

In the Great Whale River, physical protection of PyC may regulate its release from soils. Only around 10% of the total DOC is released in this river system during the ice break-up period (Hudon et al., 1996), as opposed to almost half of the DOC released from the major Arctic Rivers during the spring freshet period (Holmes et al., 2012). This observation suggests that the timing and release of organic matter from small southern boreal watersheds are driven by different controls than those operating in the major Arctic Rivers. The flux of organic matter may be normalized in these latter systems by the size of the watersheds, whereas soil heterogeneities in smaller southern boreal watersheds may show a strong influence of soil processes. For example, in southern boreal forests, accumulation of large, long-lived stocks of PyC have been observed in mineral horizons with little to no soil profile development (Miesel et al., 2015). The opposite has been reported in boreal forests containing mineral soils underlying thick active layers (Guggenberger et al., 2008). The Great Whale River watershed is itself highly heterogeneous and contains a variety of soil types including discontinuous permafrost (Bhiry et al., 2011). However, and in conjunction with high fire return intervals, the export of PyC from this watershed may be more regulated by soil processes than hydrological flow regimes, as highlighted by the lack of a clear relationship of PyC export with hydrography (**Table 3**; **Figure 3**).

## Implications for the Entire Pyrogenic Carbon Continuum

The similarity between phase distribution calculated for hightemperature PyC components (BPCA and PAHs) and lowtemperature PyC (levoglucosan) suggest that low- and hightemperature PyC may be exported similarly from river systems (**Table 2**). However, as physical weathering has been shown to be an important loss process for both high- and low-temperature charcoals, although on very different time scales (Naisse et al., 2015), understanding the ages of these exported carbon pools is necessary before final conclusions on their similarities can be determined. Additionally, given the fact that bulk DOC and POC in Arctic River systems have vastly different ages (e.g., Goñi et al., 2005; Amon et al., 2012), we must understand the sources and ages of dissolved and particulate PyC in order to fully understand how export of Py-DOC and Py-POC might be linked both temporally and throughout the PyC continuum. For example, high-temperature PyC (as determined by the CTO-375 method; PyC range covered by the method in **Figure 1**) has been found to contain two distinct sources between the particulate and dissolved phases in river systems (Wang et al., 2016), and much of this high-temperature PyC can be derived from fossil fuels (up to 33% of the DBC and around 50% of the PBC). In contrast, the sources of low-temperature PyC (derived from biomass burning) using biomarker source ratios appear to be similar in our studied rivers (S/V and L/M; suggesting gymnosperm sources in both phases). Therefore, it may become important to characterize potential variability of PyC generation during biomass burning and its effects on PyC mobilization. Isotopic approaches may help identify some of this variability (e.g., isotopic fractionation of levoglucosan occurs during atmospheric processes; Sang et al., 2016).

The apparent order of magnitude increase in the ratio of Py-DOC to Py-POC for high-temperature PyC from rivers to oceans (**Table 2**; calculated from Ziolkowski and Druffel, 2010; Coppola et al., 2014; Wagner et al., 2015) suggests that Py-POC plays an increasingly important role in PyC cycling through transport from riverine to oceanic systems. This may be due to the potential degradation of dissolved PyC during transit (Stubbins et al., 2012; Masiello and Louchouarn, 2013; Ward et al., 2014; Myers-Pigg et al., 2015) and/or the importance sorption of Py-DOC to Py-POC in marine systems (Coppola et al., 2014). Sorption processes and transfer to sediments are considered as one of the major loss processes for Py-DOC in the world's ocean (Coppola et al., 2014). Py-DOC exported from Arctic Rivers could sorb to sinking particulates once it enters the Arctic Ocean. This has implications on how much Py-DOC may survive transport and end up in regions of North Atlantic Deep Water formation (Stubbins et al., 2015), vs. what is sorbed to sinking particles and buried on the vast Arctic continental shelves. Measures of Py-POC in Arctic Ocean sediments and sinking particles (using CTO-375 method) suggest that Py-POC quickly sinks and is stored within coastal margins (Fang et al., 2016). To understand loss terms and PyC cycling, the relative proportion of all component of the PyC continuum in the dissolved and particulate phases during transfer to and within ocean systems must be further explored.

## Flux Estimates from Rivers for Arctic Pyrogenic Carbon

The percentage of the total PyC flux from these rivers in the particulate phase varies dramatically throughout the hydrograph. During low-flow stages (<2 × 10<sup>4</sup> m<sup>3</sup> s −1 ), low-temperature Py-POC was 3.1 ± 2.3% of the total PyC flux in the Yenisei River, while during high flow low-temperature Py-POC was 5.1 ± 2.8% of this flux. This is much less than a calculated distribution of Py-POC in worldwide flux estimates (∼16% of total PyC flux from rivers; Santín et al., 2016). This is also lower than previous estimates of Py-POC fluxes from the pan-Arctic (∼8% of total PyC flux) using pan-Arctic Py-POC flux estimates calculated using BC/POC ratios from Elmquist et al. (2008) and pan-Arctic POC fluxes from McClelland et al. (2016) and pan-Arctic Py-DOC fluxes from Stubbins et al. (2015). However, as our estimates measure in-situ phase distribution based on biomarker composition within rivers, they can represent a conservative estimate of the amount of Py-POC exported from the studied Arctic Rivers. Additionally, previous Py-POC flux estimates from Arctic Rivers rely on measurements of the most refractory PyC (using the CTO-375 method; see **Figure 1** for range of detection within the PyC continuum) in estuarine sediments to estimate in-situ Py-POC fluxes (see Elmquist et al. (2008) for further explanation of previous Py-POC flux calculations). As refractory PyC analytical methods (such as CTO-375 method) may overestimate PyC contribution in environmental samples (Hammes et al., 2007), higher temperature PyC can be derived from both fossil fuel combustion and biomass burning, and the most refractory portion of the PyC continuum measured by these techniques may be more particle associated than other forms of PyC, the previously reported flux estimates may overestimate the actual distribution of PyC transported from Arctic Rivers. In environments where POC export is a proportionally important part of the TOC flux, the relative influence of Py-POC on the total export of PyC from the region may be of higher significance. For example, in the Great Whale River, total POC export accounts 18.9% of the total OC export (Hudon et al., 1996), slightly higher than the average Arctic River system (∼15%; Holmes et al., 2012; McClelland et al., 2016). In this river system, Py-POC was 8.9 ± 17.5% of the total PyC flux (**Figure 3**; **Table 3**). The export of low-temperature PyC varied with the hydrograph (**Figure 3**), implying that the export of PyC in the particulate phase varies throughout the year. This is important to consider when looking at PyC in deposited sedimentary records and estimating the impact of fire on past climate, as the records of these signatures may vary with hydrographic regimes and particulate matter loadings.

Considering the changing fire regimes in the Arctic (Kelly et al., 2015), the intrinsic tie of Py-DOC with discharge in the region (Myers-Pigg et al., 2015; Stubbins et al., 2015), and changing hydrological regimes (Wrona et al., 2016), the transport of PyC from Arctic terrestrial to aquatic systems is likely to increase. As the Arctic Ocean continental margins have recently been suggested as an efficient location for PyC burial (Fang et al., 2016), understanding the phase distribution of PyC during transport through rivers and estuaries is relevant for understanding overall PyC fluxes, the potential degradation of PyC during transit, and losses

#### REFERENCES


of PyC from the active carbon cycle through burial into sediments.

#### AUTHOR CONTRIBUTIONS

All authors contributed to experimental design for this study. RT and AM collected and processed the samples. AM and PL interpreted the data and prepared the manuscript. All authors participated in revisions and approved the final manuscript.

## FUNDING

This project was funded under the TOMCAR-Permafrost Marie Curie International Reintegration Grant FP7-PEOPLE-2010- RG (project reference: 277059) within the Seventh European Community Framework Programme awarded to RT (http:// www.tomcar.fr). Yenisei River daily discharge data was provided by the Arctic Great Rivers Observatory (NSF-1107774). Travel and living expenses were also funded thanks to GDRI Car-Wet-Sib II and INP-Toulouse SMI program.

## ACKNOWLEDGMENTS

We thank the Chateaubriand Fellowship Program led by the Office for Science and Technology of the Embassy of France in the United States, which provided funding for AM to spend 6 months at EcoLab in France. We thank Nikita Tananaev for his assistance while in Igarka, including sampling, coordination and use of field facilities. The Melnikov Permafrost Institute is thanked for their hospitality and support of this research. Sample collection and transport could not have been possible without the hard work of countless amazing people, including those in Toulouse at EcoLab, the team at ULISSE, Elena Fedorova, the entire Geocryology Laboratory team in Igarka, and particularly Anatolii Pimov. We thank R. Amon, K. Kaiser, and M. Norwood for helpful comments on earlier versions of this manuscript. Figure S1 was created with assistance from M. Norwood. We also thank two reviewers for their constructive and helpful comments on this manuscript. Open access publication fees were supported in part by the Gordon and Betty Moore Foundation.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2017.00038/full#supplementary-material

Arctic Rivers. Geochim. Cosmochim. Acta 94, 217–237. doi: 10.1016/j.gca.2012. 07.015


of multidisciplinary research by Centre d'études Nordiques (CEN). Ecoscience 18, 182–203. doi: 10.2980/18-3-3469


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Myers-Pigg, Louchouarn and Teisserenc. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Impact of Wetland Decline on Decreasing Dissolved Organic Carbon Concentrations along the Mississippi River Continuum

Shuiwang Duan<sup>1</sup> \*, Yuxiang He<sup>2</sup> , Sujay S. Kaushal <sup>2</sup> , Thomas S. Bianchi <sup>3</sup> , Nicholas D. Ward3, 4 and Laodong Guo<sup>5</sup>

<sup>1</sup> Department of Geology and Earth System Science Interdisciplinary Center, University of Maryland, College Park, MD, USA, <sup>2</sup> National Water Center, National Weather Service, National Oceanic and Atmospheric Administration, Silver Spring, MD, USA, <sup>3</sup> Department of Geological Sciences, University of Florida, Gainesville, FL, USA, <sup>4</sup> Whitney Laboratory for Marine Bioscience, University of Florida, St. Augustine, FL, USA, <sup>5</sup> School of Freshwater Sciences, University of Wisconsin-Milwaukee, Milwaukee, WI, USA

#### Edited by:

Javier Arístegui, University of Las Palmas de Gran Canaria, Spain

#### Reviewed by:

X. Antón Álvarez-Salgado, Spanish National Research Council, Spain Jie Xu, South China Sea Institute of Oceanology (CAS), China

> \*Correspondence: Shuiwang Duan sduan@umd.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 26 September 2016 Accepted: 13 December 2016 Published: 09 January 2017

#### Citation:

Duan S, He Y, Kaushal SS, Bianchi TS, Ward ND and Guo L (2017) Impact of Wetland Decline on Decreasing Dissolved Organic Carbon Concentrations along the Mississippi River Continuum. Front. Mar. Sci. 3:280. doi: 10.3389/fmars.2016.00280

Prior to discharging to the ocean, large rivers constantly receive inputs of dissolved organic carbon (DOC) from tributaries or fringing floodplains and lose DOC via continuous in situ processing along distances that span thousands of kilometers. Current concepts predicting longitudinal changes in DOC mainly focus on in situ processing or exchange with fringing floodplain wetlands, while effects of heterogeneous watershed characteristics are generally ignored. We analyzed results from a 17-year time-series of DOC measurements made at seven sites and three expeditions along the entire Mississippi River main channel with DOC measurements made every 17 km. The results show a clear downstream decrease in DOC concentrations that was consistent throughout the entire study period. Downstream DOC decreases were primarily (∼63–71%) a result of constant dilutions by low-DOC tributary water controlled by watershed wetland distribution, while in situ processing played a secondary role. We estimate that from 1780 to 1980 wetland loss due to land-use alterations caused a ca. 58% decrease in in DOC concentrations in the tributaries of the Mississippi River. DOC reductions caused by watershed wetland loss likely impacted the capacity for the river to effectively remove nitrogen via denitrification, which can further exacerbate coastal hypoxia. These findings highlight the importance of watershed wetlands in regulating DOC longitudinally along the headland to ocean continuum of major rivers.

Keywords: dissolved organic carbon, wetlands, Mississippi River Basin, river continuum, in-stream processes, land use-land cover

#### INTRODUCTION

Draining an area of ∼28% of the Earth's surface, the 25 largest rivers in terms of discharge are responsible for over 40% of the global flux of DOC from rivers to the oceans (McKee, 2003). Within major river systems, DOC also plays an important role in driving microbiallymediated denitrification (Philip and Townsend, 2010) and greenhouse gas emissions (Raymond et al., 2013), and affecting the transport and toxicity of metals and trace organic contaminants (Aiken et al., 2011). A limited understanding of the fundamental processes controlling spatiotemporal variability of river DOC at the basin-scale restricts our ability to accurately predict the influence of future scenarios (e.g., land use-land cover and climate change) on fluvial DOC levels. Current concepts predicting mechanisms for DOC variations from the headwaters to the mouth of major rivers primarily focus on in situ processing (Vannote et al., 1980) and hydrologic exchange with fringing floodplain wetlands (Sedell et al., 1989). In situ processing causes DOC loss as a result of microbial degradation coupled with photochemical oxidation (Amon and Benner, 1997; Ward et al., 2013) and sorption onto minerals (Aufdenkampe et al., 2001), whereas autochthonous inputs to riverine DOC are generally more labile and thus largely recycled (Stanley et al., 2012). Fringing floodplains can considerably augment river DOC levels (Tockner et al., 1999; Hedges et al., 2000) owing to high levels of both algal and aquatic plant production. Relatively speaking, the effects of heterogeneous watershed characteristics have been generally ignored in river continuum approaches characterizing riverine DOC dynamics (Vannote et al., 1980; Sedell et al., 1989). This is particularly surprising when considering it is well known that DOC concentrations vary with climate zone (Schlesinger and Melack, 1981) and land use -land cover (LULC) (Butman et al., 2015). Wetland coverage is often the best predictor of DOC concentrations in rivers and streams (Gergel et al., 1999; D'Amore et al., 2016). However, LULC (e.g., wetland distribution) is seldom used to interpret DOC changes in large river systems. Models and theoretical concepts (e.g., SPARROW Shih et al., 2010 and the urban watershed continuum Kaushal and Belt, 2012) have been developed to link river water quality to watershed characteristics, but relative contributions of in situ vs. watershed controls on riverine DOC are rarely quantified.

Controls on river DOC may change in watersheds that are highly impacted by complex human alterations, such as North America's largest river, the Mississippi River (MR; **Figure 1**). For example, Butman et al. (2015) determined that 8% of the DOC present in rivers today is radiocarbon-depleted (i.e., millennial aged) organic carbon that has been mobilized as a result of human disturbances. The MR has been greatly altered during the last few centuries by: (1) large-scale conversion of watershed wetlands to farmlands (Dahl and Johnson, 1991), (2) construction of high and low dams in primary tributaries, and (3) an extensive network of flood-control levees (Wiener et al., 1996). Dam constructions generally increase the in situ processing time of riverine DOC, while artificial levees and large-scale conversion of watershed land use may affect DOC inputs from wetlands in floodplain and watershed. Prior studies in the Mississippi River have reported a downstream trend of decreasing DOC concentrations during three seasons from 1991 to 1992 (Leenheer et al., 1995). Based on the sparse temporal variability in the above study, it is unclear whether this trend is a consistent phenomenon in the MR (Duan et al., 2007, 2014). Such observations of a downstream decrease is different from other large river systems, where DOC concentrations increase or there are no apparent changes along the downstream gradient (Lara et al., 1998; Hedges et al., 2000; Wu et al., 2007). Reasons for these differences among the rivers (e.g., differential watershed characteristics vs. fundamental differences in in situ processing mechanisms) remain unclear.

In this study, DOC data from three intensive samplings along the Mississippi River were reanalyzed along with a 17 year time-series of DOC measurements starting in 1997 in an effort to unravel the processes regulating longitudinal changes in DOC concentrations in the MR. Further, we compare our results to longitudinal DOC changes in the world's major rivers to determine the importance of unique watershed characteristics. A combined mass balance and regression approach was used to estimate the relative contributions from in situ processing vs. tributary inputs, which are most likely controlled by watershed land use and climate. We hypothesize that watershed wetland distribution controls DOC levels of tributaries and DOC longitudinal changes along the Mississippi River. Decreases in DOC as a result of wetland loss may have large implications on the role that wetlands play in controlling DOC abundance, denitrification, and nitrogen fluxes from major rivers to coastal waters, such as the Gulf of Mexico (Schramm et al., 2009).

## MATERIALS AND METHODS

## Water Quality Data

Past measurements of DOC and temperature made during three intensive samplings along the MR main channel from 1991 to 1992, which were used here for comparison and reanalysis, were collected from U.S. Geological Survey Open-File Report 94-4191 (Leenheer et al., 1995) and Open-File Report 94-523 (Moody, 1995),respectively. These expeditions were conducted in summer 1991 (June 23 to July 1), fall 1991 (September 25 to October 4), and spring 1992 (March 25 to April 4). Samples were taken near the center of the river approximately every 17 km from downriver to upriver between New Orleans and Minneapolis. DOC concentrations were also measured in the 17 primary tributaries up to six times (twice for the 3 seasons). Water samples were collected from just below water surface and DOC was analyzed using a high-temperature catalytic oxidation method (Leenheer et al., 1995).

Long-term water quality data were collected from the U.S. Geological Survey (USGS) National Stream Quality Accounting Network (NASQAN) webpage (http://water.usgs.gov/nasqan/). The data were obtained for all available sites on the Mississippi River main stem (Clinton, Grafton, Thebes, Vicksburg, St. Francisville, Baton Rouge, and Belle Chasse; **Figure 1**) and its primary tributaries (Iowa River at Wapello, Des Moines River at Keosauqua, Illinois River at Valley City, the Missouri River at Hermann, the Ohio River at Dam 53 near Grand Chain, the Arkansas River at David D Terry Lock and Dam below Little Rock, and the Yazoo River below Steele Bayou near Long Lake). Data at most sites covered the whole period of water years of 1997 to 2013 except the Des Moines River (2009–2013), and the LMR at Vicksburg (2008–2013), Baton Rouge (2008–2013), and Belle Chasse (2007–2013). Water sampling frequency varied but generally ranged from 6 up to 24 times per year; the most frequent collections occurred during spring high flow period. Methods for sample collections and measurements of

FIGURE 1 | Map for U.S. Geological Survey monitoring stations on the Mississippi River main stem (large dots) and the primary tributaries (small dots), and distribution of wetland (in green) in the Mississippi River watershed. Thebes was the dividing point of the upper and lower Mississippi River (UMR and LMR).

water temperatures and DOC are available at Wilde (2006) and Brenton and Arnett (1993)—the same as those used in Leenheer et al. (1995).

## Estimating Contributions of Tributaries Inputs vs. In situ Processing

Long-term USGS DOC concentration data (with water discharges) were used to separate contributions of tributary inputs from that of in situ processing to the longitudinal DOC decreases in the Upper Mississippi River (UMR, from Clinton to Thebes) and the Lower Mississippi River (LMR, from Thebes to St. Francisville) based on DOC mass balances. Mean daily water discharge data were assembled from the records of the U.S. Army Corps of Engineers (USACE; http://www.usace.army.mil/) or the USGS (http://waterdata.usgs.gov/usa/nwis/). In each river section (UMR or LMR), we used conservative mixing of tributaries with main stem water to estimate DOC concentrations at the downriver station; fractions explained and unexplained by conservative mixing were tributary inputs and in-situ transformation, respectively. That is, contributions from tributary inputs were the difference between in situ DOC measurements at the upstream station and DOC estimates for the downriver station. Meanwhile, contributions of in situ transformation were the difference between the DOC estimates and measurements at the downstream station. Positive values of the contributions from tributary inputs and in situ processing suggest dilutions by tributaries and in situ loss, while negative values imply increases in concentrations by tributary inputs and in situ production/floodplain inputs.

In order to estimate DOC concentrations of the UMR and LMR at the downstream stations, daily DOC fluxes of each input station (upstream main stem or tributaries) and output station were initially estimated from discrete DOC concentration measurements and continuous average daily flow using the USGS LOADEST method (http://water.usgs.gov/software/loadest/). LOADEST estimates constituent loads in streams and rivers by developing a regression model, given a time series of streamflow, constituent concentrations, and additional data inputs. The model is well documented in the published literature and is accepted as a valid means of calculating annual solute loads from a limited number of water quality measurements. We used measurements of DOC concentrations and streamflow during the same dates for model calibrations. The models, which contained up to seven parameters, captured the dependence of concentrations on discharge and season and any long-term trends. Daily discharge data for the study period were prepared for LOADEST models so that daily fluxes of DOC were estimated for the whole study period, once the model had been calibrated. From the returning files, results from the adjusted maximum likelihood estimates (AMLE) were used to modify loading equations to correct for transformation bias. Additionally, a period-weighted approach<sup>28</sup> was used to verify above LOADEST method. In this approach, a measured DOC concentration was assumed to represent a period around which the sample was collected; this allowed for connecting measured concentrations through time by a piecewise linear function.

The estimated DOC daily loads were used for DOC mass balances for the UMR and LMR before a mean travel time from each input station to the output station was known. The travel time was obtained by finding the best correlations between daily discharge data at an input and an output station (Duan et al., 2010). More specifically, daily water discharges of the two stations were used for correlation analysis, with data of the input station moved forward a few days until the highest significance level of the correlation was achieved. The number of days that data needed to be adjusted was taken as the mean residence time between the two stations. This approach for travel time was verified by comparing water discharge inputs and output of each river section shown as in Duan et al. (2010). Once residence time from each input site was known, the daily water discharges and DOC loads of the input stations were re-assembled by moving them forward by the number of days that represented their mean travel times to the output station.

Mass balances of water discharges were initially conducted, and the unbalanced discharges were assumed to come from ungauged small tributaries. These discharges were then multiplied by mean daily DOC concentrations of ungauged tributaries (assumed to be averages of the selected tributaries) to estimate daily DOC inputs from these tributaries. Finally, total inputs of daily water discharges and DOC loads (gauged and ungauged) were then summarized, and daily DOC concentrations at the downriver stations were obtained by dividing total daily DOC inputs by total daily water inputs.

## Watershed Land Use Land Cover and Climate Data

To identify controls of watershed characteristics on DOC concentrations of the primary tributaries, correlation analyses were conducted between mean DOC concentrations of the tributaries and their watershed LULC and climatic parameters (air temperature, precipitation and runoff ratio). Relationships were tested using Pearson's correlations (p < 0.05, Pearson's correlation). DOC data collected during the three expeditions of 1991–1992 were used, because more tributaries (17 in total) were sampled during these expeditions than the USGS sites with longterm DOC data (only seven) (**Figure 1**). For most tributaries, the data were representative because DOC was measured up to six times a year across seasons. National Land Cover Dataset 1992 (NLCD 1992; http://www.epa.gov/mrlc/nlcd.html) was used for land use data collection in order to match the time when DOC concentrations were measured. NLCD 1992 data were a 21-class land cover classification scheme applied consistently over the United States.

Mean air temperature and precipitation during a close period (1990–1995) were obtained from the NCEP North American Regional Reanalysis (NARR) dataset and the forcing data for the North Land Assimilation System (NLDAS). The near-surface mean air temperature was adjusted vertically to account for terrain height difference between NARR and NLDAS. Precipitation was a product of temporal disaggregation applied directly on the NLDAS grid based on gauge-only CPC daily precipitation analysis, and orographically-adjusted according to the widely used PRISM climatology.

Sub-watersheds of the 17 primary tributaries were divided from the Mississippi River basin. Boundary polygons of each watershed were generated based on USGS 8-digit hydrologic unit code (HUC). The LULC, annual mean air temperature and precipitation of each sub-watershed were extracted from data of the whole Mississippi watershed according to its boundary polygon using the ArcGIS spatial analyst tools. Annual runoff of each sub-watershed was calculated by dividing annual water discharge with watershed drainage area, and the annual runoff was then divided by annual precipitation to estimate runoff ratio (runoff / precipitation × 100%); data of water discharge and drainage area were from http://waterdata.usgs.gov/usa/nwis/.

## RESULTS

### Longitudinal Changes in DOC Concentrations along the MR

Results showed that the downstream decrease in DOC concentrations along the main stem was consistent throughout the study period. High-resolution spatial data of 1991–1992 (**Figures 2A–C**) demonstrated that DOC concentrations decreased gradually, except for two rapid drops shortly after the confluence of major tributaries (Missouri and Ohio Rivers). DOC concentrations of major tributaries also displayed a longitudinal decreasing trend (**Figures 2A–C**; r 2 ranging from 0.52 to 0.74, p < 0.05). To be consistent with data of 1991–1992, long-term USGS DOC data (1997–2013) were presented seasonally. These new records showed that DOC concentrations in the main stem consistently decreased downstream of MR headwaters across seasons throughout the study period (**Figures 2D–G**; r 2 ranging from 0.87 to 0.96, depending on seasons, p < 0.05). The coefficients of variation were minor (C.V. = 0.4–10%) showing strong confidence.

Interestingly, the decreases in DOC concentrations were generally accompanied by downstream increases in water temperatures across seasons (one exception in **Figure 2A**) (**Figures 2B–G**; r 2 ranging from 0.81 to 0.98, p < 0.05). As a result, DOC concentrations and water temperatures were negatively correlated (r 2 ranging from 0.80 to 0.94, p < 0.05; one except of June-July 1991), but the slope changed with seasons suggesting complex temperature effects (**Figure 3**).

#### (p < 0.05). 1991–1992 measurements were obtained from the U.S. Geological Survey Open-File Report 94–4191 and Open-File Report 94–523.

## Contributions of Tributary Inputs and In situ Processing to DOC Reductions

The contributions of tributary inputs and in situ processing to downstream DOC decreases were positive across all months (**Figure 4**), indicating that the downstream reductions of DOC concentrations were a combined effect of tributary dilution (by low-DOC water) and in situ loss. The contributions from tributary dilutions were positively correlated with water discharge in the LMR (**Figure 4B**; r <sup>2</sup> = 0.81, p < 0.01), as well as in the UMR but the correlation was less significant (**Figure 4A**; two outliers of March and June; r <sup>2</sup> = 0.63, p < 0.05). The contributions of in situ processing also exhibited pronounced seasonal variations. In the LMR section, this seasonal pattern strictly followed that of water temperatures (**Figure 4D**; r 2 = 0.88, p < 0.05), but it was positively correlated with both water temperature and water discharge in the UMR (**Figure 4C**; r 2 = 0.36 and 0.55, p < 0.05).

Summing for the entire study period, we found that the effects of tributary dilution accounted for 71 and 63% of the DOC deceases in the UMR and

LMR, respectively. The effects of tributary dilution were 2.4 times and 1.7 times that of in situ processing losses.

#### Control of Watershed Wetland Distribution on DOC Decreases

Patterns of DOC spatial variability across the Mississippi tributaries tightly followed that of watershed wetland coverage, with both parameters showing a downstream decrease from north-to- south (**Figure 5A**). In particular, DOC concentrations of the 17 primary tributaries were positively correlated with watershed wetland coverage (**Figure 5D**; r <sup>2</sup> = 0.88, p < 0.05), whereas no significant correlation was found with any other type of land-use regime. Although tributary DOC concentrations also varied inversely with temperature, precipitation and runoff ratio (**Figures 5B,C**), the correlation coefficients were much lower (r <sup>2</sup> = 0.57, 0.23, and 0.03, respectively). Moreover, we found watershed wetland coverage (wetland %) can be regressed negatively with mean air temperature (Temp) and runoff ratio (runoff%), but positively with annual precipitation (Pptn) with an r <sup>2</sup> of 0.80 (Equation 1 and **Figure 5E**):

**wetland** % = **17.03** − **2.10 Temp** ( ◦**C**) <sup>+</sup> **0.031Pptn (mm)** − **0.506Runoff** % (1)

Where the standard errors of the estimates of the intercept and the three coefficients of the equation were 2.67, 0.32, 0.007, and 0.111, respectively.

## DISCUSSION

### Consistent Longitudinal Pattern of DOC Concentrations along the Mississippi River

Our data suggest that the longitudinal decreases in DOC concentrations along the Mississippi River showed a consistent pattern. The downstream decreases were not only observed in the three high-resolution spatial data of 1991–1992 (**Figures 2A–C**), but also the long-term USGS DOC data (1997–2013) with intensive measurements (biweekly to monthly) (**Figures 2D–G**). Reviews on all available longitudinal DOC data of major rivers of the world (**Table 1**) found that the consistent downstream

decreases in DOC concentrations along the MR were globally unique. Most rivers showed no obvious gradients or features (e.g., Lena; Lara et al., 1998, Columbia; Prahl et al., 1998, Paraná; Depetris and Kempe, 1993, and Yellow; Zhang et al., 2013), while other rivers displayed an increase or decrease only in a river section (e.g., Yangtze; Wu et al., 2007; Zhang et al., 2014 and Colorado; Miller, 2012). Consistent downstream increases occurred in the Amazon River (Hedges et al., 2000) and the St. Lawrence below the Great Lakes (Massicotte and Frenette, 2011). The unique longitudinal DOC pattern along the MR suggests that either fundamental in situ process are differentially expressed in the MR, or that watershed characteristics are the primary control on longitudinal DOC changes.

The UMR is one of the three primary tributaries of the Mississippi River (the other are the Missouri, and the Ohio Rivers), and it represents neither the longest tributary nor the tributary carrying the most discharge. It seems that choosing either of the other tributaries as the origin of the LMR would change this story, since both are well known to have lower DOC concentrations than the upper MR. However, the UMR was chosen as the source, like a few other rivers (e.g., the Potomac River, USA), because it flows in the same direction as the main stem and thus crosses more climate zones and/or landcover regimes. Therefore, the consistent downstream changes in DOC concentrations along the MR could be attributed to changes in climate zones and land-cover regimes reflecting DOC concentrations of tributaries.

### Tributary Inputs as a Dominant Control on the DOC Longitudinal Changes

This study used a mass balance approach for isolating physical (e.g., dilution) and biogeochemical (e.g., bio/photodegradation) controls on DOC concentrations along the MR. To complete the mass balance it was necessary to assume that DOC concentrations of unsampled small tributaries (including inputs from floodplains) were equivalent to the average concentration observed in primary tributaries of the same region. This assumption was partially satisfied. First, despite large uncertainty, there was a regional pattern of DOC concentrations, i.e., higher in UMR watershed and lower in LMR watershed (**Figure 5A**). Moreover, although DOC inputs from floodplains can be substantial in natural lowland rivers (Mulholland and Kuenzler, 1979), the impact of floodplains should be much less because the Mississippi River (especially the LMR) was leveed and >90% of the floodplain was isolated from the river (Wiener et al., 1996). Our mass balance approach was also verified with seasonal data (**Figure 4**). For example, we found that contributions of in situ transformation were positively correlated with water temperature (**Figures 4C,D**). This relationship corresponds with the observed temperature dependence of both bacterial decomposition and


photochemical oxidation (Davidson and Janssens, 2006; Porcal et al., 2015)—the two major pathways for in situ DOC loss in aquatic systems (Cole et al., 2007). The lower coefficient of correlation between DOC concentration and temperature in the UMR (4c) was probably due to additional DOC loss pathways, such as sorption onto particles (Aufdenkampe et al., 2001), DOC flocculation (Benedetti et al., 2003) and exchanges with floodplains (Tockner et al., 1999). Meanwhile, we also found that contributions from tributary dilutions on DOC decreases were all positively correlated with water discharge in both river sections (**Figures 4A,B**). This relationship matched well with the fact that DOC concentrations and water discharge are generally positively correlated in most tributaries of the Mississippi watershed (Irena et al., 2015).

Using the above mass balance approach, our results showed that tributary inputs are the primary control on riverine DOC longitudinal changes, while in situ processing play a secondary role. It was estimated that the effects of tributary dilution accounted for 71and 63% of the DOC deceases in the UMR and LMR, respectively, which is 2.4 times and 1.7

times that of in situ transformation losses, respectively. It is possible that tributaries dilute DOC concentrations in the MR because of the same longitudinal reduction phenomenon present within the tributaries themselves on a smaller scale, e.g., DOC concentrations decrease as the tributary flows toward the main channel (**Figures 2A,C**). On the other hand, although DOC concentrations and water temperature were negatively correlated downstream the river (**Figure 3**), this relationship did not support that the downriver decreases in DOC concentration was due to increasing in situ processing—that could be enhanced with higher temperatures at the downriver sites. This negative relationship did occur all the time (e.g., June-July 1991), and temperatures and DOC were not correlated if all-year data were included (**Figure 3**). In fact, when mean DOC in situ loss rates were estimated from data shown in **Figure 4** (normalized to river length), we found that the in situ loss rates decreased from 0.93µg L−1km−<sup>1</sup> in the colder UMR to 0.43µg L−1km−<sup>1</sup> in the warmer LMR, in contrast to the temperature gradient. Thus, we hypothesize that temperature-dependent in situ processing were not the primary control on the downstream distribution of DOC in the Mississippi River.

Recent studies highlight the importance of in situ processing of DOC and CO<sup>2</sup> emissions from many major river systems (Richey et al., 2002; Raymond et al., 2013). For example, it is reported that CO<sup>2</sup> emissions from the Amazon River are an order of magnitude higher than fluvial DOC flux (Raymond et al., 2013). This finding, however, does not conflict with the results of this study. That is, although DOC in situ processing can be substantial, it may only modify the magnitude of DOC spatial variability but does not alter DOC longitudinal patterns along a river network, per se. Otherwise, one would expect that DOC concentrations would longitudinally decrease for most large rivers worldwide, which is not observed. Therefore, longitudinal DOC variability in major river systems are likely highly dependent on the structure of tributary networks and the LULC regime of the surrounding watershed.

## Wetland Distribution Effects on River DOC Longitudinal Patterns

Data from this study showed that tributary DOC concentrations were tightly correlated with watershed wetland coverages, both of which decreased from headwater watersheds to the watersheds of the LMR (**Figure 5A**). Tributary DOC concentrations also varied inversely with temperature, precipitation and runoff ratio with less significance (**Figures 5B,C**), but it was more likely that these variables were controls for the formation of wetlands (Equation 1). That is, the occurrences of large areas of wetlands in the headwater sub-basins probably attributed to topographical depressions, cool climates, and abundant moisture in this region. Moreover, the negative relationship between wetland extent and air temperature also explains the complex relationship between DOC concentrations and water temperature, which varied seasonally (**Figure 3**). Therefore, wetland distribution (occurring mainly in headwaters) is likely the primary control on tributary DOC concentrations and downstream DOC decreases along the main stem of the MR. The tight coupling of tributary DOC concentrations with watershed wetland coverage observed here was also consistent with prior reviews showing occurrence of maximal DOC concentrations in wetlands (Schlesinger and Melack, 1981; Spitzy and Leenheer, 1991). This is also consistent with findings showing that wetland coverage was the best predictor of DOC concentration in rivers and streams (Mulholland and Kuenzler, 1979; Massicotte and Frenette, 2011).

With this relationship determined, wetland distribution can be used to interpret longitudinal patterns of DOC concentrations along major world rivers. For example, in more naturally flowing systems that have higher connectivity to floodplain wetlands, like the Amazon River, DOC increases downstream concurrent with high rates of in situ processing as a result of floodplain wetland inputs to the river (Hedges et al., 2000; Ward et al., 2015). Wetlands also play an important role in driving CO<sup>2</sup> production in large rivers (Abril et al., 2014; Borges et al., 2015) due to aquatic plant respiration and enhanced in situ processing due to both breakdown of wetland-derived OC and processes, such as priming effects, which can stimulate the breakdown of terrestrially-derived OC in the presence of fresh autochthonous OC (Guenet et al., 2014; Bianchi et al., 2015; Ward et al., 2016). In contrast, in rivers where wetlands are primarily located in headwater sub-watersheds and floodplain wetland extent is limited (as a result of human development; Wiener et al., 1996), such as the MR (**Figure 1**), DOC concentrations decrease gradually from headwaters to the river mouth due to tributary dilution. Therefore, the effects of wetland distribution should be highlighted when we examine the controls on longitudinal changes in DOC concentrations along large rivers with respect to global and regional carbon budgets.

## Historical Wetland Loss and Effect on Mississippi Watershed C and N

A significant fraction of wetland extent in the UMR has been lost during the last few centuries (1780–1980) due to large-scale conversion of wetlands to farmlands in the Ohio/MR Valleys and dam constructions in the UMR for navigation (Shih et al., 2010). In the LMR, a large fraction of floodplain wetlands have also been replaced with arable land, and human-made levees have resulted in a reduction of about 90% of the local inputs from the remaining floodplain wetlands in the region (Wiener et al., 1996). Based on EPA estimates of U.S. wetland extent in 1780 and 1980 for each state (Dahl, 1990), we calculated wetland losses for each tributary using ARC GIS. In total, wetland coverage of the MR watershed decreased from 8.8 to 3.1%, showing a wetland loss as high as 65% during this period. As previously discussed, such a drastic alteration to wetland extent likely played a key role in shaping the amount and composition of DOC that was exported from the watershed over the last several centuries, and helped form the downstream trends in DOC variability observed here in the present day. For example, based on the regression of tributary DOC concentrations with watershed wetland coverage (**Figure 5D**), we estimate that mean DOC concentrations of the MR tributaries decreased from 6.35 mg L−<sup>1</sup> in 1780 to 2.66 mg L−<sup>1</sup> in 1980 on average, having decreased by 58% or up to 3.69 mg L−<sup>1</sup> . Of course, there are many other factors in the basin that also affect carbon dynamics in the MR, including agricultural and urban development and channel modifications (e.g., dam constructions) throughout the UMR (Wiener et al., 1996). However, our data show agricultural or urban land cover was not correlated with river DOC concentrations. Dam construction may increase residence time for DOC in situ processing, but the effect of damming of riverine DOC is generally complicated (Wehr et al., 1997; Nadon et al., 2015) and thus less effective relative to watershed wetland loss.

Historical wetland loss and the associated DOC reductions may have further implications for the regional cycles of nutrients (nitrogen, phosphorus and silicon; Struyf and Conley, 2009). Reductions in wetlands would have local effects (e.g., nutrient sinks) as well as downstream effects from DOC export. For local effects, nitrogen, phosphorus and dissolved silicon can be retained via biological uptake or sedimentation, and high DOC exerts an important control on the occurrence of denitrification (the major process for nitrogen removal) by providing an energy source and anaerobic conditions (Philip and Townsend, 2010). Large amounts of DOC exported from wetland may cause anaerobic conditions in downstream rivers and nitrogen removal via denitrification. Thus, historical wetland loss in the UMR watershed and the LMR floodplain may have greatly reduced the capacity of this river system to retain nitrogen and other nutrients. Consequently, high nutrient yields from the croplands of MR watershed due to increased fertilizer application could be further enhanced by the wetland loss to some extent (Mitsch et al., 2005; Schramm et al., 2009). This finding has major implications for the implementation of restoration strategies to reduce nutrient fluxes from the MR to the Gulf of Mexico, which unfortunately continues to be ignored (Rabotyagov et al., 2014). In some cases these inundated wetland environments act as important sinks for river nitrogen, and their loss through land-use change can exacerbate hypoxia in coastal waters.

#### FUTURE PERSPECTIVES

Consistent long-term monitoring of aquatic systems, along with mechanistic modeling is necessary for constraining how

#### REFERENCES


earth systems will react to changing conditions in the future (e.g., climate, land use land cover, and population density). While detailed mechanisms can be difficult to disentangle in large river systems, the basin-integrated signal present in the main channel of a large river allows the use of simple models, such as those presented here to elucidate large scale, broadly defined processes (e.g., in situ processing vs. tributary inputs) across large spatial scales (e.g., continental). Constraining the mechanisms that drive biogeochemical fluxes from the land to the sea on a conceptual and quantitative basis is essential to understanding how natural systems interact, the influence that human activities have on these processes, and how these processes have/will change in the future.

#### AUTHOR CONTRIBUTIONS

SD, YH, and SK conceived of the modeling approach. SD compiled the DOC data from the literature and performed the model analyses. SD, TB, NW, and LG performed comparisons of the Mississippi River to other large rivers worldwide. All authors contributed to the writing and critical evaluation of the manuscript.

#### FUNDING

This research was supported by NASA grant NASA NNX11AM28G, NSF Awards EAR 1521224, DBI 0640300, CBET 1058502, and EAR-1426844. The Jon and Beverly Thompson Endowed Chair of Geological Science at the University of Florida.

#### ACKNOWLEDGMENTS

Stuart Findlay provided helpful comments on an earlier version of the manuscript. We thank the USGS for the historic data evaluated here. Open access publication fees were supported in part by the Gordon and Betty Moore Foundation.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Duan, He, Kaushal, Bianchi, Ward and Guo. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Microbially-Mediated Transformations of Estuarine Dissolved Organic Matter

Patricia M. Medeiros <sup>1</sup> \*, Michael Seidel 1, 2, Scott M. Gifford<sup>3</sup> , Ford Ballantyne<sup>4</sup> , Thorsten Dittmar <sup>2</sup> , William B. Whitman<sup>5</sup> and Mary Ann Moran<sup>1</sup>

*<sup>1</sup> Department of Marine Sciences, University of Georgia, Athens, GA, USA, <sup>2</sup> Research Group for Marine Geochemistry (ICBM-MPI Bridging Group), Institute for Chemistry and Biology of the Marine Environment, Carl von Ossietzky University of Oldenburg, Oldenburg, Germany, <sup>3</sup> Department of Marine Sciences, University of North Carolina, Chapel Hill, NC, USA, <sup>4</sup> Odum School of Ecology, University of Georgia, Athens, GA, USA, <sup>5</sup> Department of Microbiology, University of Georgia, Athens, GA, USA*

Microbially-mediated transformations of dissolved organic matter (DOM) in a marsh-dominated estuarine system were investigated at the molecular level using ultrahigh resolution mass spectrometry. In addition to observing spatial and temporal variability in DOM sources in the estuary, multiple incubations with endogenous microorganisms identified the influence of DOM composition on biodegradation. A clear microbial preference for degradation of compounds associated with marine DOM relative to those of terrestrial origin was observed, resulting in an overall shift of the remaining DOM toward a stronger terrigenous signature. During short, 1-day long incubations of samples rich in marine DOM, the molecular formulae that were enriched had slightly smaller mass (20–30 Da) and number of carbon atoms compared to the molecular formulae that were depleted. Over longer time scales (70 days), the mean differences in molecular mass between formulae that were depleted and enriched were substantially larger (∼270 Da). The differences in elemental composition over daily time scales were consistent with transformations in functional groups; over longer time scales, the differences in elemental composition may be related to progressive transformations of functional groups of intermediate products and/or other reactions. Our results infused new data toward the understanding of DOM processing by bacterioplankton in estuarine systems.

#### Edited by:

*Toshi Nagata, University of Tokyo, Japan*

#### Reviewed by:

*Dave Kirchman, University of Delaware, USA Hiroshi Ogawa, University of Tokyo, Japan*

> \*Correspondence: *Patricia M. Medeiros medeiros@uga.edu*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *12 December 2016* Accepted: *27 February 2017* Published: *14 March 2017*

#### Citation:

*Medeiros PM, Seidel M, Gifford SM, Ballantyne F, Dittmar T, Whitman WB and Moran MA (2017) Microbially-Mediated Transformations of Estuarine Dissolved Organic Matter. Front. Mar. Sci. 4:69. doi: 10.3389/fmars.2017.00069* Keywords: dissolved organic matter, biodegradation, estuarine processes, DOM composition, FT-ICR MS

## INTRODUCTION

Bacterioplankton control the flux of dissolved organic matter (DOM) into the coastal microbial food web and influence the release of inorganic carbon to atmospheric and offshore reservoirs. Thousands of compounds make up the estuarine DOM pool, each with different biological turnover rates (Poretsky et al., 2010). This complex DOM pool is processed by a diverse community of heterotrophic bacteria composed of hundreds of taxa (Giovannoni and Stingl, 2005) with varying ecological strategies for the uptake and metabolism of carbon using largely uncharacterized biochemical pathways (Cottrell and Kirchman, 2000; Mou et al., 2007). While measures of changes in chemical characteristics and bulk metabolic rates have provided considerable insights into DOC turnover (Kirchman et al., 2001; Repeta et al., 2002; Benner and Kaiser, 2003; Hertkorn et al., 2006), alone they are insufficient to fully understand the transformation of this complex mixture.

Riverine input can be a large source of DOM of terrigenous origin to estuaries and coastal areas (Hedges et al., 1997; Fichot and Benner, 2014; Medeiros et al., 2016). As riverine DOM is transported toward the ocean, its quantity and composition is often altered (Cole et al., 2007) by a variety of chemical, biological and physical processes (Tzortziou et al., 2008), making estuaries hot spots of DOM cycling. Significant gradients exist in the composition of DOM in estuaries (Sleighter and Hatcher, 2008; Medeiros et al., 2015c; Seidel et al., 2015). In general, DOC concentrations and the DOM aromaticity decrease toward the ocean (Abdulla et al., 2013) at the same time that DOM lability increases (D'Andrilli et al., 2015). Despite intense cycling, a large portion of compounds persist during estuarine passage (Osterholz et al., 2016).

Understanding transformations of the DOM pool in estuaries is important to constraint the characteristics of the DOM introduced to the coastal zone (Bauer and Bianchi, 2011) that may ultimately reach the deep sea (Opsahl and Benner, 1997). Characterizing DOM degradation remains challenging, however, due to the complex nature of DOM, the variety of microbial metabolic pathways, and the various environmental conditions that regulate bacterial metabolism (Eichinger et al., 2011). Despite these challenges, much has been learned by tracking changes in DOC concentration and DOM composition during incubation experiments, including the variation in biomineralization rates as a function of DOM source (e.g., Moran and Zepp, 1997; Obernosterer and Benner, 2004), the relationship of microbial activity to the production of labile (e.g., Kawasaki and Benner, 2006) and refractory (e.g., Ogawa et al., 2001; Jiao et al., 2010; Lechtenfeld et al., 2015) DOM, and how bacterial-derived refractory material compares structurally to refractory DOM in the ocean (e.g., Osterholz et al., 2015). Studies using ultrahigh resolution mass spectrometry have revealed additional molecular-level transformations in DOM as a result of microorganism activity (Kujawinski et al., 2004), including a significant decrease in the molecular diversity of terrigenous DOM (Seidel et al., 2015) and a preferential degradation of oxygen-rich molecules on a time scale of several days (Kim et al., 2006; Medeiros et al., 2015c; Seidel et al., 2015).

Here, we use incubation experiments lasting from 24 h to 70 days to identify microbially-mediated transformations in DOM composition based on ultrahigh resolution mass spectrometry analysis. The incubations were pursued using water from three sites along a marsh-dominated estuary characterized by strong spatial and temporal variability in DOM sources, including terrigenous, marine, and salt marsh inputs (Medeiros et al., 2015a). The estuarine system is characterized by large interannual variability, and some of the incubations captured a record drought during fall 2012, when river discharge was the lowest recorded since measurements began in 1931 (Medeiros et al., 2015a). The two main goals of this study were to (1) identify how DOM from different sources influence the patterns of DOM transformation at the molecular level due to biodegradation, and to (2) identify the dominant changes in DOM composition resulting from microbially-mediated transformations over time scales from hours to months.

#### METHODS

### Sample Collection and Microbial Incubations

DOM was collected in the Altamaha-Doboy-Sapelo Estuary off the southeastern U.S. coast in November 2012 and May 2013. Samples were collected in triplicate at three locations: the Altamaha River (salinity S = 0 to 11; strong terrigenous inputs), the mouth of Doboy Sound (S = 22.9 to 34.6; large marine influence) and near the head of Sapelo Sound (S = 9.1 to 31.5; mixture of terrigenous, marine and salt marsh inputs; Medeiros et al., 2015a) (Figure S1). All samples were collected during high tide. Immediately after collection, triplicate water samples from each site were filtered sequentially through Whatman GF/D filters (pre-combusted at 450◦C for 5 h; nominal 2.7 µm pore size) and 0.2 µm Pall Supor membrane filters into pre-combusted amber glass bottles. Fifty milliliters aliquots of the filtrate from each sampling site were set aside in separated beakers for the preparation of microbial inocula. For that, the 0.2 µm filters from all locations were aseptically cut into pieces and equal areas of the filters were pooled into 50 mL aliquots of 0.2 µm filtrate from each of the three sites and stirred for 30 min. The resulting filtrates, now containing microbes from all sites, were added back to the appropriate set of bottles. This ensured that functional capabilities of the microbes were similar during all incubations and changes in DOM composition could therefore be attributed predominantly to the different initial composition of the DOM pools. We note that fine particles that passed through the GF/D filter but were retained in the 0.2 µm membrane were likely also introduced to the bottles. After the addition of inorganic nutrients (20 µM Na2PO<sup>4</sup> and 50 µM NH4Cl), a triplicate set from each sampling site was immediately filtered (0.2 µm) in order to characterize the initial condition for each experiment. Aliquots were stored frozen for DOC analysis, and the remaining filtrates were acidified to pH 2 (using HCl) and DOM was extracted using solid phase extraction (SPE) cartridges (Agilent Bond Elut PPL) as in Dittmar et al. (2008). We refer to those samples as T0.

The remaining triplicate sets were incubated in the dark at the temperature measured at the time of collection for up to 70 days. At days 35 and 70, samples were filtered, acidified, and DOM was extracted using PPL cartridges as described previously. These samples are referred to as T35 and T70. Shorter-term incubations lasting 24 h were also conducted with water collected at the mouth of Doboy Sound in February and in December of 2011 (referred to as T1). Aliquots for DOC analysis were collected for all incubations as described above (i.e., prior to acidification, stored frozen).

#### Chemical Analyses

In all cases, DOC concentration in water samples and extracts (i.e., dried and redissolved in ultrapure water) was measured with a Shimadzu TOC-VCPH analyzer. Analytical accuracy and precision were tested against the Consensus Reference Material (Hansell, 2005) and were better than 5%. SPE extraction efficiency across all samples was 75 ± 5% of the DOC. Bulk δ <sup>13</sup>C of extracted DOC (SPE-DOC) was measured with a Finnigan MAT 251 isotope ratio mass spectrometer after complete drying. Procedural blanks did not yield detectable amounts of carbon isotope contamination. The molecular composition of DOM extracts (15 mg C L−<sup>1</sup> in 1:1 methanol/water) was analyzed using a 15T Fourier transform ion cyclotron resonance mass spectrometer (FT-ICR MS; Bruker Daltonics) with electrospray ionization (negative mode) as in Seidel et al. (2014). Spectra were internally calibrated with >100 known CxHyO<sup>z</sup> molecular formulae over the mass range in the samples. With this calibration procedure, a mass error of <0.1 ppm was achieved. Before each sample set, blank checks with methanol and ultrapure water were measured. Molecular formulae were calculated in the mass range between 150 and 850 Da by applying the following restrictions: <sup>12</sup>C1−<sup>130</sup> <sup>1</sup>H1−200O1−<sup>150</sup> <sup>14</sup>N0−4S0−2P0−2. Assignment of molecular formulae was done considering a maximum mass error of 0.5 ppm and using the criteria described by Rossel et al. (2013). Only compounds with a signal-to-noise ratio of 4 or higher were used in the analysis to eliminate inter-sample variability based on peaks that were close to the limit of detection. Additional details of molecular formulae assignment are given in Seidel et al. (2014). A sample from the North Equatorial Pacific Intermediate Water collected at station NELHA off Big Island in Hawaii at 674 m water depth (Osterholz et al., 2015; Riedel et al., 2016) was also analyzed repeatedly (41 times) in order to characterize differences in DOM composition as identified by FT-ICR MS that arise solely due to instrument variability.

Patterns of variability in FT-ICR MS-derived DOM composition were distinguished by principal component (PC) analysis using all replicates simultaneously as in Medeiros et al. (2015a,b,c). All PCs shown here are significantly different (95% confidence level) from results of PC analysis of spatially and temporally uncorrelated random processes (Overland and Preisendorfer, 1982). The Wilcoxon rank-sum test was used for comparisons between groups of compounds, as in Osterholz et al. (2016).

#### RESULTS

#### Spatial and Temporal Variations in DOM Composition

A detailed characterization of spatial and temporal differences in DOM composition in this system has been presented in Medeiros et al. (2015a). We briefly summarize results here for comparison with microbially-mediated transformations described in the next section. DOC concentrations were substantially lower near the ocean at Doboy Sound than at the other sites (Table S1) and increased at all locations in May 2013 when river discharge was high. The pre-incubation samples (T0; solid symbols) were spread along the first PC (PC1) axis (**Figure 1A**), which separated samples with high salinity (e.g., Doboy in November 2012) from those with low salinity (e.g., Altamaha in May 2013) (Table S1) (see caption of **Figure 1** for example of interpretation of the analysis). The correlation coefficient between PC1 and salinity for the pre-incubation samples was −0.87 (p < 0.01). The loading of PC1 (**Figure 1B**) shows a pattern typical of river-to-ocean

transects (Medeiros et al., 2015c), with high salinity samples enriched with compounds with high H/C ratios compared to lowsalinity samples, which is consistent with the higher abundance of aliphatic structures in marine DOM (Sleighter and Hatcher, 2008; Osterholz et al., 2016). This is also consistent with the strong gradient in DOM sources in the system associated with terrigenous vs. oceanic inputs (Medeiros et al., 2015a).

For any given location, samples from May 2013 had higher DOC concentrations and DOM with a more terrigenous character than samples from November 2012 (e.g., samples from Altamaha River in May 2013 had higher PC1 scores than samples from Altamaha River in November 2012; the same is true for samples from Sapelo to Doboy Sounds; **Figure 1A**), a pattern consistent with their lighter δ <sup>13</sup>C values (Table S1). This difference probably resulted from a severe drought in fall 2012, when the Altamaha River discharge was anomalously low (Medeiros et al., 2015a). During that time, salinity at all sampling sites was higher than in May 2013 (Table S1), which is consistent with the more marine character of the DOM observed in November 2012.

## Microbially-Mediated Transformations in DOM Composition

In addition to capturing the gradient in DOM composition in the estuary (section Spatial and Temporal Variations in DOM Composition), the PC analysis shown in **Figure 1** also captured microbially-mediated transformations of the DOM. The largest transformations in DOM as captured by the first 2 principal components [i.e., the largest distances in PC space between T0 (solid symbols) and T70 (open symbols) for a given incubation in **Figure 1A**] occurred in samples with a more marine character (i.e., Doboy Sound in November 2012, black circles), decreasing progressively in samples with a more terrigenous signature. Incubations of Altamaha River water in May 2013 (green squares), when discharge was near the seasonal maximum (Medeiros et al., 2015a), yielded no significant changes in DOM composition as detected by the first 2 PCs (a different pattern of DOM transformation was observed in that case; see Figure S2).

It is interesting to note that microbially-mediated transformations resulted in changes on both the first and second PCs. The PC1 scores at the end of the incubations (T70) were generally higher than those at the start of the incubations (T0). Since large PC1 scores are associated with terrigenous DOM (see section Spatial and Temporal Variations in DOM Composition), this indicates that microbial degradation resulted in DOM with a more terrigenous character. This presumably resulted from preferential degradation of marine DOM, which is indeed thought to be more labile than freshwater DOM (D'Andrilli et al., 2015). As marine DOM (i.e., molecular formulae shown in blue in **Figure 1B**) was degraded during the experiment, terrigenous DOM (i.e., molecular formulae shown in red in **Figure 1B**) became relatively more important compared to the initial state. The shift toward a terrigenous signature (i.e., a shift toward high values along the PC1 axis) was stronger in the samples that had a stronger marine signature initially (**Figure 1A**). For the incubations of Doboy Sound water in November 2012, in particular, PC1 changed substantially during the incubation, representing 53% of the distance along the PC1 axis between marine (solid black circle, PC1 = −2.45 ± 0.03) and terrigenous (solid green square, PC1 = +0.86 ± 0.05) DOM. In other words, PC1 (**Figures 1A,B**) captured the fraction of the DOM that is statistically different between the sites and is related to the gradient in DOM composition due to terrigenous vs. marine inputs (as shown in section Spatial and Temporal Variations in DOM Composition). Biodegradation during the incubation of Doboy Sound water from November 2012 resulted in large enough changes in that fraction captured by PC1 that the DOM composition as identified by FT-ICR MS at the end of the experiment (open black circles in **Figure 1A**) was nearly as similar to riverine DOM (solid green squares) as it was to the original pre-incubated marine DOM (solid black circles). We note that photo-degradation generally produces the opposite effect, with riverine DOM resembling marine DOM in its broad molecular composition after extensive photo-degradation mainly due to removal of aromatics (Riedel et al., 2016).

Pre- and post-incubation samples were also characterized by different PC2 scores (**Figure 1A**). Microbially-mediated transformations therefore also resulted in a shift in DOM composition that did not fall along the terrigenous-marine gradient captured by PC1. The loading of PC2 reveals that molecules characterized by low O/C and high H/C ratios (molecules in red in **Figure 1C**) were relatively enriched at the end of the incubations. A detailed investigation of those changes is presented in the next section.

## DOM Transformations at Doboy Sound

These analyses show that microbial degradation resulted in shifts in DOM composition so that the pool as a whole lost some of its marine characteristics and became more similar to terrigenous DOM. However, the analyses included samples from different locations and times. To examine the specific chemical formula changes occurring during biodegradation, the PC analysis was repeated considering only incubations of DOM from Doboy Sound in November 2012. This removed effects of spatial and temporal variability in DOM composition from the analysis, so that transformations could be primarily attributed to biodegradation.

DOC concentrations decreased progressively during the course of the incubation (Table S1), indicating remineralization of a fraction of the DOM pool. The score of the first principal component (referred to as PC1DOB−long) increased with time (**Figure 2A**), and compounds associated with molecular formulae shown in blue in **Figure 2B** (negative loading) decreased in relative abundance during microbial processing while those associated with formulae shown in red (positive loading) increased with processing. As such, compounds that decreased or increased in relative abundance had distinct chemical characteristics. There was a relative enrichment of molecules that occupy the upper left corner of the van Krevelen diagram after the incubation (**Figure 2B**), i.e., a shift toward low O/C and high H/C ratios, which is consistent with molecular transformations observed in dark incubations of DOM from the Amazon River mouth (Medeiros et al., 2015c; Seidel et al., 2015). Additionally, while formulae whose relative abundances strongly decreased after the incubation (i.e., more negative loadings of PC1DOB−long in **Figure 2B**) were characterized by high masses (553.5 ± 78.7 Da), formulae whose relative abundances were enriched after the incubation (i.e., more positive PC1DOB−long loading in **Figure 2B**) were characterized by significantly lower masses (286.0 ± 93.9 Da) (**Figure 3A** and **Table 1**; Wilcoxon rank-sum test, p < 0.01). Molecular formulae associated with compounds that were preferentially depleted during the incubation were also characterized by a significantly higher numbers of carbon, hydrogen and oxygen atoms compared to compounds whose relative abundance enriched due to microbially-mediated transformations in DOM composition (**Figures 3B–D**; Wilcoxon rank-sum test, p < 0.01). They were also characterized by lower aromaticity indices (AI<sup>∗</sup> ; a measure of the aromaticity of the molecules; Koch and Dittmar, 2006, 2016; **Figure 3H**), although the differences are small (but still significant; Wilcoxon rank-sum test, p < 0.01). Lastly, the transformations in DOM composition resulting from biodegradation (**Figure 2B**) were compared with the DOM composition variability associated with differences in terrigenous

and marine inputs (**Figure 1B**) on a molecular basis. Molecular formulae associated with compounds whose relative abundances decreased after the microbial transformation (blue dots in **Figure 2B**) were generally also associated with compounds enriched in marine DOM (blue dots in **Figure 1B**), although there is substantial scatter around the bin averages (**Figure 3E**). Molecular formulae associated with compounds whose relative abundances increased after the incubation, on the other hand, were associated with compounds enriched in terrigenous DOM, with much smaller scatter around the bin averages. Thus, the marine DOM appeared to be preferentially transformed during the incubations, leaving behind relatively more terrigenous DOM, which is consistent with results shown in **Figure 1**. The large scatter around the formulae associated with the compounds depleted during the incubation (**Figure 3E**) indicated that, although there is a preference for biotransformation of marine DOM, a fraction of the compounds enriched in terrigenous DOM were also transformed during the incubation.

Repeating the PC analysis using only samples from days 0 to 35 (T0 and T35), T0 and T70, or T35 and T70 led to results similar to those obtained using all time points together. This indicates that the dominant pattern of DOM compositional change associated with biodegradation remained similar during the course of the incubation, at least at the monthly time scale captured by the sampling. Moreover, when the T0, T35 and T70 samples from Doboy Sound were analyzed together, approximately 60% of the total change in PC1DOB−long occurred in the first 35 days (i.e., from −1.29 ± 0.13 to 0.15 ± 0.27 along PC1DOB−long axis; **Figure 2A**), with the remaining 40% of the change (i.e., from 0.15 ± 0.27 to 1.10 ± 0.09; **Figure 2A**) occurring in the second half of the incubation. This indicates that active DOM transformations occurred throughout the experiment.

#### Changes in DOM Composition Over Short Time Scales

We have also investigated changes in DOM composition during the 24-h long incubations (T0-T1; **Table 1**). At those short time

incubation of Doboy Sound water in November 2012. Loading of PC1DOB−long shown on x-axis in all panels are values from Figure 2B. Fifty uniformly spaced bins were selected spanning the range of the values of the loading of PC1DOB−long. Then, for each bin, the average <sup>±</sup>1 standard deviation of each chemical characteristic [molecular mass (A), number of carbon (B), hydrogen (C), and oxygen (D) atoms, oxygen/carbon (F) and hydrogen/carbon (G) ratios] were computed. Y-axis on panel (E) distinguishing marine and terrigenous DOM are values from Figure 1B. AI\*: modified aromaticity index (H) (Koch and Dittmar, 2006, 2016).

scales, the overall change in DOC concentrations were quite small (Table S1). Molecular formulae which were enriched after the 1-day experiments had chemical characteristics that were


TABLE 1 | Chemical characteristics of formulae whose relative abundances were enriched vs. depleted during incubations of Doboy Sound samples.

*Values given are average* ±*1 standard deviation. Differences between chemical characteristic of enriched and depleted formulae are also shown.*

relatively enriched at the end of the incubations.

similar to those whose relative abundances decreased, occupying similar regions in van Krevelen space (**Figure 4**). For example, molecular formulae that were preferentially depleted after the 1-day incubations had masses of 485–500 Da, while formulae that were preferentially enriched had masses of 460–475 Da (**Table 1**; those differences were statistically significant; Wilcoxon rank-sum test, p < 0.01). Thus, the difference in the average mass of molecular formulae whose relative abundance decreased vs. increased during the short incubations (∼1 mass = 25–30 Da) was substantially smaller than the difference observed in the incubation lasting 70 days (∼1 mass = 270 Da; **Table 1**). Likewise, the differences in the number of carbon and oxygen atoms between compounds whose relative abundances decreased vs. increased in the 1-day incubations were also small (23 vs. 20–22 carbon atoms, respectively; unaltered for oxygen atoms; **Table 1**).

#### Changes in DOM Composition at Sapelo Sound and the Altamaha River

We also pursued a separate PC analysis for the other sets of replicates from each long incubation (e.g., from Sapelo Sound in May 2013, from Altamaha River in November 2012), similarly to the analysis presented in **Figures 2**, **3**. The analysis revealed a progressive shift in the attributes of transformed DOM (**Figure 5**). In the 70-day incubation of DOM from the Altamaha River in May 2013 (i.e., water rich in terrigenous material; **Figure 1**), the compounds whose relative abundance enriched and decreased during the incubation had similar molecular

masses (**Figure 5**). In that regard, results were qualitatively similar to the 1-day incubation (T0-T1) of the more marineinfluenced Doboy Sound DOM (**Table 1**). As the marine character of the sample increased (as identified in **Figure 1A**), the mass difference between decreasing vs. increasing compounds also increased (**Figure 5**). For the incubation of Doboy Sound water from November 2012, increasing the duration of the experiment from 35 to 70 days led to a slight (but statistically significant) increase in the mass difference between compounds depleted vs. enriched.

## DISCUSSION

#### Temporal Changes in DOM Transformation

Since the average mass difference between compounds whose relative abundance significantly increased vs. decreased in the short incubation of Doboy Sound water is small (average of 20–30 Da; **Table 1**) and they often occupy similar regions in van Krevelen space (**Figure 4**), it is possible that those differences are reflecting transformations of functional groups. Side chain oxidation (e.g., Filley et al., 2002), demethylation, or hydroxylation (e.g., Crawford and Crawford, 1980) of compounds present in the initial DOM pool (i.e., at T0) would lead to a reduction in the abundance of those compounds at the end of the incubation and to an increase in the abundance of compounds with slightly lower molecular masses. For example, vanillic acid (∼168 Da) can be converted via demethylation to protocatechuic acid (∼154 Da) (Merkens et al., 2005). In this case, the difference in the mass of the compounds that were consumed and produced is small at 14.0157 Da (-CH2). If multiple progressive transformations occur, however, then the difference in the mass of the compounds consumed and produced will be larger (Bugg et al., 2011). As an example, the lignin model biphenyl compound 5,5′ -dehydrodivanillate (∼332 Da) can be transformed by bacterial degradation via an O-demethylation followed by a subsequent ring cleavage (Masai et al., 2007). If the resulting product is further hydrolyzed, decarboxylated, and then demethylated, protocatechuic acid is generated. In this case, the change in the mass between compounds consumed and produced progressively increases with each transformation (i.e., O-demethylation, ring cleavage, etc), reaching ∼178 Da for the end result (i.e., comparing 5,5′ dehydrodivanillate and protocatechuic acid). Although, those examples come from investigations of microbial degradation of terrigenous material in soils, they provide illustrations of how biodegradation of DOM may result in small shifts in masses initially, and to large shifts in masses later on associated with additional, progressive transformations of the intermediate products. This scenario is consistent with the progression observed for microbially-mediated transformations in DOM composition in the incubations shown here. For Doboy Sound water, for example, the short term incubations yielded small differences in the average mass of the compounds whose relative abundances significantly decreased vs. increased (**Table 1**), which could be associated with the removal of a functional group (e.g., demethylation). If those resulting products are further transformed progressively via modification of additional functional groups (e.g., demethylation followed by hydroxylation and decarboxylation), then a large difference in the average mass of the compounds whose relative abundances significantly decreased vs. increased would be expected, which is consistent with results from the long-term incubations (**Table 1**). Alternatively, the large difference in masses between consumed and produced compounds in the long-term incubation could also be a result of different reactions. It will be interesting to see if emerging tools in analytical chemistry, microbiology and informatics (Moran et al., 2016) will allow for future studies to identify the specific pathways that occur in the transformation of DOM.

#### Microbially-Mediated DOM Compositional Changes at the Molecular Level

In the previous section, we hypothesized that progressive modifications of functional groups could explain why the differences in the masses of compounds whose relative abundance decreased vs. increased was small in short-term incubations, and comparatively larger in long-term experiments. Here, we compare the elemental composition of the compounds that changed significantly during the incubations to identify if those changes are consistent with that hypothesis.

The molecular formulae associated with the compounds whose relative abundance changed during the 1-day incubations have masses that are similar to each other. This does not necessarily mean, however, that they can be related to each other via transformations of side functional groups. Compounds with similar masses can have very different elemental compositions, and FT-ICR MS analysis does not give direct information about structure (e.g., Longnecker et al., 2015). We therefore investigated the possibility of transformations of functional groups using a strictly probabilistic approach. The analysis began by first selecting all molecular formulae associated with compounds whose relative abundance significantly increased after a given incubation. For the 1-day incubation (triplicate) from February 2011, for example, there were 496 such molecular formulae. For each of these, we computed the difference between its elemental composition and the elemental composition of all molecular formulae associated with compounds whose relative abundance decreased during the incubation (there were 408 such formulae in the February 2011 incubation). Therefore, the February 2011 calculation yielded a total of 202,368 (496 × 408) possible composition differences of potential substrates and products. Lastly, we sorted those composition differences to identify how often a potential change in elemental composition appeared (see Figure S3 for an example of the calculation). The rationale behind this calculation is that, if the microbial community is transforming the DOM via modification of a specific functional group, then it is more likely that a difference in elemental composition between compounds whose relative abundance decreased vs. increased equivalent to that modification will appear more often in the difference matrix. For example, if the microbial community is actively demethylating (i.e., targeting –CH3), then it is more likely that a difference in elemental composition of a CH<sup>2</sup> (e.g., replacement of a CH<sup>3</sup> group by an H) between compounds whose relative abundance changed will be common. If, on the other hand, the differences in molecular formulae between relatively depleted and enriched compounds appear random, then it seems less likely that the microbial community is biasing their activity toward specific molecular modifications. If the most frequent composition differences in the 1-day incubations involve additions and removals of large numbers of carbon and heteroatoms, then the similarity in masses (**Table 1**) is likely simply coincidental and unlikely to have resulted from direct transformation of side functional groups. For example, although C16H6O<sup>9</sup> and C13H13O7NS have a small mass difference of 14.9599 Da, the former cannot be converted into the latter via simple transformations of side functional groups given the large difference in their elemental composition.

The calculations revealed that, for the 1-day incubations, the most common differences in elemental composition were those involving a small number of C, H, and/or O atoms (e.g., removal or addition of a CH2; removal of an O) (Table S2) and a small mass (Figure S4). This indicates that compounds whose relative abundances were significantly decreased vs. increased in the short incubations not only had similar masses (**Table 1**) and occupied similar regions in van Krevelen space (**Figure 4**), but many of them also had similar elemental compositions (i.e., small differences in the number of atoms in Table S2). Of the 200 most frequently observed composition differences shown in Table S2, 82 appeared in both the February and the December 2011 short incubations. If the analysis is expanded to include the 800 most frequently observed composition differences, then 268 appear in both incubations. Some of the most common TABLE 2 | Most common potential transformations in elemental compositions that would have resulted in a decrease in molecular mass after short-term incubations.


*<sup>a</sup>Average of short-term incubations of Doboy Sound water (February and December 2011).*

composition differences that appeared in both the February and the December 2011 short incubations are listed in **Table 2**. For the 70-day incubation, on the other hand, the most frequent composition differences involve a net removal of a much larger number of C, H, and O (e.g., net removal of C16H8O15, Table S2) and a larger mass (Figure S4). These could be related to changes in functional groups occurring sequentially (e.g., demethylation followed by hydroxylation and decarboxylation, whose cumulative result would be a large change in elemental composition) and/or other reactions. There is no overlap between the composition differences appearing in the short and long incubations. It is likely that similar changes do occur in the long incubations, but because of their cumulative effect (i.e., changes occurring on intermediate products) the net change in elemental composition is different than in the short incubations. These results are consistent with our previous hypothesis that changes in DOM composition in the 1-day incubations may be related to transformations in side functional groups (with a small change in elemental composition and mass), while changes in the long incubations may be related to progressive transformations (and thus with a larger change in elemental composition and mass). We note that previous incubations of coastal water lasting for up to 9 days have not revealed a significant impact of incubation time on DOM composition (Kujawinski et al., 2016). Our analysis of short- (1 day) and long-term (35 or 70 days) incubations indicate that significant differences can occur over these longer time scales.

The patterns just described for short- and long-term incubations are not due to instrument variability or noise. First, the principal component analysis separated the pre-incubation samples from those collected at the end of the experiments in both cases (**Figures 2A**, **4A,C**). If the transformation in DOM composition shown in the van Krevelen diagrams (**Figures 2B**, **4B,D**) were simply due to instrument noise, there would be no reason to expect separation between pre- and postincubation samples in the principal component analysis. Second, we repeated the analysis using all 41 injections of the sample from the North Equatorial Pacific Intermediate Water collected at station NELHA off Hawaii. The differences in elemental composition between those samples (which in this case were entirely due to instrument variability) were mostly random, with small and large changes in molecular mass occurring frequently.

It is important to note some of the limitations of the analysis. The calculation provides a check for consistency only, since it is possible that even when the difference in elemental composition is small (small number of C, H, and/or O atoms), the structure of the compounds that underlies the predicted formula may not allow for that particular transformation. In addition, the analytical methods used here are focused on the investigation of compounds with molecular mass larger than about 150–160 Da. As reviewed by Kujawinski (2011), previous studies have shown that small molecules that are ubiquitous in cellular biomass can be easily assimilated into cells through dedicated transport systems. Since those compounds fall outside our analytical window, however, the transformations in DOM composition reported here are only representative of the fraction in the 150– 850 Da range that can be detected by FT-ICR MS analysis. The analyses are also only representative of the fraction of the DOM that is retained in the solid phase extraction resin. Saccharides, for example, are known to be consumed by microorganisms (e.g., Rich et al., 1996), but are generally not well captured by the resin used here. As such, some of the DOC decrease observed in the incubations is likely related to a fraction of the DOM that is not captured by the FT-ICR MS analysis. The statistically significant transformations in composition reported in the shortterm incubations reveal, however, that at least a portion of the labile DOM was captured by the solid phase extraction. Lastly, we note that we have focused here on transformations of estuarine DOM due to microbial degradation. In situ transformations are influenced by additional processes, including photochemistry and flocculation.

In summary, our analyses of microbial-mediated changes in DOM composition in an estuarine system characterized

#### REFERENCES


by strong terrigenous and marine inputs identified significant transformations. Analyses at the molecular level revealed a general preference for degradation of compounds that are enriched in marine DOM, which resulted in the remaining DOM becoming relatively more terrigenous. Transformations observed during short-term incubations resulted in a small difference in the average mass of compounds whose relative abundance decreased vs. increased, and a novel probabilistic approach revealed that changes in elemental composition were consistent with modifications of functional groups. Over longer time scales, the differences in mass were significantly larger, possibly a result of additional changes in function groups of intermediate products. Despite limitations, and given the challenges of actually identifying transformations in the complex DOM mixture, the analysis presented here serves as a step toward shedding light into these elusive processes. Future studies combining this probabilistic approach, NMR analysis and other techniques that provide direct information about structure, and biological gene expression at a highly-resolved scale will be particularly useful to tackle the biological complexity of carbon turnover.

#### AUTHOR CONTRIBUTIONS

PM conceived and designed the research; PM, MS, and TD generated and processed FT-ICR MS data. SG, FB, WW, and MM contributed materials/analysis tools. PM analyzed the data. PM wrote the paper; all authors commented on the manuscript.

#### ACKNOWLEDGMENTS

We thank D. Kirchman and H. Ogawa for their constructive comments and suggestions, which led to a much improved manuscript. This research was supported by the National Science Foundation through the Sapelo Island Microbial Carbon Observatory (SIMCO, OCE-1356010). Additional support was provided by the NSF—Georgia Coastal Ecosystems Long Term Ecological Research program (GCE-LTER, OCE-1237140). We thank the University of Georgia Marine Institute at Sapelo Island for logistical support. This is University of Georgia Marine Institute contribution number 1058.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2017.00069/full#supplementary-material


world rivers. Front. Earth. Sci. 4:85. doi: 10.3389/feart.2016. 00085


transform ion cyclotron resonance mass spectrometry. Mar. Chem. 110, 140–152. doi: 10.1016/j.marchem.2008.04.008

Tzortziou, M., Neale, P., Osburn, C., Megonigal, J., Maie, N., and Jaffé, R. (2008). Tidal marshes as a source of optically and chemically distinctive colored dissolved organic matter in the Chesapeake Bay. Limnol. Oceanogr. 53, 148–159. doi: 10.4319/lo.2008.53.1.0148

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Medeiros, Seidel, Gifford, Ballantyne, Dittmar, Whitman and Moran. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Composition and Transformation of Dissolved Organic Matter in the Baltic Sea

Michael Seidel <sup>1</sup> \*, Marcus Manecki 1, 2, Daniel P. R. Herlemann<sup>2</sup> , Barbara Deutsch3 † , Detlef Schulz-Bull <sup>2</sup> , Klaus Jürgens <sup>2</sup> and Thorsten Dittmar <sup>1</sup>

<sup>1</sup> Research Group for Marine Geochemistry (ICBM-MPI Bridging Group), Institute for Chemistry and Biology of the Marine Environment (ICBM), Carl von Ossietzky University of Oldenburg, Oldenburg, Germany, <sup>2</sup> Leibniz Institute for Baltic Sea Research, Warnemünde (IOW), Germany, <sup>3</sup> Baltic Sea Centre, Stockholm University, Stockholm, Sweden

#### Edited by:

Toshi Nagata, University of Tokyo, Japan

#### Reviewed by:

Clare Elizabeth Davis, University of Liverpool, UK Youhei Yamashita, Hokkaido University, Japan

\*Correspondence: Michael Seidel m.seidel@uni-oldenburg.de

† Present Address: Barbara Deutsch, Biota—Institut für Ökologische Forschung und Planung GmbH, Bützow, Germany

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Earth Science

Received: 07 November 2016 Accepted: 06 April 2017 Published: 02 May 2017

#### Citation:

Seidel M, Manecki M, Herlemann DPR, Deutsch B, Schulz-Bull D, Jürgens K and Dittmar T (2017) Composition and Transformation of Dissolved Organic Matter in the Baltic Sea. Front. Earth Sci. 5:31. doi: 10.3389/feart.2017.00031 The processing of terrestrial dissolved organic matter (DOM) in coastal shelf seas is an important part of the global carbon cycle, yet, it is still not well understood. One of the largest brackish shelf seas, the Baltic Sea in northern Europe, is characterized by high freshwater input from sub-arctic rivers and limited water exchange with the Atlantic Ocean via the North Sea. We studied the molecular and isotopic composition and turnover of solid-phase extractable (SPE) DOM and its transformation along the salinity and redox continuum of the Baltic Sea during spring and autumn. We applied ultrahigh-resolution mass spectrometry and other geochemical and biological approaches. Our data demonstrate a large influx of terrestrial riverine DOM, especially into the northern part of the Baltic Sea. The DOM composition in the central Baltic Sea changed seasonally and was mainly related to autochthonous production by phytoplankton in spring. Especially in the northern, river-dominated basins, a major fraction of riverine DOM was removed, likely by bio- and photo-degradation. We estimate that the removal rate of terrestrial DOM in the Baltic Sea (Bothnian Bay to the Danish Straits/Kattegat area) is 1.6–1.9 Tg C per year which is 43–51% of the total riverine input. The export of terrestrial DOM from the Danish Straits/Kattegat area toward the North Sea is 1.8–2.1 Tg C per year. Due to the long residence time of terrestrial DOM in the Baltic Sea (total of ca. 12 years), seasonal variations caused by bio- and photo-transformations and riverine discharge are dampened, resulting in a relatively invariant DOM molecular and isotopic signature exported to the North Sea. In the deep stagnant basins of the Baltic Sea, the DOM composition and dissolved organic nitrogen concentrations changed seasonally, likely because of vertical particle transport and subsequent degradation releasing DOM. DOM in the deep anoxic basins was also enriched in sulfur-containing organic molecules, pointing to abiotic sulfurization of DOM under sulfidic conditions.

Keywords: Baltic Sea, sub-arctic rivers, discharge, dissolved organic matter, ultrahigh-resolution mass spectrometry, seasonal variation

## INTRODUCTION

Shelf seas are often strongly influenced by riverine inputs of nutrients and organic matter. Although they represent only a small part of the total ocean, about 20% of the marine organic matter production and 80% of organic matter deposition to ocean sediments occurs there (Borges, 2005). Yet, the carbon cycling in coastal waters is still not well understood, partly because the different carbon sources and sinks are highly complex (Gattuso et al., 1998; Ward et al., 2017).

In coastal shelf seas and estuaries, dissolved organic matter (DOM) is derived from land plants, riverine and autochthonous (marine) primary production (Stedmon et al., 2010; Bauer and Bianchi, 2011; Osburn and Stedmon, 2011; Deutsch et al., 2012). Removal and transformation of terrestrial DOM occurs through processes such as flocculation (Sholkovitz et al., 1978), microbial and photo-degradation (Hernes and Benner, 2003; Obernosterer and Benner, 2004; Gonsior et al., 2009; Spencer et al., 2009; Stubbins et al., 2010). Complete remineralization releases inorganic chemical species that then serve as nutrients for aquatic microbes and phytoplankton (Vähätalo and Zepp, 2005). High concentrations of chromophoric DOM in many estuaries limit primary production due the lower light penetration in the water column (Urtizberea et al., 2013). Studying and untangling these features still poses a major challenge because these processes often occur simultaneously, and because of the dynamic nature of estuarine systems (Bauer and Bianchi, 2011).

The Baltic Sea is a shelf sea with high inputs of nutrients from surrounding rivers that stimulate large phytoplankton blooms from spring to autumn, which effectively take up CO2. At the same time, high primary production of organic matter (OM) stimulates heterotrophic activity, which is further enhanced by the input of terrestrial DOM from rivers. The net uptake or release of atmospheric CO<sup>2</sup> differs for the different subbasins of the Baltic Sea (Thomas et al., 2010; Kulinski and ´ Pempkowiak, 2011; Gustafsson et al., 2014; Ylöstalo et al., 2016): the central Baltic Sea is a CO<sup>2</sup> sink, while the northern Bothnian Bay is a weak source of CO<sup>2</sup> to the atmosphere (Löffler et al., 2012; Schneider et al., 2014).

It is likely that increasing temperature and precipitation will lead to a higher riverine export of terrestrial DOM (Voss et al., 2011; Räike et al., 2016). Understanding the fate of this enhanced input of terrestrial carbon in sub-arctic shelf seas along marineterrestrial and redox-gradients is therefore important. The Baltic Sea is an ideal system to study the biogeochemical cycling of DOM in terrestrial-to-marine and in oxic-to-anoxic gradients in detail, because of its estuarine character in combination with long water residence times ranging from months to years (Savchuk, 2005). The Baltic Sea in northern Europe is a semienclosed, brackish marginal sea. It is characterized by limited water exchange with the Atlantic Ocean via the North Sea and high freshwater input from rivers. The shallowness and infrequent inflow of saline and dense North Sea water causes a permanent stratification in the central Baltic Proper at water depths >60 m (Kulinski and Pempkowiak, 2012 ´ ). The northern drainage basin is dominated by sub-arctic boreal forests while the southern basin is dominated by agriculture and high population density. Consequently, parts of the southern Baltic Sea are threatened by eutrophication and extension of bottom water anoxia (Thomas et al., 2010). Hypoxic conditions are reached when oxygen concentrations fall below 2 mL L−<sup>1</sup> and alternative terminal electron acceptors, such as nitrate and sulfate, are used during heterotrophic degradation of organic matter by microbes (Conley et al., 2009). Microbial sulfate reduction causes sulfidic conditions in the deep anoxic basins of the central Baltic Sea (Savchuk, 2005; Voss et al., 2011).

Previous studies have used molecular biomarkers, isotopes, and optical properties to differentiate marine and terrestrial sources of DOM in the Baltic Sea (Højerslev et al., 1996; Bianchi et al., 1997; Stedmon et al., 2000, 2010; Alling et al., 2008; Osburn and Stedmon, 2011; Deutsch et al., 2012). For example, the concentration of suspended lignin phenols (unique biomarkers for vascular plants, e.g., Hernes and Benner, 2003) decreased from north to the south due to high freshwater inflow into the northern Baltic Sea (Bianchi et al., 1997). Stable carbon isotope ratios of high-molecular weight (HMW) DOM revealed a nonconservative behavior and removal of terrestrial HMW DOM in the Baltic Sea (Alling et al., 2008; Deutsch et al., 2012). Residence times for terrestrial HMW DOM ranged from 2.8 years in the northern to 4.5 years in the southern Baltic Sea, and more than 50% of it was proposed to be removed in the coastal Baltic Sea (Deutsch et al., 2012). Osburn and Stedmon (2011) used optical and chemical DOM properties and estimated that 0.8 Tg yr−<sup>1</sup> terrestrial dissolved organic carbon (DOC) was exported from the Baltic to the North Sea. The molecular composition, the transformation and the export of low molecular weight terrestrial DOM, however, remains unexplored.

Ultrahigh-resolution Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) is a powerful tool to deepen our understanding of DOM processing in estuarine settings, because it resolves thousands of molecular formulae in the very complex DOM mixtures. Consequently, this technique was used to study the transformation of DOM in a wide range of aquatic systems, including terrestrial-to-marine and oxicto-anoxic transition zones (Koch et al., 2005; Kim et al., 2006; Sleighter and Hatcher, 2008; Schmidt et al., 2009; Longnecker and Kujawinski, 2011; Lechtenfeld et al., 2013). For example, in the Delaware Estuary (USA), compositional differences at the molecular level were not reflected in changes of DOC concentrations (Osterholz et al., 2016). Several other studies have shown that the transformation of terrestrial and the new production of autochthonous (marine) DOM compounds leave characteristic molecular imprints (Sleighter and Hatcher, 2008; Kujawinski et al., 2009; Medeiros et al., 2015a,b; Seidel et al., 2015b). Aromatic terrestrial DOM compounds generally decrease along salinity gradients while the DOM heteroatom content increases due to autochthonous production (Sleighter and Hatcher, 2008; Osterholz et al., 2016). Likewise, the alteration of DOM under different redox-regimes can leave identifiable molecular imprints such as the relative enrichment of sulfurcontaining compounds (Schmidt et al., 2009; Seidel et al., 2014, 2015a; Sleighter et al., 2014; Gomez-Saez et al., 2016).

Here we used ultrahigh-resolution mass spectrometry to determine the DOM molecular composition in the Baltic Sea from the sub-arctic north to the boreal south during spring and autumn sampling and relate it to geochemical and biological tracers. We aimed to answer the following research questions: What is the composition and the fate of DOM from marine and terrestrial sources in the Baltic Sea? How do the anoxic conditions influence the DOM composition and transformations? How do DOM and carbon fluxes to the North Sea vary seasonally?

## MATERIALS AND METHODS

#### Study Site

The Baltic Sea consists of five major basins, the Bothnian Bay, the Bothnian Sea, the Baltic Proper, the Gulf of Riga, and the Gulf of Finland (**Figure 1**). The freshwater inflow to the Baltic Sea is ∼480 km<sup>3</sup> yr−<sup>1</sup> , of which 200 and 100 km<sup>3</sup> yr−<sup>1</sup> are received by the Bothnian Bay (e.g., inflow of the Kalix River) and the Gulf of Finland (Neva River), respectively (Voss et al., 2011). Surface salinity ranges from two in the northernmost Bothnian Bay to >20 in the Kattegat and >30 in the Skagerrak. Deep-water salinity is 3–7 in the Bothnian Sea >50 m depth and 10–13 in the central Baltic Sea >60 m depth (Kullenberg and Jacobsen, 1981; Voss et al., 2011). Due to the shallow connection, water exchange between the Baltic Sea and the North Sea is limited. Following a typical estuarine circulation pattern, North Sea water enters through the Skagerrak and Kattegat at the bottom while brackish Baltic Sea water builds an outflow of water on top. The halocline in the central Baltic Sea at about 60 m limits vertical water mixing, and bottom water exchange between the deep basins of the Baltic Sea is further limited by sills. This causes the development of bottom water anoxia in the deep basins such as the Gotland and Landsort Deeps.

## Sampling, Chemical, and Biological Analyses

Samples were collected during two cruises with R/V Meteor, in November/December 2011 (leg M86/1, autumn), and in May/June 2012 (leg M87/3, spring; **Figure 1**). Sampling in 2012 covered the spring flood of sub-arctic rivers draining large permafrost areas in the Northern Baltic region.

Conductivity, temperature, chlorophyll a fluorescence and dissolved oxygen concentrations of the water samples were measured in situ using a conductivity-temperature-depth sensor (Sea-Bird 911 CTD), a fluorometer (Wetlabs ECO FLNTU), and a dissolved oxygen sensor (Sea-Bird SBE43) connected to the sampling rosette.

Dissolved inorganic phosphate, nitrite, nitrate, ammonium, and silicate concentrations were measured in filtered samples following standard photometric methods (Grasshoff et al., 1999), using a continuous-flow analyzer (FLOWSYS, Alliance Instruments). The samples were filtered through pre-combusted Whatman GF/F filters and stored at −20◦C until analysis. Ammonium was determined directly after sampling using a UV mini 1,240 photometer (Shimadzu). Precision was <10%, and detection limits were: nitrite 0.05 µM, nitrate 0.125 µM, ammonium 0.5 µM, phosphate 0.125 µM, silicate 0.5 µM. Dissolved organic nitrogen concentrations (DON) were calculated as difference between total dissolved nitrogen (TDN) and nitrate + nitrite + ammonium concentrations.

DOC and TDN were measured in the filtered and acidified (pH 2, HCl, p.a.) water samples by high temperature catalytic oxidation on a Shimadzu TOC-VCPH instrument. Analytical accuracy and precision were tested against the deep-sea reference sample (D. Hansell, University of Miami, USA) and were better than 5%.

DOM was extracted from 250 mL of filtered and acidified (HCl, pH 2, p.a.) water as described by Dittmar et al. (2008) using solid phase extraction (SPE) with cartridges filled with a styrene divinyl benzene polymer (Agilent Bond Elut PPL, 1 g). The methanol extracts were stored frozen in the dark until further analysis. The extraction efficiencies were determined as SPE-DOC divided by bulk DOC concentrations. SPE-DOC was determined by drying an aliquot of the SPE-DOM extract (at 40◦C) and re-dissolving it in ultrapure water.

The stable carbon isotope composition was determined on SPE-DOM. Extracts containing ca. 20 µg of DOC were pipetted into Sn caps (IVA, Germany) and subsequently dried at 40◦C in an oven. The isotopic composition was analyzed in duplicates on an isotope-ratio-monitoring mass spectrometer (Finnigan MAT 252, Bremen, Germany) via a Conflo II split interface. The stable carbon isotope ratios are reported in δ notation relative to the Vienna Pee Dee Belemnite. Precision and accuracy was <0.5‰ and procedural blanks did not yield detectable amounts of carbon isotope contamination.

For bacterial enumeration, 4 mL samples were fixed for 1 h with 400 µL of 1% (final concentration) paraformaldehyde and 0.5% (final concentration) glutaraldehyde, shock frozen in liquid nitrogen, and stored at −80◦C until processed by flow cytometry. Samples were measured on a FacsCalibur bench cytometer (Becton Dickinson) using a modification of the method of Gasol et al. (1999). Briefly, 0.2-µm-filtered SYBR Green solution (2.4 M potassium citrate, 0.2 M dimethyl sulfoxide and 5 µL SYBR Green) was mixed with 300 µL of the sample. The mixture was incubated for 30 min in the dark and measured for 3 min in a flow cytometer at a medium, previously determined flow rate. The diluted bead solution was microscopically evaluated against the measurements received by the flow cytometry. The flow diagrams were evaluated using the software CellQuestPro. Cells with chlorophyll autofluorescence (including cells of Synechococcus, picoeukaryotes, and nanoeukaryotes) were measured without SYBR Green staining based on autofluorescence in the FL2 (phycoerythrin) and FL3 (chlorophyll a) detectors of the FacsCalibur bench cytometer as described by Gasol et al. (1999).

### FT-ICR-MS Analysis

The methanol extracts were diluted with ultrapure water and methanol (MS grade) to yield a DOC concentration of 10 mg L −1 in 1:1 methanol:water (v/v) for the analysis with FT-ICR-MS (15 Tesla solariX, Bruker Daltonik). We used electrospray ionization (ESI) in negative mode. Instrument settings and molecular formulae assignments are described in detail in Seidel et al. (2014). The capillary voltage was 4 kV in negative mode. Ions were accumulated in the hexapole for 0.3 s and data were acquired in broadband mode using 4 megaword data sets and

a scanning range of 150–2,000 Da with 500 scans accumulated per mass spectrum. Mass spectra were calibrated internally with a list of known compounds in the targeted mass range (achieved mass accuracy <0.1 ppm). Molecular formulae were assigned to peaks with a signal-to-noise ratio >4 applying the criteria described by Koch et al. (2007) with a mass tolerance of <0.5 ppm. The signal intensities of peaks with assigned molecular formulae were normalized to the sum of all peak intensities with identified molecular formulae in each sample, and normalized peak intensities were multiplied by a factor of 10,000. The aromaticity and the degree of unsaturation of compounds were assessed using the modified aromaticity index (AImod; Koch and Dittmar, 2006, 2016). This parameter was developed to identify aromatic and condensed aromatic structures in DOM molecular formulae data acquired by ultrahigh-resolution MS. AImod values >0.5 are indicative for aromatic compounds, AImod values ≥0.67 are unambiguous indicators for condensed aromatic compounds. Intensity-weighted averages of AImod, number of atoms per molecular formula (carbon, C, hydrogen, H, oxygen, O, nitrogen, N, sulfur, S, and phosphorus, P), and molar ratios (hydrogento-carbon, H/C and oxygen-to-carbon, O/C) were calculated for each sample by considering the peak intensity of each assigned molecular formula as described in Seidel et al. (2014).

Molecular compound groups were assigned to molecular formulae as described in Seidel et al. (2014): group (1), polycyclic aromatics (PCAs, AImod >0.66), including condensed combustion-derived dissolved black carbon if C > 15 (Dittmar and Koch, 2006), (2) highly aromatic compounds with aliphatic side chains (HACs, Koch and Dittmar, 2006) (0.66 ≥ AImod ≥ 0.50), (3) highly unsaturated compounds (HUCs, AImod < 0.50 and H/C < 1.5), including lignin degradation products (Stenson et al., 2003) and carboxyl-rich alicyclic molecules (CRAM, Hertkorn et al., 2006), (4) unsaturated aliphatic compounds (2.0 ≥ H/C > 1.5, N = 0), (5) saturated compounds, including lipids (H/C > 2.0 and O/C < 0.9), (6) saturated compounds (H/C > 2.0) with O/C > 0.9, including carbohydrate molecular formulae, and (7) unsaturated aliphatic compounds containing N, including peptide molecular formulae (2.0 > H/C ≥ 1.5, N > 0). Further, we separated the molecular formulae according to their heteroatom content, i.e., (8) molecular formulae that contain N, (9) molecular formulae that contain S, (10) molecular formulae that contain P, and (11) molecular formulae that contain N and S. The molecular categorizing using ultrahighresolution MS provides information about likely structures behind the identified molecular formulae. The molecular structure assignment for molecular formulae is ambiguous because several different isomers may exist for a single molecular formula. However, the assignment of compound groups based on molecular level information obtained by FT-ICR-MS has been successfully used to identify biogeochemical processing of DOM (e.g., Kim et al., 2003; Schmidt et al., 2009). The proportion of each compound group is reported as the percentage of molecular formulae of a given compound group relative to the total number of molecular formulae detected in the sample.

### Statistical Analyses and Terrestrial DOC Flux Calculations

Principal coordinate analysis (PCoA) was performed on a Bray-Curtis dissimilarity matrix of the normalized peak intensities of all detected DOM molecular formulae. DOM molecular endmembers were identified using Ward's hierarchical clustering based on a Bray-Curtis dissimilarity matrix and PCoA. The DOM compound groups and environmental data (including DOC concentrations, stable carbon isotope composition of SPE-DOM, chlorophyll concentrations, concentrations of dissolved nitrogen species, silicate, phosphate, oxygen, salinity, and water temperature) were fitted post-hoc to the PCoA scores using the envfit function of the vegan package (Oksanen et al., 2015) within the R statistical platform (R core team, 2015). The correlation of environmental parameters to the DOM molecular composition (PCoA) was tested with 10,000 permutations and was considered significant if p < 0.1. Pearson correlations were used to assess the linear relationship between DOM compound groups and environmental parameters (α level 0.05). All contour plots and maps were produced using the Ocean Data View software (Schlitzer, 2016).

We determined how the concentrations of the measured nutrients, DOC, the <sup>13</sup>C isotopic composition of SPE-DOM and the relative abundance of DOM compound groups deviated from simple linear mixing of marine and terrestrial endmembers by using the correlation and regression analyses described previously by Seidel et al. (2015a). For the present study, we modified this approach by scaling the relative abundances of the molecular compound groups to the DOC concentrations (i.e., by multiplying the relative abundance of each compound group to the DOC concentration of each sample) before normalizing the data to values between 0 (minimum) and 1 (maximum). In short, this analysis gives a relative measure 1mix indicating the percentage that a measured parameter deviates from values expected from simple linear mixing of two endmembers (if significantly correlated to salinity, p ≤ 0.05) or averaged measured values (if endmembers had similar values). High 1mix values (>10%) describe a relatively high variability of the parameter in comparison to values expected from a linear mixing model or in relation to the average measured value along the salinity gradient. It is very important to note that this approach is not intended to describe the Baltic Sea as a simple two-endmember mixing model, i.e., a system where the terrestrial DOM from the northern basin rivers (e.g., Kalix or Neva Rivers) mixes with the marine DOM from the North Sea in the south. As mentioned before, almost two-thirds of the Baltic Sea riverine freshwater inflow occurs in the northern basins but additional terrestrial DOM is supplied to the southern Baltic Sea by rivers such as the Oder, Nemunas, and Vistula Rivers. Our mixing analysis was therefore intended as a firstorder approach to assess the dynamics of the DOM composition along the estuarine gradient when considering the two most extreme endmembers (sub-arctic terrestrial vs. marine DOM) of the Baltic Sea. In this analysis, deviations from the mixing of these endmembers can thus be due to removal or production of DOM compounds but also due to enrichment or depletion of the DOM compounds caused by to the input of freshwater with different DOC concentrations compared to the northern rivers along the salinity gradient.

The relative proportion of terrestrial SPE-DOM was calculated using δ <sup>13</sup>C SPE-DOM under consideration of terrestrial (δ <sup>13</sup>Cterr) and marine/autochthonous (δ <sup>13</sup>Cmarine) endmembers:

$$\text{terrestential SPEC-DOM } (\%) = \frac{\delta^{13} \text{C}\_{sample} - \delta^{13} \text{C}\_{marine}}{\delta^{13} \text{C}\_{terr} - \delta^{13} \text{C}\_{marine}} \times 100$$

The terrestrial and marine endmembers were determined by extrapolating the range of δ <sup>13</sup>C SPE-DOM values and DOC concentrations along the salinity gradient to salinities 0 and 35, respectively. Accordingly, the endmembers used for calculating the relative proportion of terrestrial SPE-DOM were salinity 0, δ <sup>13</sup>C SPE-DOM −28‰, DOC 400 µM and salinity 35, δ <sup>13</sup>C SPE-DOM −22‰, DOC 100 µM, respectively.

To extend our simplified mixing analysis, we also calculated the terrestrial DOC fluxes between the basins using a box model approach where we considered the in- and outflow for the basins of the Bothnian Bay, Bothnian Sea, Baltic Proper (including Gulf of Finland and Gulf of Riga), Danish Straits, and Kattegat area. The terrestrial DOC fluxes were calculated using the DOC concentrations (without PPL extraction), the relative proportions of terrestrial SPE-DOC from this study (with PPL extraction), and water flows between basins reported previously by Savchuk (2005). The net terrestrial DOC balance per basin (loss rate) was calculated as the difference of influx and outflux, i.e., the terrestrial DOC input by rivers (Gustafsson et al., 2014) and adjacent basins minus the outflow to adjacent basins. The residence times of terrestrial DOC were calculated for each basin by dividing the total stock of terrestrial DOC in each basin by all inputs of terrestrial DOC into the basin (sum of the inputs from adjacent basins and rivers). The reported values are basin averaged means ± standard deviations (of all concentration and isotope data determined in each basin). We note that for calculating fluxes and residence times of terrestrial DOC, we used the terrestrial SPE-DOC values (calculated percentage of terrestrial DOC using <sup>13</sup>C isotopic composition of DOC after PPL extraction) and the overall DOC pool (without PPL extraction). Although, PPL extraction has a poor recovery for very low molecular weight ionic compounds (Raeke et al., 2016), it provides a major fraction of the overall DOM (Green et al., 2014). As in previous studies (Seidel et al., 2015a,b; Osterholz et al., 2016), our extraction efficiencies were not correlated to salinity. We therefore assume that representative fractions of the marine and terrestrial DOM pools were extracted along the salinity gradient and that the terrestrial SPE-DOC values can be applied to infer the contribution of terrestrial DOC to the overall DOC pool.

#### RESULTS

#### General Water Characteristics

Temperatures were highest in the seawater entering from the North Sea through the Skagerrak and Kattegat. A steeper temperature gradient developed in spring compared with the temperature gradient in autumn (**Figures 2A,B**). Surface salinity increased from 2.7 to 26 (autumn) and 35 (spring) from the Kalix River outflow stations (Bothnian Bay) toward the marine water inflow at the Skagerrak and Kattegat. In the Bothnian Bay and Sea, salinity was <6 throughout the water column while in the Gotland Deep and Landsort Deep the water was more saline with salinities between 10 and 12 (**Figures 2C,D**). Dissolved oxygen concentrations were highest in the surface layers from the Bothnian Bay to the Skagerrak (6–10 mL L−<sup>1</sup> ). Oxygen concentrations were lower in the deep central Baltic Sea with hypoxic conditions (<1.4 mL L−<sup>1</sup> ) >100 m water depth (Gotland and Landsort Deeps). Anoxic conditions prevailed in >120 m (Gotland Deep) and >140 m depth (Landsort Deep), respectively (**Figures 2E,F**).

FIGURE 2 | Contour depth plots of the sampling transects. Water temperature (A,B), salinity (C,D), dissolved oxygen concentrations (E,F), concentrations of chlorophyll a (G,H), phosphate (I,J), and silicate (K,L) are shown for autumn (left) and spring (right) sampling. Non-linear depth scale. Skag., Skagerrak; Kat., Kattegat; Dan. Str., Danish Straits; Both. Sea/Bay, Bothnian Sea and Bothnian Bay.

#### Inorganic Nutrients and Cell Numbers

Chlorophyll a concentrations were highest in the surface water of the Kattegat area and the southern Baltic Proper (stations S4 to TF245, **Figure 1**) with values between 0.6 and 1 mg m−<sup>3</sup> in autumn (**Figure 2G**). In spring, chlorophyll a concentrations were high in the surface waters of the Bothnian Sea (1.8 mg m−<sup>3</sup> at station At4) and the southern Baltic Proper (1.3 mg m−<sup>3</sup> at station Mo5) and increased to values of up to 3.5 mg m−<sup>3</sup> at the Landsort Deep (station TF284, **Figure 1**) at 14 m depth (**Figure 2H**). For the autumn sampling campaign, the number of primary producers was also assessed by enumeration of cells with chlorophyll autofluorescence (including cells of Synechococcus, picoeukaryotes, and nanoeukaryotes, Supplementary Figure 1A). These data matched closely the chlorophyll data that were assessed using the fluorescence detector connected to the sampling rosette, i.e., highest concentrations of primary producers were found in the southern Baltic Proper (Supplementary Figure 1A). In addition, the concentration of prokaryotes was assessed by flow cytometry. Prokaryotic cell numbers were highest in the surface waters of the southern Baltic Proper, in the Bothnian Bay and in the hypoxic/anoxic water masses of the Gotland Deep and Landsort Deep (Supplementary Figure 1B).

Phosphate and silicate concentrations were highest in the anoxic Gotland and Landsort Deeps (**Figures 2I–L**). Phosphate concentrations in the Landsort Deep were up to 5.1 µM (spring) and 6.5 µM (autumn) and in the Gotland Deep up to 3.9 µM (spring) and 3.5 µM (autumn). Silicate concentrations in the Gotland Deep were up to 97 µM (spring) and 93 µM (autumn) and in the Landsort Deep up to 53 µM (spring) and 60 µM (autumn). Phosphate concentrations were ≤0.5 µM (below detection limit) in autumn and spring in the surface samples throughout the Baltic Sea (**Figures 2I,J**). Silicate concentrations were elevated in the surface waters of the Bothnian Bay (37– 40 µM and 35–43 µM in autumn and spring, respectively) but depleted in the surface samples of the Baltic Proper (4.2–13 µM and 0–15 µM in autumn and spring, respectively, **Figures 2K,L**).

TDN concentrations in autumn were high at the surface in the Bothnian Bay (station Mo14 with 30 µM TDN) while in the Baltic Proper, surface TDN concentrations were lower with 14–18 µM (autumn, Supplementary Figure 2A) and 12– 18 µM (spring, Supplementary Figure 2B). TDN concentrations were highest in the bottom water of the Gotland Deep with 55 and 60 µM in spring and autumn, respectively (Supplementary Figures 2A,B). In the bottom water of the Landsort Deep, TDN concentrations were 21 µM (spring) and 22 µM (autumn, Supplementary Figures 2A,B), respectively. Nitrate concentrations were highest in the Bothnian Bay with values of up to 8.2 µM at station S11 in autumn (**Figure 3A**) and 6.4 µM in spring (**Figure 3B**). In the Baltic Proper, nitrate concentrations decreased to levels below the detection limit (0.125 µM) in the surface samples (**Figure 3B**). Nitrite concentrations were generally low (<0.5 µM) with the highest values during autumn (Supplementary Figures 2C,D). Ammonium concentrations were highest in the anoxic waters of the Gotland and Landsort Deeps (**Figures 3C,D**). The ammonium concentrations positively correlated with silicate and phosphate concentrations in autumn (Supplementary Figure 3) and spring (Supplementary Figure 4). Ammonium concentrations were lower in spring compared to autumn (Landsort Deep 6.4 µM compared to 11 µM, and Gotland Deep 11 µM compared to 31 µM in **Figures 3C,D**). DON concentrations were high in the Baltic Proper (surface concentrations 15–17 µM and 9.2–17 µM in autumn and spring, respectively, **Figures 3E,F**). High DON concentrations were also found in the bottom water of the anoxic Gotland (29 µM in autumn and 28 µM in spring) and Landsort Deeps (11 µM in autumn and 15 µM in spring, **Figures 3E,F**). The TDN concentrations positively correlated to DON concentrations in autumn (Supplementary Figure 3) and spring (Supplementary Figure 4) while DON also positively correlated with ammonium concentrations in spring (Supplementary Figure 4).

## DOC Concentrations and Isotopic Composition

DOC concentrations were highest in the Bothnian Bay and Bothnian Sea close to the Kalix River outflow with 379 µM during the spring flood (Bothnian Bay station At4, surface, **Figure 1**). During autumn, highest DOC concentrations occurred in the Baltic Proper and the Bothnian Sea (301–341 µM from station Mo7 to S10, surface to 70 m depth) while in the Bothnian Bay, DOC concentrations were 304–308 µM (stations S11 and Mo14, 11–41 m depth, **Figures 1**, **4A**). Toward the Kattegat (station S4), DOC concentrations decreased to 187 µM in autumn (**Figure 4A**). In spring, DOC concentrations decreased to 118 µM in the Skagerrak (station S1, **Figures 1**, **4B**).

For both sampling campaigns, samples from the Bothnian Bay were characterized by depleted δ <sup>13</sup>C SPE-DOM signatures (between −27.9 and −27.1‰ in autumn at stations Mo14 and S11, and between −27.4 and −26.7‰ in spring at stations At4 and S11). Samples from the North Sea and Kattegat were more enriched (−25.2‰ in autumn at station S3, and −23.2‰ in spring at station S1, **Figures 1**, **4C,D**). The isotopic composition of samples from the central Baltic Proper were between −27.0 and −25.2‰ (autumn) and −27.3 and −25.3‰ (spring, **Figures 4C,D**). The relative abundance of terrestrial SPE-DOC ranged from almost 100% close to the Kalix River outflow in the Bothnian Bay and decreased to ca. 75% in the Bothnian Sea (**Figures 4E,F**). Within the Baltic Proper basin, terrestrial SPE-DOC was uniformly distributed with values between 70 and 75%. A decrease to 53% was found in the Kattegat bottom water (inflowing North Sea water) while the surface water (outflowing Baltic Sea water) contained 69% (autumn) and 78% (spring) terrestrial SPE-DOC, respectively. The most marine station at Skagerrak in the North Sea (station S1 in **Figure 1**, salinity 34.9 in spring) contained 5.5% terrestrial SPE-DOC in the bottom and 24% terrestrial SPE-DOC in the surface water (**Figure 4F**).

## Distribution of DOM Molecular Compound Groups

SPE-DOM extraction efficiencies were 77 ± 10% (n = 51, whereby few samples with extraction efficiencies above 90 and below 40% were not considered because they were likely

and Bothnian Bay.

erroneous). Extraction efficiencies were not correlated to salinity (p > 0.05, r = −0.27, n = 51). This supports our assumption that there was no preferential extraction of terrestrial over marine compounds from the DOM that is extracted with the PPL method. In total, ca. 6,000 molecular formulae were identified in the SPE-DOM. From the Kalix River outflow toward the Bothnian Sea, most molecular formulae were highly aromatic and highly unsaturated molecular formulae (groups 1–3 were up to 90% of all molecular formulae, Supplementary Figures 5A–F). Condensed and highly aromatic compounds (groups 1 and 2) were most abundant in the Bothnian Bay and Bothnian Sea, decreasing from 7.2% (group 1) and 16% (group 2) to 4.5% (group 1) and 14% (group 2) in the Baltic Proper. In this area, also the highest percentage of aromatic compounds, an indicator for terrestrial DOM, was found (weighted-average AImod in **Figures 5A,B**). Accordingly, salinity and δ <sup>13</sup>C SPE-DOM values were negatively correlated to condensed aromatics (group 1, in autumn and spring) or highly aromatic compounds (group 2, in spring only) (Supplementary Figures 3, 4). The relative abundances of highly unsaturated compounds (group 3, Supplementary Figures 5E,F) and unsaturated aliphatics (group 4, Supplementary Figures 5G,H) were highest in the marine inflow area of the Kattegat with values of up to 72% (group 3) and 15% (group 4), respectively. The relative abundance of saturated compounds (group 5, Supplementary Figures 5I,J) was highest in autumn in the marine inflow waters at the Kattegat (up to 1.4%), while in spring, values were low throughout the Baltic Sea (<0.6%). Saturated compounds with high O/C ratios (group 6) were close to zero and were therefore excluded from further analyses. Unsaturated compounds containing N (group 7, including peptide molecular formulae, Supplementary Figures 5K,L) contributed up to 1.2% throughout the Baltic Sea in autumn and were <1% in spring. In our regression analyses, all compound groups (scaled to DOC concentrations), apart from saturated compounds (group 5), were negatively correlated to salinity in autumn (p < 0.05, Supplementary Figure 6). In spring, compound groups 1–4 negatively correlated with salinity whereas groups 5–11 did not (Supplementary Figure 6). Group 7 compounds positively correlated with DON concentrations in spring (Supplementary Figure 4), when DON concentrations were highest in the anoxic Landsort and Gotland Deeps (**Figure 3F**).

N-containing compounds (group 8) were most abundant in the marine inflow water at the Kattegat (up to 26% of all molecular formulae; **Figures 5C,D**). S-containing formulae (group 9, **Figures 5E,F**) were abundant in the marine inflow water (up to 22% of all molecular formulae) and in the anoxic

bottom water of the Gotland Deep and Landsort Deep (18–22% of all molecular formulae). P-containing formulae (group 10) were most abundant in the marine inflow water in autumn (up to 6% of all molecular formulae; **Figures 5G,H**).

Total DOC concentrations negatively correlated with salinity (p < 0.01) in spring (r = −0.70, n = 41) and autumn (r = −0.75, n = 36; **Figure 6A**). Likewise, terrestrial DOC concentrations (based on δ <sup>13</sup>C SPE-DOM) negatively correlated with salinity in both seasons (p < 0.01; Supplementary Figure 6). The DOCconcentration-scaled relative abundances of highly aromatic compounds (group 2, **Figure 6B**) and unsaturated aliphatic compounds (group 4, **Figure 6C**) were also negatively correlated to salinity (p < 0.01). In spring, the DOC-concentration-scaled relative abundances of unsaturated aliphatics (compound group 4) deviated more from linear mixing compared to autumn (**Figure 6C**).

#### Statistical Analyses

In PCoA analysis, the samples were correlated to eight (autumn, **Figure 7A**) and nine (spring, **Figure 7B**) environmental parameters (p < 0.1). In the ordination plots, the first two axes explained between 68 and 74% of the DOM molecular variability (spring and autumn, respectively). PCoA and cluster analyses revealed three DOM molecular clusters that generally followed the salinity and oxygen gradients, i.e., (1) terrestrial-oxic, (2) brackish-to-marine oxic, and (3) brackish suboxic-to-anoxic (**Figure 7**). In PCoA, the projections of sampling points onto the vectors depict correlations with the corresponding environmental parameters. In autumn and spring, the identified clusters correlated with salinity, δ <sup>13</sup>C SPE-DOM, water temperature (brackish-to-marine oxic cluster, blue circles in **Figure 7**), DOC, nitrate and dissolved oxygen concentrations (terrestrial-oxic cluster, green circles in **Figure 7**), as well as depth and phosphate concentrations (suboxic-toanoxic cluster, red circles in **Figure 7**). In spring, samples from the brackish-to-marine samples also correlated with DON concentrations (**Figure 7B**).

In autumn, marine-influenced samples correlated with the percentage of DOM groups 4, 5, 8, 9, and 10 (unsaturated aliphatics, saturated, N-, S-, and P-containing compounds, **Figure 7A**). Terrestrially influenced samples correlated with groups 1, 2, 3, 7, and 11 (polycyclic aromatic, highly aromatic, highly unsaturated, unsaturated aliphatic with N, and NS-containing compounds, **Figure 7A**). In spring, marineinfluenced samples correlated with DOM compound groups 3, 4, 7, 8, 9, 10, and 11 (highly unsaturated, unsaturated aliphatic with/without N, saturated, N-, S-, P-, and NS-containing compounds, **Figure 7B**). More terrestrial influenced samples

FIGURE 5 | Contour depth plots of the sampling transects. Weighted averages of DOM aromaticity (AImod; A,B) and compound groups of DOM containing nitrogen (C,D), sulfur (E,F), or phosphorus (G,H) (DOM groups 8, 9, and 10) are shown for autumn (left) and spring (right) sampling. Averages are weighted by FT-ICR-MS signal intensity. Non-linear depth scale. Skag., Skagerrak; Kat., Kattegat; Dan. Str., Danish Straits; Both. Sea/Bay, Bothnian Sea and Bothnian Bay.

FIGURE 6 | DOC concentrations and the distribution of DOM molecular compound groups along the salinity gradient. Relationship of bulk DOC concentrations (A), the relative abundance of highly aromatic compounds (B) and unsaturated aliphatic compounds (C; molecular formulae of groups 2 and 4, respectively) and salinity for autumn (blue squares) and spring (red circles) samples. The solid lines represent linear regressions. The relative abundance of the compound groups was scaled to the DOC concentrations and normalized to their minimum and maximum values. Bulk DOC concentrations and the highly aromatic compounds strongly correlated with salinity indicative for mixing of riverine and marine DOM. Unsaturated aliphatic compounds were more variable compared to linear mixing suggesting complex processing along the salinity gradient.

FIGURE 7 | Principal coordinate analyses (PCoA) based on Bray-Curtis dissimilarities of the relative abundance of DOM molecular formulae. Shown are the PCoA for autumn (A) and spring (B) samples, respectively. The percentages give the DOM molecular variability as explained by the axes. Colored circles identify the three main DOM clusters that were found in terrestrial-oxic (green circles), marine-to-brackish oxic (blue circles), and brackish suboxic-to-anoxic conditions (red circles). Environmental parameters (significantly correlated with p ≤ 0.1, p ≤ 0.05, and p ≤ 0.01 blue, red, and dark red arrows, respectively), DOM compound groups (groups g1–g10 specified in legend shown in B; black arrows) and intensity weighted averages of O/C or H/C ratios and masses ("O/C," "H/C," "Mass") and average number of molecular formulae per sample ("Formulae") were fitted onto the ordination. Correlations with compound group 6 (saturated compounds with high O/C ratios) were not significant and are therefore not shown. The projections of sampling points onto the vector arrows show maximum correlations with the corresponding molecular compound groups, environmental, and molecular parameters.

correlated with DOM groups 1 and 2 (polycyclic aromatic and highly aromatic compounds, **Figure 7B**).

### Comparison of DOM Molecular Endmembers

The characteristic DOM molecular signatures of the terrestrial, marine, oxic, and anoxic water masses were described by calculating the molar averages of the enriched molecular formulae of the endmembers that were identified in PCoA and cluster analyses. These were the samples from stations Mo14 and At-4 (Bothnian Bay surface close to the Kalix River outflow; terrestrial endmembers), S4 (19 m depth at Kattegat), and S1 (surface close to Skagerrak; marine endmembers) and TF284 (Landsort Deep at 430 m depth; anoxic endmembers) for the autumn and spring cruises, respectively (see **Figure 1** for sampling positions). Molecular formulae enriched in terrestrial DOM had higher masses, more carbon and more oxygen atoms, higher O/C and lower H/C ratios than DOM from marine or from anoxic water masses (**Table 1**). Enriched marine and DOM from anoxic water masses contained 7- to 33-fold more nitrogen, sulfur, and phosphorus (N-, S-, and P-) atoms compared to terrestrial DOM (**Table 1**). In spring, N-atoms in marine compounds were 2-fold higher compared to autumn (**Table 1**). Marine water masses in autumn were almost twice as much enriched in S-atoms compared to anoxic DOM compounds. In spring however, S-atoms were more enriched in anoxic DOM compared to marine DOM (**Table 1**). The S-content of marine DOM varied between the seasons with higher values in autumn compared to spring. Molecular formulae relatively enriched in the terrestrial endmember water in the Bothnian Bay were mainly polycyclic aromatics, highly aromatic, and highly unsaturated compounds (groups 1–3 were 97% of the enriched compounds, Supplementary Table 1). Unsaturated aliphatic compounds, with or without nitrogen and saturated compounds (groups 4–7), contributed comparatively more to the molecular formulae enriched in the marine and anoxic DOM samples (15–32% in spring and 13–22% in autumn, Supplementary Table 1).

We characterized molecular formulae that were relatively enriched in the seawater bottom inflow in 17–19 m depth at station S4 (i.e., marine compounds that are transported from the North Sea to the Baltic Sea) and brackish surface outflow at station S4 (i.e., terrestrial compounds that are exported from the Baltic Sea to the North Sea; **Table 2**) through the Kattegat and Skagerrak (station S4 in **Figure 1**). The outflowing water had a higher number of molecular formulae than North Sea water entering the Baltic Sea ("formulae" in **Table 2**). The outflowing water was relatively enriched with molecular formulae with higher masses and more carbon atoms while the inflowing water was enriched with molecular formulae containing nitrogen, sulfur and phosphorus heteroatoms. These trends were more pronounced in spring compared to autumn.

#### Terrestrial DOC Fluxes

The relative proportion of terrestrial DOC decreased from 90 ± 3% and 86 ± 5% in the Bothnian Bay to 60 ± 8% and 66 ± 17% in the western Baltic Sea at the Skagerrak/Kattegat area (stations S1, S3, and S4 in **Figure 1**) in autumn and spring, respectively (**Table 3**). Using our box model approach, we estimated a loss of terrestrial DOC in the Baltic Sea (sum of values from Bothnian Bay to Danish Straits) of 1.6–1.9 Tg C yr−<sup>1</sup> (43–51% of total riverine input). The Baltic Proper (including terrestrial DOC inputs from the Gulf of Riga and Gulf of Finland) was the biggest sink for terrestrial DOC with 1.8 ± 0.8 Tg C yr−<sup>1</sup> (autumn) and 2.5 ± 0.9 Tg C yr−<sup>1</sup> (spring) net removal. Removal of

TABLE 1 | Characteristics of DOM molecular formulae (means of molecular parameters ± SD) that were relatively enriched in the terrestrial (riverine sources), marine (autochthonous sources), and anoxic (deep basins) end members in autumn and spring as identified by principal coordinate and cluster analyses.


<sup>a</sup>Number of enriched molecular formulae ("Formulae"), intensity-weighted means of molecular masses of enriched molecular formulae ("Mass," in Dalton), intensity-weighted means of the number of atoms per molecular formula (C, H, O, N, S, P), molar ratios (O/C, H/C), and of aromaticity index (AImod).

<sup>b</sup>Stations Mo14 and At-4 (surface of Bothnian Bay close to Kalix River outflow; terrestrial DOM endmembers), S4 (19 m depth in Kattegat) and S1 (surface close to Skagerrak; marine DOM endmembers) and TF284 (Landsort Deep in 430 m depth; anoxic DOM endmembers) were used for the autumn and spring samples, respectively. See Figure 1 for sampling positions and Figure 7 for distribution of clusters in PCoA.

terrestrial DOC in the Bothnian Bay was 0.3 ± 0.1 Tg C yr−<sup>1</sup> (autumn) and 0.3 ± 0.3 Tg C yr−<sup>1</sup> in spring (**Table 3**). The terrestrial DOC balances in the Bothnian Sea and the Danish

TABLE 2 | Characteristics of DOM molecular formulae (mean of molecular parameters ± SD) that were enriched in inflowing bottom water from the North Sea (into the Baltic Sea) and outflowing surface water from the Baltic Sea (into the North Sea) through the Kattegat area.


<sup>a</sup>Number of enriched molecular formulae ("Formulae"), intensity-weighted mean of molecular masses of enriched molecular formulae ("Mass," in Dalton), intensity-weighted means of the number of atoms per molecular formula (C, H, O, N, S, P), molar ratios (O/C, H/C), and of aromaticity index (AImod).

b Inflowing North Sea bottom water (17–19 m depth) and outflowing Baltic Sea surface water at station S4 (Kattegat, see Figure 1 for sampling positions).

Straits were slightly negative or close to zero (**Table 3**). The export of terrestrial DOC from the Danish Straits into the Kattegat area was between 1.8 and 2.1 Tg C yr−<sup>1</sup> in autumn and spring, respectively. The fluxes were slightly higher in spring compared to autumn. The residence times of terrestrial DOC were 3.7–3.8 years in the Bothnian Bay, 3.2–3.4 years in the Bothnian Sea, and 4.5–4.9 years in the Baltic Proper (**Table 3**). The residence time of terrestrial DOC in the Danish Straits and Kattegat area was 0.3–0.4 years (**Table 3**).

#### DISCUSSION

### Bulk Geochemical and Biological Parameters

The distribution of the chemical constituents in the Baltic Sea water was largely driven by mixing of marine and riverine water masses and by the vertical redox zonation of the water column. The thermal stratification was weaker in autumn compared to spring (water temperature in **Figures 2A,B**), indicating enhanced convective overturning in the surface mixed layer. Reduced stratification was probably due to a combination of stronger wind-induced mixing, surface cooling as well as reduced inflow of riverine freshwater in autumn (Nausch et al., 2012). In winter, the water column of the Baltic Sea is vertically mixed down to the halocline, whereas in spring, a seasonal thermocline generally forms at about 20 m depth (Eilola and Stigebrandt, 1998). Water masses below the halocline in the deep stagnant basins is only replaced during major inflow events, which have not occurred between 2004 and 2014 (Mohrholz et al., 2015), i.e.,



<sup>a</sup>Riverine DOC input per basin from Gustafsson et al. (2014); Baltic Proper includes Gulf of Finland and Gulf of Riga.

<sup>b</sup>Terrestrial DOC balance per basin; positive (negative) numbers indicate removal (input) of terrestrial DOC compared to the terrestrial DOC input from adjacent rivers and basins; NA, not available.

<sup>c</sup>Average residence time (in years, yrs) of terrestrial DOC (terr. DOC) for each basin.

<sup>d</sup>Average or sum of values from Bothnian Bay to Danish Straits.

\*Residence times for the combined Danish Straits and Kattegat area.

during and around the time our samples were taken (2011 and 2012).

Nitrogen is provided to the Baltic Sea by riverine and atmospheric inputs, nitrogen fixation and nutrient recycling within the water column (Thomas et al., 2010). During the autumn months, stronger wind-induced mixing provides the surface and subsurface mixed layers with recycled nutrients from below the thermocline. At the same time in late autumn, the overall phytoplankton biomass (as indicated by chlorophyll a concentrations, **Figures 2G,H**) was lower. Consequently, higher depth-integrated nitrate concentrations were found in the Bothnian Bay in autumn compared to spring (**Figures 3A,B**).

Nitrate (**Figures 3A,B**) and silicate concentrations (**Figures 2K,L**) were higher in the surface layers of the Bothnian Sea compared to the central Baltic Sea, where these nutrients were depleted in the photic zone, probably due to phytoplankton growth. Silicate and nitrate were still detectable in the areas with high freshwater input such as the Bothnian Bay, where rivers are important nutrient sources (Savchuk, 2005). Dissolved silica and dissolved phosphate are mainly derived from riverine inputs and they are removed as biogenic particles by growth of phytoplankton such as diatoms (Rahm and Danielsson, 2007; Conley et al., 2008). Sinking particles export silica, nitrogen and phosphorus into the deeper water layers where the particles settle and partly dissolve (Struck et al., 2004; Savchuk, 2005). Below the halocline in the stagnant anoxic Gotland and Landsort Deeps, dissolved silicate (**Figures 2K,L**), phosphate (**Figures 2I,J**), and ammonium (**Figures 3C,D**) can accumulate (Nausch and Nehring, 1996; Nausch et al., 2012).

The DOC concentrations of the terrestrial endmembers in the Bothnian Bay were slightly elevated in spring (**Figures 4A,B**). This was likely due to the flushing of soils and fast response to precipitation events and snow melt. In general, boreal and sub-arctic river systems are characterized by a pronounced seasonality of their hydrograph, because of elevated spring peak flow due to snowmelt (Smedberg et al., 2006). But considering the long residence time of terrestrial DOC in the various basins of the Baltic (see Discussion below), the effect of seasonal runoff is dampened on a basin scale. DOC concentrations strongly correlated with salinity during both seasons but with noticeable deviations from two-endmember mixing (**Figure 6A**). Positive deviations of DOC concentrations indicate release of DOM during the spring phytoplankton bloom whereas negative deviations suggest degradation of terrestrial DOC. Similar to bulk DOC, the concentration of terrestrial DOC was largely driven by dilution with seawater along the salinity gradient, but noticeable deviations from simple mixing indicate processing of terrestrial DOM along the salinity gradient (Supplementary Figure 6). It is important to keep in mind that in our simplified linear mixing model, we assumed that all terrestrial and marine sources in the Baltic Sea have the same endmember characteristics in terms of stable carbon isotopes and DOC concentrations. This may not be the case, because the various rivers draining into the Baltic Sea differ, especially in their DOC concentrations (e.g., Deutsch et al., 2012). This adds uncertainty to our two-endmember mixing model.

Nonetheless, our considerations are consistent with previous findings. On average, 20–60 µM particulate organic carbon is seasonally produced by phytoplankton in the Baltic Proper during the spring and summer blooms (Nausch et al., 2012), a part of which is likely released in dissolved form. Removal of terrestrial DOM was previously observed in the Baltic Sea (Alling et al., 2010; Deutsch et al., 2012). The bioavailability of terrestrial DOM depends strongly on the catchment area (Asmala et al., 2013; Riedel et al., 2016). However, a more significant sink for terrestrial DOM in the Baltic Sea, and coastal oceans in general, is probably photo-mineralization and bacterial uptake of the photoproduced labile DOM (Miller and Zepp, 1995; Aarnos et al., 2012). In fact, Aarnos et al. (2012) calculated that the total annual photochemical mineralization exceeds the annual input of photoreactive riverine DOM, suggesting that also autochthonous DOM is photo-transformed in the Baltic Sea. The biodegradation of terrestrial DOC is likely to be significant on time scales of weeks to months due to the semi-labile to refractory character of terrestrial DOC that is exported to the Baltic Sea (Herlemann et al., 2014; Kulinski et al., 2016 ´ ). The net heterotrophy that was observed in parts of the Baltic Sea suggests that a part of the terrestrial DOC pool is important for supporting bacterial mineralization (Algesten et al., 2006).

## DOM Composition in Relation to Marine and Terrestrial Sources

In the riverine-influenced samples, we found a stable carbon isotopic composition of SPE-DOM that is typical for C3 land plants, i.e., between −28 to −25‰, whereas the marineinfluenced samples were in the range for marine phytoplankton OM, i.e., around −22‰ (Benner et al., 1997; Guo et al., 2003). C3 plants, such as woody gymnosperms, are a main source for the terrigenous DOM that is discharged into the northern Baltic Sea (Bianchi et al., 1997; Deutsch et al., 2012). This is supported by our finding that most of the riverine DOM consisted of highly aromatic and highly unsaturated molecular formulae, which include lignin degradation products (Stenson et al., 2003), polyphenols (Koch and Dittmar, 2006), but also combustion-derived compounds (Dittmar and Koch, 2006). Molecular formulae with a comparatively low hydrogen content (low H/C ratios) are a general feature of terrestrial DOM due to the high content of aromatic compounds such as degraded lignins and tannins (Sleighter and Hatcher, 2008; Schmidt et al., 2009) and mobilized black carbon from soils (Dittmar et al., 2012; Jaffé et al., 2013).

The non-targeted DOM molecular analyses using ultrahighresolution MS allows to distinguish different DOM sources and complex transformations in estuaries and river plumes (Medeiros et al., 2015a,b; Seidel et al., 2015b; Osterholz et al., 2016). Using the PCoA multivariate statistical approach, we identified the molecular composition of the endmembers of the autochthonous marine and allochthonous terrestrial sources, as well as oxic-to-anoxic transformations in the Baltic Sea. The terrestrial endmembers had a higher number of molecular formulae (508–788 enriched terrestrial compounds compared to 163 and 197 enriched compounds in the marine endmembers, **Table 1**) and contributed more aromatic compounds as indicated by the higher aromaticity (higher AImod, **Figures 5A,B**) compared to the marine DOM endmembers. The marine endmembers were enriched with saturated compounds (lower AImod values) compared to the terrestrial endmembers (**Table 1**). The saturation of DOM compounds increased also along other river-to-ocean transects (Sleighter and Hatcher, 2008; Medeiros et al., 2015b; Seidel et al., 2015b). This is because marine DOM is enriched in aliphatic structures compared to freshwater samples that are enriched in aromatic compounds. Our data are consistent with these previously observed trends.

The molecular formulae that were enriched in the marine endmembers were characterized by higher contents of N-, S-, and P-atoms compared to the enriched terrestrial molecular formulae (**Table 1**). A higher proportion of DOM from phytoplankton and microbial biomass that is enriched in N, P and S-heteroatoms, such as peptides, nucleotides, and lipid degradation products, have also been previously observed in marine DOM (Kujawinski et al., 2004; Sleighter and Hatcher, 2008; Schmidt et al., 2011; Gonsior et al., 2011a; Seidel et al., 2015a). For example, in the Amazon plume and Delaware estuary, the increase of aliphatic compounds was related to addition of DOM from phytoplankton and the biotransformation of terrestrial DOM (Medeiros et al., 2015b; Osterholz et al., 2016). In line with these previous findings, our data demonstrate that the autochthonous production of DOM in the Baltic Sea and possibly also biotransformation of terrestrial compounds are a source for aliphatic N-, S-, and Pheteroatom containing compounds to the estuarine DOM pool.

The seasonal changes between the enriched compounds of the terrestrial endmembers were comparatively small (**Table 1**). The terrestrial DOM endmembers were from stations close to the Kalix River outflow in the Bothnian Bay (**Figure 1**). This suggests that at this location the seasonal DOM molecular changes were relatively minor and that the input of mobilized land-plant derived material was the main source of riverine DOM that was transported to the Bothnian Bay. In contrast, the changes between the marine endmembers and the anoxic endmembers were more pronounced. For example, during the spring phytoplankton bloom, the average N-content was 2-fold higher in the marine DOM compared to autumn (**Table 1**). As mentioned previously, the higher N- content in the molecular formulae can be attributed to DOM from autochthonous (marine) sources (e.g., Sleighter and Hatcher, 2008), particularly in spring, with higher phytoplankton biomass.

A previous study has shown that (chromophoric) DOM is transported from the North Sea (German Bight) into the Baltic Sea via the Jutland Coastal current through the Kattegat area (Stedmon et al., 2010). The sulfidic porewater in the tidal flats of the North Sea is strongly enriched with sulfurized DOM and the porewater discharge is a DOM source to the water column (Seidel et al., 2014). The Skagerrak/Kattegat area (**Figures 5E,F**), the marine endmember samples (**Table 1**), and the DOM from seawater flowing into the Baltic Sea (**Table 2**) were relatively enriched with S-containing DOM (particularly during autumn, when the phytoplankton biomass was lower compared to spring). This may indicate that the S-rich DOM from the North Sea tidal flats is transported into the Skagerrak/Kattegat area. Another possible source for the S-containing compounds is the discharge from the more anthropogenically influenced rivers of the southwestern Baltic Sea. Compared to the less populated subarctic north, more dissolved organic sulfur compounds may be introduced by wastewater (Gonsior et al., 2011b; Wagner et al., 2015) in the densely populated southwestern basins.

The distribution of the molecular DOM compound groups along the salinity gradient was strongly influenced by mixing (**Figure 6**; Supplementary Figure 6). However, similar to bulk and terrestrial DOC concentrations, there was considerable scatter around the mixing lines, which indicates active processing of DOM in the Baltic Sea. Especially DOM compound groups that were identified as typical marine (compounds groups 5–11) showed largest deviations from conservative mixing in spring, which can be associated to autochthonous production in the spring blooms. The high abundance of aromatic compounds makes terrestrial DOM highly susceptible to photochemical degradation (Hernes and Benner, 2003; Gonsior et al., 2009; Stubbins et al., 2010) which would explain the observed depletion of the aromatic compound groups (groups 1 and 2) at mid salinity in both seasons.

During its transit through the Baltic Sea, DOM became more enriched in autochthonous (phytoplankton-produced) compounds, which was more pronounced in spring compared to autumn (**Table 1**). A net accumulation of more aliphatic compounds (**Figure 6C**) and marine DOC (**Figure 4F**) was observed in spring, which is likely due to new addition of OM compounds during the phytoplankton bloom. In combination with the quantitative DOC data and the analysis of the <sup>13</sup>C isotopic composition of SPE-DOM, we found indications for processing of terrestrial DOC in the Baltic Sea. A general trend for the Baltic Sea may be inferred: marine DOC accumulates due to the high phytoplankton primary production in spring whereas net (terrestrial) DOC removal occurs in autumn (without a noticeable accumulation of marine DOC). Thus, our data suggest that the carbon budget (here as DOC depletion/accumulation) can shift between net CO<sup>2</sup> uptake (conversion to OM) and net release (remineralization of OM to CO2) which varies between the basins but also seasonally (Thomas et al., 2010; Kulinski and ´ Pempkowiak, 2011; Gustafsson et al., 2014; Ylöstalo et al., 2016). In line with that, Algesten et al. (2006) previously suggested net heterotrophy in the Gulf of Bothnia because of terrestrial DOC processing.

## DOM Transport from the Baltic Sea to the North Sea

The riverine inflow from the northern Bothnian Bay (e.g., Kalix River) and Bothnian Sea provide ca. 1.2 Tg terrestrial DOC per year (Gustafsson et al., 2014). For that area, our box model calculations resulted in terrestrial DOC removal rates of up to 0.3 Tg C per year (**Table 3**). Rivers draining into the Baltic Proper, the Gulf of Finland (Neva River) and the Gulf of Riga contribute ca. 2.4 Tg terrestrial DOC per year (Gustafsson et al., 2014). In that area, terrestrial DOC removal rates were between 1.8 and 2.5 Tg C per year. Overall, we estimated that 43–51% of the total riverine input of the 3.7 Tg terrestrial DOC is removed between the Bothnian Sea and the Danish Straits/Kattegat area. The fact that we estimated zero and even negative net balances for some of the lower basins highlights that their carbon budget may be in close balance, but it also emphasizes that the terrestrial DOC budgets (such as the DOC input terms) are still not well-constrained.

Our calculated export of 1.8–2.1 Tg terrestrial DOC per year to the Danish Straits (for autumn and spring, respectively) is comparable to the previously reported net flux of ca. 3 Tg total organic carbon per year by Gustafsson et al. (2014). It is, however, higher than the 0.8 Tg C per year which was based on optical measurements of chromophoric DOM (Osburn and Stedmon, 2011). In a previous study, Deutsch et al. (2012) found that the relative proportion of terrestrial HMW DOM decreased to 43% which is in line with our data. We calculated residence times for terrestrial DOC (i.e., the time that terrestrial DOC spends on average in each basin) of 3.7–4.9 years (**Table 3**) which also agrees well with previously reported values (Alling et al., 2008; Deutsch et al., 2012; Gustafsson et al., 2014). The water residence times (5.1, 3.2, and 4.4 years for Bothnian Bay, Bothnian Sea, and Baltic Proper, respectively; Savchuk, 2005) are higher or in the same range, indicating removal of terrestrial DOC on the time scales of water transport through the Baltic Sea. Yet, a significant fraction of the terrestrial DOC was resistant to removal in the Baltic Sea and was transported to the North Sea and presumably to the Atlantic Ocean. The residence time of terrestrial DOC in the Danish Straits and Kattegat area is relatively short (0.3–0.4 years, **Table 3**), probably because this area is relatively shallow and more mixed compared to the deeper basins of the Baltic Sea. However, the similar fluxes of terrestrial DOC to the North Sea in spring and autumn suggest that seasonal changes of OM production and riverine discharge are dampened. We propose that this is due to the long residence time of terrestrial DOC in the Baltic Sea.

After passage through the Baltic Sea, the outflowing DOM was more enriched in autochthonous (marine) compounds than the terrestrial endmember (but still contained a more terrestrial signature than the North Sea inflow). There was a substantial increase in N (6- to 8-fold increase), S (5- to 7-fold increase), and P (3- to 7-fold increase) heteroatoms in the molecular formulae after transit through the Baltic Sea compared to the riverine inflow at the northern Bothnian Bay (comparing the terrestrial endmember in **Table 1** to the outflow in **Table 2** in spring and autumn, respectively), suggesting that primary production and biotransformation in the Baltic Sea act as sources of these compounds. At the same time, aromatic terrestrial compounds were depleted in the outflow (AImod decreased from 0.49 and 0.51 to 0.33 and 0.32 in spring and autumn, respectively, compared to the terrestrial endmember; **Tables 1, 2**).

The identified processes appear to be applicable to other regions and to larger scales, although their relative extent will certainly depend on seasonal and spatial variabilities such as redox zonation, river discharge, circulation patterns, and DOM residence times. Arctic shelf seas, for example, show similar river runoff schemes as the Baltic Sea, with pronounced spring flood events. Furthermore, more than 50% of terrestrial DOC is removed on the Eurasian shelf before reaching the Arctic Ocean. Yet, this removal occurs on shorter time scales (2–5 years; Alling et al., 2010; Letscher et al., 2011) compared to the Baltic Sea (11.8 years, **Table 3**; and Alling et al., 2008; Stedmon et al., 2010; Deutsch et al., 2012). It remains to be resolved how these fluxes and rates change interannually, or if they are susceptible to changes in land use and an increase of (labile) terrestrial DOC export from sub-arctic rivers.

## DOM Alterations in the Hypoxic Basins of the Baltic Sea

We observed an enrichment of DON in the central Baltic Sea (**Figures 3E,F**). This area is influenced by the discharge of rivers draining watersheds with high agricultural land use which could be important DON sources (Stålnacke et al., 1999; Voss et al., 2011). We also observed high chlorophyll concentrations in the regions with high DON concentrations (**Figures 2G,H**) and high abundances of phytoplankton (data for autumn only, Supplementary Figure 1). Thus, DON release from autochthonous sources such as cyanobacteria (Voss et al., 2005, 2011) is another likely explanation for the observed DON patterns. The DON concentrations positively correlated with unsaturated aliphatic N-containing compounds in spring (group 7, Supplementary Figure 4) which include peptide molecular formulae (Kujawinski et al., 2004; Sleighter and Hatcher, 2008) being consistent with an autochthonous production of DON.

The DON concentrations below the halocline were in the same range as in the photic zone in spring (**Figure 3F**). The similar DON concentrations in surface and bottom waters may reflect a recalcitrant background of DON. It also indicates DON transport below the halocline into the anoxic basins, probably by sinking particles and subsequent release of DON due to degradation processes. The anoxic endmember contained also more marine SPE-DOC in spring (33%) compared to autumn (27%; **Figures 4E,F**). Associated with this trend was a higher abundance of typical marine DOM compounds groups in spring compared to fall in the anoxic basins (compound groups 4, 5, and 7 with/without N were 32% in spring but only 13% in autumn, Supplementary Table 1). Such seasonal changes in the stagnant anoxic basins are remarkable and indicate active processing and seasonal inputs into these basins. The particle export to the deep anoxic basins is likely coupled with the seasonal changes of organic matter composition in the overlaying water column, for example, the higher autochthonous production in spring compared to autumn. In turn, this could mean that if the export of terrestrial organic matter from rivers should change (Voss et al., 2011; Räike et al., 2016), it may also affect the DOM composition in the deep anoxic basins of the Baltic Sea.

Our data also highlight additional modification of the DOM in the anoxic basins. The DOM compounds of the deep anoxic basins were more enriched with sulfur-containing compounds compared to the terrestrial endmember (**Table 1**). Abiotic sulfurization reactions between sulfide species and DOM have been proposed previously to be responsible for the observed elevated dissolved organic sulfur compounds in sulfidic environments (Schmidt et al., 2009; Seidel et al., 2014; Gomez-Saez et al., 2016). Potential abiotic sulfurization reactions include the exchange of an oxygen by a sulfur atom or exchanging H2O, H2, and/or O<sup>2</sup> by addition of H2S per DOM molecular formulae (Schmidt et al., 2009; Gomez-Saez et al., 2016). It has further been hypothesized that abiotic sulfurization reactions may increase the recalcitrant character of DOM, which may explain why these compounds accumulate under anoxic conditions (Seidel et al., 2014).

#### CONCLUSIONS

We identified the molecular imprints of the sources and transformations of DOM in the Baltic Sea. Our box model calculations demonstrated that in both, spring and autumn, the Baltic Sea was a net sink for terrestrial DOM, including condensed and highly aromatic DOM compounds. We also found a net accumulation of autochthonous DOM in the Baltic Proper in spring, likely resulting from OM production during the phytoplankton bloom. Phytoplankton production and possibly also the bio- and photo-transformation of terrestrial compounds are sources for aliphatic, nitrogen-, sulfur-, and phosphorousrich compounds to the Baltic Sea. Seasonal changes of the DOM composition and dissolved organic nitrogen concentrations were found in the deep, stagnant, and anoxic basins, indicating vertical transport of organic particles and organic matter processing in the basins. DOM in the anoxic basins was enriched with sulfur compounds. Abiotic sulfurization reactions in the sulfidic basins are proposed as a likely reason. The inflow of sulfur-rich DOM from North Sea tidal flats and the discharge from the more anthropogenically influenced rivers are likely contributing additional sulfur-containing DOM to the southwestern Baltic Sea and the Kattegat area. Using a box model approach, we estimated that between 43 and 51% of the total terrestrial DOC that is supplied by rivers is removed in the Baltic Sea, while 1.8–2.1 Tg terrestrial DOC is exported per year to the North Sea. These fluxes were comparable for spring and autumn. This suggests that, due to the long residence times of terrestrial DOC in the Baltic Sea (3.7 years in the Bothnian Bay to 6.9 years in the Baltic Proper), seasonal changes of biotransformation and riverine

#### REFERENCES


discharge of terrestrial DOM become largely undetectable. More research is needed to reliably predict how the outflow of terrestrial DOM will change over longer time spans if the riverine export of terrestrial DOM from the sub-arctic north should increase in the future.

#### AUTHOR CONTRIBUTIONS

MS, MM, and TD conceived the study; MS and MM performed chemical analyses; KJ and DH provided microbiological data; MS, TD, and BD performed flux calculations; DH and DS provided nutrient data; MS wrote the manuscript with significant contributions from all authors.

## FUNDING

This work was supported by the ATKiM Project from the Leibniz Society (SAW-2011-IOW-3) to DH, KJ, and MM as well as by the German Science Foundation (DFG, JU 367/15- 1) to KJ, and by the Leibniz Institute for Baltic Sea Research Warnemünde.

### ACKNOWLEDGMENTS

We thank the captains and crews of R/V Meteor for their help during the sampling campaigns (legs M86 and M87). We are also thankful to Matthias Friebe, Katrin Klaproth, and Ina Ulber (University of Oldenburg, Germany) for technical assistance. We further acknowledge the reviewers and the editor for their thoughtful and valuable comments that led to an improved manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/feart. 2017.00031/full#supplementary-material


organic carbon cycle. Geophys. Res. Lett. 22, 417–420. doi: 10.1029/94GL 03344


matter composition in an intertidal sandy beach. Mar. Chem. 176, 150–163. doi: 10.1016/j.marchem.2015.08.011


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Seidel, Manecki, Herlemann, Deutsch, Schulz-Bull, Jürgens and Dittmar. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Environmental Drivers of Dissolved Organic Matter Molecular Composition in the Delaware Estuary

Helena Osterholz <sup>1</sup> \*, David L. Kirchman<sup>2</sup> , Jutta Niggemann<sup>1</sup> and Thorsten Dittmar <sup>1</sup>

*1 ICBM-MPI Bridging Group for Marine Geochemistry, Institute for Chemistry and Biology of the Marine Environment, Carl von Ossietzky University, Oldenburg, Germany, <sup>2</sup> School of Marine Science and Policy, University of Delaware, Lewes, DE, USA*

Estuaries as connectors of freshwater and marine aquatic systems are hotspots of biogeochemical element cycling. In one of the best studied temperate estuaries, the Delaware Estuary (USA), we investigated the variability of dissolved organic matter (DOM) over five sampling cruises along the salinity gradient in August and November of 3 consecutive years. Dissolved organic carbon (DOC) concentrations were more variable in the upper reaches of the estuary (245 ± 49µmol DOC L−<sup>1</sup> ) than at the mouth of the estuary (129 ± 14µmol L−<sup>1</sup> ). Bulk DOC decreased conservatively along the transect in November but was non-conservative with increased DOC concentrations mid-estuary in August. Detailed analysis of the solid-phase extractable DOM pool via ultrahigh resolution mass spectrometry (Fourier-transform ion cyclotron resonance mass spectrometry, FT-ICR-MS) revealed compositional differences at the molecular level that were not reflected in changes in concentration. Besides the mixing of terrestrial and marine endmember signatures, river discharge levels and biological activity impacted DOM molecular composition. DOM composition changed less between August and November than along the salinity gradient. Relative contributions of presumed photolabile DOM compounds did not reveal non-conservative behavior indicative of photochemical processing, suggesting that on the timescales of estuarine mixing photochemical removal of molecules plays a minor role in the turbid Delaware Bay. Overall, a large portion of molecular formulae overlapped between sampling campaigns and persisted during estuarine passage. Extending the analysis to the structural level via the fragmentation of molecular masses in the FT-ICR-MS, we found that the relative abundance of isomers along the salinity gradient did not change, indicating a high structural similarity of aquatic DOM independent of the origin. These results point toward a recalcitrant character of the DOM supplied by the Delaware River. We demonstrate that in addition to bulk DOC quantification, detailed information on molecular composition is essential for constraining sources of DOM and to identify the processes that impact estuarine DOM, thereby controlling amount and composition of DOM eventually discharged to the ocean through estuaries.

Keywords: dissolved organic matter, Delaware Estuary, FT-ICR-MS, discharge, conservative mixing, seasonal variation, collision-induced fragmentation

#### Edited by:

*Nicholas David Ward, University of Florida, USA*

#### Reviewed by:

*Yina Liu, Pacific Northwest National Laboratory, USA Elizabeth Ann Canuel, Virginia Institute of Marine Science, USA*

\*Correspondence:

*Helena Osterholz helena.osterholz@uni-oldenburg.de*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Earth Science*

Received: *26 August 2016* Accepted: *20 October 2016* Published: *17 November 2016*

#### Citation:

*Osterholz H, Kirchman DL, Niggemann J and Dittmar T (2016) Environmental Drivers of Dissolved Organic Matter Molecular Composition in the Delaware Estuary. Front. Earth Sci. 4:95. doi: 10.3389/feart.2016.00095*

## INTRODUCTION

Large amounts of terrigenous organic matter are channeled through rivers and estuaries into the oceans—approximately 0.25 × 10<sup>15</sup> g are transported in the form of dissolved organic carbon (DOC) annually (Hedges et al., 1997). The true amount of the DOC exported from land is likely higher as some regions (for example northern high-latitude rivers, Holmes et al., 2011), diffuse flows and flood events (Cole et al., 2007; Raymond and Saiers, 2010) are not well constrained to date. During transit, the size and composition of the riverine dissolved organic matter (DOM) pool is transformed through a multitude of biotic and abiotic processes. Despite decades of research on estuarine and coastal DOM cycling, we still lack understanding of the mechanisms and magnitudes of these modifications, thereby possibly resulting in misinterpretation of characteristics of the riverine DOM discharged to the coastal seas and ultimately reaching the deep ocean (Bauer and Bianchi, 2011).

Generally, bulk DOC concentrations decrease along estuaries from the river to the ocean while at the same time, the composition of the DOM pool changes. On molecular level, the aromaticity of the DOM decreases (Abdulla et al., 2013), while molecular weight, carbohydrate content (Abdulla et al., 2010), heteroelement content (Sleighter and Hatcher, 2008) and lability increase (D'Andrilli et al., 2015) with increasing salinity. Positive and negative deviations from a conservative mixing line defined by high-DOC river water and the low-DOC marine endmember are frequently observed (Cadée, 1982; Middelburg and Herman, 2007; Sharp et al., 2009). Estuaries are complex systems where, in addition to mixing of fresh- and marine waters, diffuse sources such as subterranean groundwater discharge (Taniguchi et al., 2002), aeolian and anthropogenic input (Liu et al., 2005; Tzortziou et al., 2015) add to the intricacy of biogeochemical cycling. Abiotic processes shaping the DOM pool along the estuarine gradient of pH, ionic strength and turbidity include the adsorption on and desorption from mineral surfaces (Keil et al., 1994; Mayer, 1994), flocculation (Eisma, 1986), aggregation or precipitation. Loss of organic carbon from the dissolved phase in the estuarine turbidity maximum (e.g., Miller, 1999), where the concentration of particulate organic carbon is usually the highest, is accompanied by a loss in aromaticity and nitrogencontaining compounds, as these compounds are preferentially adsorbed (Aufdenkampe et al., 2001; Riedel et al., 2012). Photodegradation processes, although generally thought to be minor in riverine systems due to the low transmission (Spencer et al., 2009), exhibit the highest impact on aromatic moieties of terrestrial origin (Stubbins et al., 2010). By decreasing the average molecular weight and increasing the contribution of aliphatic compounds, the bioavailability of the DOM can increase through photochemically mediated breakdown of molecules especially in tropical regions (Medeiros et al., 2015b). Biological processes, in temperate regions influenced by seasonality, such as selective uptake and transformation by heterotrophic microbes (Azam et al., 1994; Bourgoin and Tremblay, 2010), as well as addition of compounds via autochthonous production (Pennock and Sharp, 1986) likewise shape the estuarine DOM pool. Medeiros et al. (2015b) for example showed that phytoplankton-derived DOM inputs introduce saturated compounds in the tropical Amazon plume. DOM export by the Yukon River to the Bering Sea was highly stable over seasons, with enhanced DOM photo- and biolability in spring due to higher inputs of carbohydrates and aromatic moieties (Cao et al., 2016). Previous studies have furthermore shown that the DOC concentration in rivers increases with discharge (Schiff et al., 1998; Raymond and Saiers, 2010), superimposing the influence of most other environmental drivers. At high discharge, riverine DOM composition tends to carry a stronger terrigenous signal (Spencer et al., 2008). A similar observation was made by Medeiros et al. (2015a) who found strong seasonal variability in molecular DOM composition in the marsh-dominated Altamaha-Doboy-Sapelo estuary. This variability was driven mainly by river flow, leading to a stronger terrigenous signature at high discharge.

In our study, we investigate estuarine carbon cycling in the tidal Delaware River and Bay, located on the east coast of the USA. The freshwater flow into the estuary is dominated by the Delaware River, which supplies nearly 60% of the freshwater to the bay; the Schuylkill and other smaller rivers supply another 14% (Sutton et al., 1996). In total, these rivers account for >80% of freshwater input into the estuary. The water column of the Delaware Estuary is usually well-mixed, although stratification can build up near the mouth of the bay, particularly during summer (Garvine et al., 1992). The bay has an estimated flushing time of about 80 to 180 days (Ketchum, 1952; Sharp et al., 1982; Cifuentes et al., 1990). In parts heavily urbanized, its waters carry a high nutrient loading from the Philadelphia area and saline waters reach as far as ∼125 km from the mouth (Biggs et al., 1983). The estuary has been extensively monitored over the last 30 years with a focus on carbon and nutrient fluxes (Sharp et al., 2009). Different sources of the DOM and particulate organic matter (POM) have been identified along the estuary: isotope and biomarker studies of ultrafiltered DOM have identified a terrestrial signal in the river and turbid middle estuary transitioning to a more algal and zooplankton-dominated signal in the lower estuary and coastal ocean (Mannino and Harvey, 1999, 2000b). Further compositional changes of the DOM pool have been examined at the level of chemical classes, e.g., total amino acids and polysaccharides (Hoch and Kirchman, 1995; Mannino and Harvey, 2000a; Kirchman and Borch, 2003), which in addition to mixing of marine and terrestrial signatures reveal a seasonal imprint as well as processing in the estuarine turbidity maximum. Of the rivers draining into the Middle Atlantic Bight, the Delaware River however carries a high proportion of allochthonous DOC and aged carbon (Hossler and Bauer, 2013).

We characterized the main drivers of DOM concentration and composition between August and November over 3 years (five sampling cruises) along the salinity gradient of the Delaware estuary, thereby allowing inferences about the modified terrestrial signal ultimately reaching the Atlantic Ocean. In addition to bulk DOC concentrations, we assessed the compositional variation of solid-phase extractable (SPE) DOM using ultrahigh resolution mass spectrometry (Fourier-Transform Ion Cyclotron Resonance Mass Spectrometry, FT-ICR-MS), which allowed the assignment of thousands of different molecular formulae to a water sample (Stenson et al., 2003; Koch et al., 2005, 2007). Further, we applied collision-induced fragmentation to reveal possible structural changes within the SPE-DOM pool along the salinity gradient that might be hidden on molecular formula level. The different levels of analytical resolution provided unprecedented insight into the DOM cycling in the Delaware estuary. Overall, we used the new opportunities presented by FT-ICR-MS to build on the previous body of work in the Delaware and other study systems and to gain a more detailed understanding of DOM at the molecular level.

#### MATERIALS AND METHODS

#### Site Description and Sample Collection

Sampling was carried out from about 160 km upstream from the mouth of the estuary to a station just outside of the Delaware Bay during five cruises over three years (November 7–10 2011, August 9–13 2012, November 12–16 2012, August 3–7 2013, and November 18–21 2013). Water was obtained by a CTD rosette-mounted Niskin bottle from ∼0.5 m depth. Ammonium, nitrate, phosphate, and silicate concentrations were measured with a SEAL Analytical AA3 Continuous Segmented Flow Analyzer using standard procedures (Sharp et al., 2009). The concentration of total chlorophyll a was estimated in acetone extracts by fluorometry. Light attenuation was estimated from the intensity of photosynthetically active radiance over a depth profile measured with a Biospherical PNF-210 radiometer. A microcentrifuge approach was used to estimate leucine incorporation (added concentration of 20 nM) (Kirchman, 2001). Sampling locations and environmental parameters are summarized in Table S1 and Figure S1.

River discharge at Trenton and Schuylkill gage stations was obtained from the USGS Water Information System (waterdata.usgs.gov). We calculated the sum of discharge for 7, 14, 28, 56, and 112 days before the start of each sampling campaign and found the same general trend (Figure S2). The summed discharge of 112 days was chosen for further calculations because it lies well within the residence time of the estuary which is between 80 to 180 days (Ketchum, 1952; Sharp et al., 1982; Cifuentes et al., 1990).

#### Dissolved Organic Matter Analysis

For quantification and molecular analysis of the DOM, water was filtered through precombusted glass fiber filters (GF/F, Whatman), acidified to pH 2 with 25% HCl (p.a. grade), and stored at 4◦C in the dark before being shipped to Germany for analysis. Concentrations of DOC and total dissolved nitrogen (TDN) were measured via high temperature catalytic oxidation on a Shimadzu TOC-VCPH analyzer equipped with a TDN module. Accuracy and precision were monitored with consensus reference material (DSR, D. Hansell, University of Miami) and were better than 5%. DON was estimated by subtracting nitrate and ammonium concentrations from TDN. The DOM was solidphase extracted according to Dittmar et al. (2008) using 1 g PPL columns (Agilent Bond Elut), and the resulting methanol extracts were stored at −20◦C in the dark. The concentration of extractable DOC was determined from the extracts after complete removal of the methanol and dissolution in ultrapure water. Methanol extracts were diluted to a DOC concentration of 20 mg C L−<sup>1</sup> in a methanol:water mixture of 1:1 (v/v) and submitted to Fourier Transform Ion Cyclotron Resonance Mass Spectrometry on a solariX FT-ICR-MS (Bruker Daltonik GmbH) equipped with an electrospray ionization source (Bruker Apollo II) applied in negative mode. 500 scans in a mass window from 150 to 2000 Da were accumulated and molecular formulae were calculated with the following restrictions: <sup>12</sup>C1-<sup>130</sup> <sup>1</sup>H1-200O1-<sup>50</sup> <sup>14</sup>N0-4S0-2P0-<sup>1</sup> to masses above the method detection limit after Riedel and Dittmar (2014). Additionally, masses detected in less than three samples were removed prior to further analysis.

Each sample was normalized to the sum of FT-ICR-MS signal intensities and corrected for dilution per sample. The modified aromaticity index (AImod, Koch and Dittmar, 2006 as corrected and reported in the erratum published in 2016), double bond equivalents (DBE) and weighted molar ratios were calculated for each sample. The molecular formulae were assigned to compound groups according to their O/C and H/C ratios as described in Seidel et al. (2014). Here, the seven main groups with their subcategories include (1) polycyclic aromatics (PCA, AImod > 0.66) subdivided into PCA containing more than 15 C atoms indicative of dissolved black carbon, PCA with less than 15 C atoms, or PCA containing a heteroelement; (2) polyphenols (0.5 < AImod ≤ 0.66) with high (O/C ≥ 0.5) or low oxygen (O/C < 0.5) content; (3) highly unsaturated aliphatics (AImod ≤ 0.5, H/C < 1.5, O/C < 0.9) with high (O/C ≥ 0.5) or low (O/C < 0.5) oxygen content; (4) unsaturated aliphatics (1.5 < H/C ≤ 2, O/C < 0.9, N = 0) with high (O/C ≥ 0.5) or low (O/C < 0.5) oxygen content; (5) saturated fatty acids (H/C > 2, O/C < 0.9) with or without heteroelements; (6) carbohydrate-like formulae (O/C > 0.9) with or without (N,S,P = 0) heteroelements; and (7) peptide-like molecular formulae (1.5 < H/C < 2, O/C < 0.9, N > 0). The assignment of a formula to a compound class is not unambiguous since the grouping does not take into account all possible isomers. Calculated and intensity-weighted molecular composition and molecular categories per sample are provided in Table S1.

In order to investigate the structural DOM composition along the salinity gradient, we fragmented six nominal masses of two homologous series (m/z 269, 283, 297, 341, 365, and 379 Da) of three samples from the sampling campaign in November 2012 (salinities 0.8, 14.7, and 30.7) via collision-induced fragmentation in the FT-ICR-MS. The nominal masses were chosen to compare to a previous publication using the same approach (Osterholz et al., 2015), to span a broad mass range and to have a high relative signal intensity suitable for fragmentation. The November 2012 cruise was chosen for the fragmentation experiment as the DOC concentrations behaved conservatively, being representative of the conditions in the estuary most of the time except for bloom or very high discharge conditions. For the fragmentation, methanol extracts were diluted to a DOC concentration of 60 mg C L−<sup>1</sup> in a methanol:water mixture of 1:1 (v/v). The collision voltage was chosen to yield similar signal intensities of the most intense fragments and varied between 13 V and 18 V. 150 scans were accumulated for each analysis performed in duplicates. Molecular formula assignment to fragmentation mass spectra was done as described above. We identified 423 fragments resulting from the neutral losses of H2O, CH4O, 2H2O, CO2, CO2+H2O, CO2+CH4O, 2CO2, 2CO2+H2O, 2CO2+CH4O, 3CO2, 3CO2+H2O or 3CO2+CH4O. Only molecular formulae that were detected in the two replicate analyses were analyzed in further detail. The relative signal intensity of each fragment ion was divided by the sum of the intensities of all major fragment signal intensities plus the precursor ion signal intensity for better comparison between samples. The normalized fragment intensities were then grouped by neutral loss and compared between samples.

#### Statistical Analysis

Conservative mixing lines (**Figures 1**, **2**) were constructed using the respective mean concentrations/values of the two samples of highest and lowest salinity. The deviation of each sample from the theoretical mixing line was then calculated and averaged over the whole transect. A high value therefore represents a high variability of the parameter in relation to the values expected for a conservative mixing scenario, similar as described in Seidel et al. (2015a).

Multivariate statistical analyses were performed using R (version 3.2.2) and the package vegan (Oksanen et al., 2015). All analyses were run on normalized intensities of signals with molecular formula assignment which were corrected for dilution. Four extraction efficiencies (on carbon basis) above 100 and below 40% were likely erroneous and were replaced by the average value of extraction efficiencies of other samples (67%). Principal coordinate analyses (PCoA) were performed on Bray Curtis distance matrices (Bray and Curtis, 1957). Environmental vectors were fitted to the respective PCoA scores using the envfit function in the vegan package. Spearman rank correlations were calculated between single molecular formulae intensities and environmental parameters. Only correlations between a formula and an environmental parameter with a p < 0.01 were considered in further analysis and discussion. The non-parametric Wilcoxon rank-sum test was used for comparisons between groups of samples.

#### RESULTS

#### Environmental Characterization of the Delaware Bay

DOC, TDN and nitrate concentrations as well as turbidity generally decreased along the salinity gradient (**Figure 1**, Figure S1). Several biological parameters influenced by water temperature and light availability varied greatly between the two months (August and November) we sampled. These parameters included microbial abundance, chlorophyll a concentration and leucine incorporation as well as silicate concentrations (**Figure 2**, Figure S1). Freshwater discharge did not vary consistently between the two seasons.

Most environmental parameters deviated from conservative mixing along the salinity gradient. Nitrate concentrations, for example, were always high in the oligohaline zone, decreasing toward the ocean with a negative deviation from conservative mixing at intermediate salinities. Most of the total dissolved nitrogen in the Delaware estuary is inorganic. DON concentration did not vary consistently along the estuarine salinity gradient or with season.

Silicate concentrations were depleted at the freshwater end of the estuary in August, while in November concentrations were high because the river contributes large amounts of dissolved silicate into the estuary. Chlorophyll a concentration (**Figure 2**) and leucine incorporation (Figure S1) differed between the two campaigns that took place in August, yet covaried significantly (ρ = 0.82, p < 0.01) with highest values observed in the oligo- and polyhaline zones in August 2013 and overall high values during August 2012.

DOC concentrations decreased linearly (conservatively) during estuarine mixing in November (**Figure 1**). In August, however, changes in DOC concentrations were non-conservative; deviations from conservative mixing were significantly higher in August than in November (Wilcoxon rank-sum test, p < 0.01). Overall, DOC concentrations were more variable at the freshwater end; the coefficient of variation (CV) at low salinity was 20% (salinity < 1, mean DOC 245 ± 49µmol L−<sup>1</sup> , n = 14) while it was 11% in marine waters (salinity > 30, mean DOC 129 ± 14µmol L−<sup>1</sup> , n = 14). At salinities of less than 1, DOC concentrations were significantly lower in August than in November by 1.4-fold (Wilcoxon rank-sum test, p < 0.01). DOC concentrations in marine waters did not vary substantially between the 2 months. DOC concentrations at the high or low salinity extremes did not correlate significantly with discharge.

A turbidity maximum was apparent at salinities between 0 and 5 during all cruises; turbidity then decreased toward the ocean. Turbidity patterns were neither related to discharge nor revealed an impact on phytoplankton growth or DOC concentration.

## Dissolved Organic Matter Molecular Composition

In the whole dataset consisting of 121 solid-phase extracted DOM samples, 12666 unique molecular formulae (MF) within the mass range of 154–885 Da were identified. Of those, 3093 MF (24%) were present in all 121 samples. Further, 78% of all MF were detected at least once during each of the five cruises. Considering each cruise separately, 34–45% of the MF were observed in every sample (November 2011: n = 22, August 2012: n = 26, November 2012: n = 25, August 2013: n = 26, November 2013: n = 22).

Principal coordinate analysis was performed on a Bray-Curtis dissimilarity matrix (Bray and Curtis, 1957) calculated using relative signal intensities of all samples (**Figure 3**). The two major axes of variation together encompassed almost 80% of the variability of the dataset. Salinity was highly positively correlated to principal coordinate PC1 (ρ = 0.78, p < 0.001, **Figure 1**), whereas silicate and DOC concentration were negatively correlated with PC1 (ρ = −0.95 and −0.57, respectively, p < 0.001). Discharge was strongly associated

FIGURE 1 | DOC concentration and characteristics of DOM molecular composition (rows) vs. salinity for all five cruises (columns). PC1 scores are from Figure 3 (1 was added to each score for data representation); H/C ratio, N- and S-content of molecular formulae were weighted by peak intensity; AImod, modified aromaticity index; HU, percentage of highly unsaturated compounds. Solid lines represent conservative mixing lines calculated using averages of the two samples of lowest and highest salinities, respectively. Numbers in italics denote percentage of deviation between the conservative mixing line and the measured values (see section Materials and Methods).

with PC2 (ρ = −0.75, p < 0.001, **Figure 4**), the second axis of variation comprising 25% of the variability of the DOM molecular data. Microbial abundance, bacterial production (leucine incorporation) and chlorophyll a concentration were correlated with the third and/or fourth PCs, encompassing 6 and 3% of the variation in the data, respectively (Figures S3, S5E,F).

denote standard error. Spearman rank correlations for each parameter vs. chlorophyll *a* concentration are given. Significance of correlations is *p* < 0.01.

Three environmental parameters (salinity, discharge, chlorophyll a concentration) were chosen for a more detailed investigation of changes in molecular DOM composition. These three parameters each represent one group of co-correlating environmental variables identified via hierarchical clustering (Figure S4).

Almost half of all identified molecular formulae, capturing 85 ± 5% of the total signal intensity of each sample, decreased in intensity with increasing salinity. The overall aromaticity of the SPE-DOM, assessed via the modified aromaticity index AImod, strongly decreased (**Figure 1**, **Table 1**). A molecular formulae based measure for terrigenous input, Iterr proposed by Medeiros et al. (2016), decreased from ∼0.5 at the freshwater end to ∼0.2 in the most marine waters and was highly negatively correlated with salinity (ρ = −0.79, p < 0.001). The relative contribution of sulfur to DOM molecular composition exhibited a weak relationship with salinity, overall decreasing toward the ocean. SPE-DOM N-containing molecular formulae, average molecular mass and H/C ratio all increased with increasing salinity (**Figure 1**, Figures S5A,B, Table S1). The relative abundance of highly unsaturated compounds, representing the largest compound class, likewise increased. The index for labile compounds above the molecular lability boundary (MLB) of H/C > 1.5, the MLBwL proposed by D'Andrilli et al. (2015), slightly increased toward the ocean for all cruises except in November 2012 (ρ = 0.23, p < 0.05).

At high discharge in the Delaware Bay, we found high values for O/C ratio, molecular mass and highly unsaturated compounds (**Figure 4**). The subgroup of oxygen-rich highly unsaturated compounds emphasizes this trend (**Figure 4**, **Table 1**, Figures S5C,D). We observed only a slight increase in aromaticity, while the Iterr did not correlate significantly with discharge (p > 0.05). The MLBwL was highly negatively linked with discharge (ρ = −0.80, p < 0.001). About 2000 molecular formulae were positively affected by discharge and another 3000 were negatively affected, in total accounting for ∼60% of total signal intensity (**Table 1**).

Chlorophyll a concentration and microbial activity were highly correlated; therefore only chlorophyll a was chosen for


TABLE 1 | Characteristics of molecular formulae correlating positively (POS) or negatively (NEG) with salinity, discharge and chlorophyll a concentration as well as of those molecules exhibiting no correlation to either one of the three environmental parameters.

*Values represent formulae counts (MF), intensity-weighted molecular mass in Dalton (MW), element contribution (C, H, O, N, S, P), molar ratios (O/C, H/C), molecular indices (AImod, DBE) and relative contribution to compound classes (PCA, polycyclic aromatics; PCA*<*15 or* ≥*15, PCA with less than or more/equal to 15 C atoms; -X, containing heteroelements (N, S, P); PP, polyphenols; HU, highly unsaturated compounds; UA, unsaturated aliphatics; FA, fatty acids).*

further investigation. Overall, chlorophyll a concentration was variable along the salinity gradient, but consistently displayed positive deviations from conservative mixing in August and negative deviations in November (**Figure 2**). Only a few molecules varied consistently with chlorophyll a concentration and accounted for ∼6% of the total signal intensity (Figure S5E, **Table 1**). The molecular variation associated with the chlorophyll a gradient was, however, markedly different from random correlations as tested with a dataset of normally distributed, random variables (Figure S5H). During periods of high phytoplankton biomass and microbial production, DOM was enriched in N, S, O, unsaturated aliphatics, polyphenolic compounds as well as sugar- and peptide-like molecular formulae (**Table 1**). To explore the correlations in more detail, the deviations from conservative mixing of DOC concentrations and molecular parameters were calculated (1DOC, 1N, 1O/C); significant increases are shown in **Figure 2**. The MLBwL and Iterr did not correlate significantly with chlorophyll a concentrations (p > 0.05).

The 471 ± 126 molecular formulae that correlated with one of the three factors (p > 0.01) accounted for ∼3% of the total signal intensity and were widely distributed over the van Krevelen space occupied by a typical DOM sample, but were less prevalent in the low-H/C and high-O/C region (Figure S5G).

### Dissolved Organic Matter Structural Analysis

Fragmentation of six nominal masses from three samples along the salinity gradient in Nov 2012 was used to detect potential preferential removal or addition of DOM structural features along the salinity gradient of the Delaware Estuary. A total of

dashed lines denote standard error. Spearman rank correlations for each parameter vs. discharge are given (ρ). Significance of correlations is *p* < 0.001, except for HU where *p* < 0.05. HU, highly unsaturated compounds; PC2 scores are taken from Figure 3.

423 detected fragments were normalized and grouped by neutral loss (H2O, CO2, CH4O as well as combinations and multiples thereof). We did not detect any of the OCH<sup>2</sup> functionalities described by Liu et al. (2011). These neutral losses represent a component of lignin biopolymers generated from wood extract and C18-extracted black waters and were thus presumed to occur in the terrigenous DOM. In pairwise comparisons of the relative signal intensities per neutral loss, a ratio of 1 indicates that no differences in neutral losses were observed between samples. This in turn suggests that a formula with the same molecular mass has a similar structure, regardless of its sampling location within the salinity gradient of the estuary. Although there were some deviations from the ratio of 1 among samples taken at different salinities, the fragmentation patterns did not differ significantly (Wilcoxon rank-sum test, p > 0.05, **Figure 5**).

#### DISCUSSION

In order to investigate the environmental drivers of DOM quantity and quality in a temperate estuarine system, we combined quantitative analysis of DOC concentrations with ultrahigh resolution mass spectrometric analysis of solid-phase extractable (SPE) DOM from samples from the entire salinity gradient of the Delaware estuary over 3 years. The analytical window provided by the combination of PPL-SPE and FT-ICR-MS allowed us to examine a large, otherwise inaccessible portion of the DOM pool; in this study on average 67 ± 6% of the DOC was captured with our solid-phase extraction method. The PPL columns used here poorly recover very small and polar compounds (Raeke et al., 2016), but absorb a wide range of compounds of varying polarity and are thought to provide a major fraction of the natural DOM (Green et al., 2014; Li et al., 2016). The extraction efficiency did not significantly change along the salinity gradient of the estuary, indicating that our method captured a reproducible fraction of the estuarine DOM pool. FT-ICR-MS analysis of our SPE-DOM samples allowed us to gain valuable information on molecular composition and explore variation with biogeochemical properties. Additional insights into DOM structural features were revealed with in-cell fragmentation.

## Mixing of Terrigenous and Marine DOM along the Salinity Gradient

On its way from soils to the ocean margins, DOM can undergo transformations in molecular composition that are not reflected in the DOC concentrations along the salinity gradient. Besides a slight decrease in molecular formulae richness (ρ = −0.29, p < 0.01) and the Shannon diversity index (ρ = −0.44, p < 0.001) along the salinity gradient, condensed aromatic compounds, AImod, and polyphenols indicating DOM of terrigenous origin were higher in the low salinity region. Proportions of highly unsaturated compounds and molecular mass increased with increasing salinity, as did the contributions of nitrogen-containing molecular formulae and sugar- and peptide-like compounds that are thought to be products of more recent primary production (Amon et al., 2001). Our results are overall in line with previous studies applying ultrahigh resolution mass spectrometry to molecularly characterize DOM. These have shown that DOM tends to become more aliphatic, has a higher heteroelement content (N, S, P), and contains

lower abundances of oxygen-rich molecules along gradients of increasing salinity in temperate (Sleighter and Hatcher, 2008) and tropical estuaries (Medeiros et al., 2015b; Seidel et al., 2015b). Highly unsaturated compounds, however, are thought to mainly comprise soil-derived "humics" and other highly unsaturated compounds according to Šantl-Temkiv et al. (2013). It has to be kept in mind that FT-ICR-MS detects relative signal intensities and grouping into molecular classes is reported in proportions, i.e., if one compound class such as the polyphenols strongly decreases in a sample, another compound class or classes must increase. The less oxygenated polyphenols decrease more strongly with increasing salinity than the oxygen-rich subgroup (**Table 1**) which includes many phenolic acids. This finding illustrates that the classification into broad compound groups does not unambiguously identify the origin and fate of the molecules. Further, although unsaturated aliphatic compounds have previously been shown to be abundant in the ocean supposedly due to the DOM from marine organisms (Sleighter and Hatcher, 2008; Seidel et al., 2015a), they exhibited no relationship with salinity in our study.

Data on DON concentration and the molecular composition data give a different picture about sources of DOM in the estuary. DON concentrations were similar along the whole transect whereas increased DON concentrations have been observed previously in freshwater parts of estuaries (Badr et al., 2008). However, there was a strong positive correlation between relative intensities of N-containing molecular masses and salinity (ρ = 0.77, p < 0.001). This is consistent with the hypothesis that marine DOM is mainly derived from autochthonous production and contains more heteroelements than DOM from terrestrial sources (Sleighter and Hatcher, 2008). In addition to a higher N content of the SPE-DOM molecular formulae, we also observed a slightly positive correlation between the relative abundance of P in the assigned molecular formulae and salinity (ρ = 0.18, p < 0.05).

We propose that variation in the endmembers (river and ocean) is preserved more clearly in the DOM molecular composition than in DOC concentration. The coefficient of variation (CV) of DOC concentrations was 20% (245 ± 49µmol L −1 ) in waters with salinities <1 and 11% (129 ± 14µmol L−<sup>1</sup> ) in waters with salinity > 30. In contrast, the CV of the Bray-Curtis dissimilarity of the SPE-DOM samples was similar for the low and high salinity samples (35 and 31%, respectively). Put differently, DOC concentrations decreased along the salinity gradient and reached similarly low levels in the marine waters while DOM molecular composition remained diverse among the five cruises. This discordance in trends of concentration and composition could be caused by the removal of DOM fractions not captured with our SPE-method. A more likely explanation is the consistent removal or dilution or both of SPE-DOM over the whole captured mass range, supported by the strong correlation of DOM molecular composition with salinity (e.g., **Figures 1**, **3**).

The fact that few molecules increase in relative intensity toward the ocean indicates that dilution and/or degradation processes are by far more important than new production of DOM compounds in shaping DOM composition in this system. Many compounds present at the freshwater source were detected along the whole salinity gradient (34–45%, calculated separately for each cruise). Together with the high percentage of compounds that were detected repeatedly during each cruise (78%), these data indicate a common source of DOM from land that is preserved due to the recalcitrant nature of these compounds. Hossler and Bauer (2013) indeed report a high degree of allochthonous (40–96%) and aged DOC (1– 23%) in the Delaware River and Bay. A second possibility is that some compounds are actually removed while others with different structures but the same molecular formulae are produced during the transit through the estuary, thus masking a lower-level DOM variability in the system (Sleighter and Hatcher, 2008). Since structure can be deduced from molecular formula only in extreme cases, i.e., for molecules with very low H/C ratios (Koch and Dittmar, 2006), we conducted fragmentation experiments applying FT-ICR-MS/MS to three samples along the salinity gradient at low, mid and high salinity of the November 2012 cruise. Collision-induced fragmentation has been shown to distinguish samples with similar molecular formulae fingerprints, but different structural

composition (Osterholz et al., 2015). The neutral losses of H2O, CO2, and CH4O as well as combinations and multiples thereof are markedly different in relative intensity and exhibit ratios up to 4 when comparing relative fragment intensities of fresh, biologically produced and aged oceanic DOM. The same neutral losses were detected in the fragmentation experiments with Delaware Estuary DOM (**Figure 5**). However, none of the fragment intensities differed significantly between samples. This finding indicates that structures in aquatic DOM were highly similar, irrespective of their origin from marine or terrestrial sources. For our study, we infer that during estuarine mixing, there were no changes in DOM composition at the structural level that were overlooked at molecular formula level.

#### The Influence of Discharge Levels

During our five sampling campaigns we did not observe correlations between discharge regimes and DOC concentrations. This remained true even if only the lowsalinity samples (salinity < 1, n = 14) were taken into account, which would be most directly affected by discharge. For forested streams in the eastern United States as well as the Amazon, high DOC concentrations at the onset of high discharge events and low DOC concentrations at low discharge have been reported (Raymond and Saiers, 2010; Ward et al., 2013). Other studies using a similar analysis approach to the one used here have detected positive correlations between DOC concentration and discharge (Raymond and Bauer, 2001; Medeiros et al., 2015b). The Delaware Estuary in general shows a weak axial salinity response to freshwater input, probably due to vertical shear flow dispersion and the action of lateral shear coupled to a strong lateral salinity gradient (Garvine et al., 1992). DOC, mostly exhibiting conservative behavior, accordingly showed no strong influence of discharge.

Although discharge did not affect DOC concentrations it affected DOM molecular composition. According to correlation analysis of the principal coordinates, discharge was identified as the second most important driver of DOM molecular composition responsible for about a quarter of the total variability. We observed a pronounced shift in DOM composition between low/normal discharge situations where the discharge was around or below the yearly mean flow of ∼ 400 m<sup>3</sup> s <sup>−</sup><sup>1</sup> provided by the Delaware and Schuylkill River (Polis and Kupferman, 1973) and high discharge regimes (> 400 m<sup>3</sup> s −1 , **Figure 4**). DOM associated with high river discharge was enriched in condensed aromatics, oxygen-rich polyphenols, oxygen-rich highly unsaturated compounds and a higher AImod, all indicative of higher terrigenous DOM input (Meyers-Schulte and Hedges, 1986; Koch et al., 2005). The specific positive correlation with oxygen-rich aromatic compounds presumably relates to the input of lignin- and tannin-like compounds with high O/C ratio (**Table 1**, **Figure 4**). Lignins are complex aromatic biopolymers that make up a major component of vascular plant biomass (Sarkanen and Ludwig, 1971). They are an important component of riverine DOM (Ertel et al., 1986). Tannins are produced by some algae but primarily are constituents of higher plant tissue (Hedges and Weliky, 1989). They are thought to be quickly removed from aquatic environments via decomposition, precipitation, and sorption to sediments (Maie et al., 2008). Litterfall and plant debris are degraded in forest soils, from which rainfall events can mobilize tannins and lignin, leading to high concentrations in the Delaware River when discharge is high. The average mass of the compounds associated with discharge rates was higher than of those compounds negatively correlating with discharge. This is opposite of the expected trend as terrigenous SPE-DOM tends to have a lower molecular weight when compared to marine DOM analyzed via FT-ICR-MS (Koch et al., 2005; Medeiros et al., 2015b). Tannins, in molecular weight from 500 to 3000 Da (Bate-Smith, 1968), and lignin polymers of similar or higher molecular weight increased in abundance at high discharge and therefore could contribute to this increase in average molecular mass of the entire DOM pool. Forest and wetland coverage of the watershed decreases from the upper basin toward the bay but nevertheless accounts for the major land use of the Delaware basin (49 and 8%, respectively, Partnership for the Delaware Estuary, 2012). Forests and wetlands are possible sources for this terrigenous DOM. Another much debated source of DOC to the Delaware Estuary are the Spartina alterniflora-dominated wetlands surrounding especially the lower estuary. Hossler and Bauer (2012) reported a contribution of carbon originating from plants with a C4-photosynthetic mechanism to DOC of up to 51% in the Delaware River. Since the production of maize, a C<sup>4</sup> plant, is low in the Delaware watershed, the salt marshes might be a large source of this C4-organic matter. According to Benner et al. (1987) and Cifuentes (1991) and references therein, the marshes do not seem to export large quantities of organic matter to the estuarine water column. These sources, however, cannot be deciphered through the methods applied in our study.

Further, the river tended to carry more S-containing DOM to the freshwater end of the estuary at low to intermediate discharge scenarios (ρ = −0.54, p < 0.001, **Figure 4**, **Table 1**). The anthropogenically influenced portion of the Delaware River (land cover: 26% agriculture, 15% developed land, Partnership for the Delaware Estuary, 2012) especially in the central and lower watershed, may introduce dissolved organic sulfur (DOS) compounds originating from e.g., wastewater input and agriculture (Gonsior et al., 2011; Wagner et al., 2015), which would be less diluted at low river discharge. In the same manner, DOS compounds formed during early diagenesis in sediments could be released into the water column (Seidel et al., 2014) or be exported from the salt marshes surrounding the Delaware Bay (Luther et al., 1986). The Delaware Bay nevertheless carried SPE-DOS in ranges comparable to less anthropogenically influenced rivers such as Lena, Amazon and Congo (Wagner et al., 2015).

#### Seasonal DOM Imprint Due to Biological Processes

The impact of biological processes on DOM concentration and composition was assessed through chlorophyll a concentration as a proxy for phytoplankton biomass and bacterial activity measured via leucine incorporation. Due to their high cocorrelation, these two factors cannot be distinguished in our correlative approach (ρ = 0.82, p < 0.001). There were no significant correlations between DOC and chlorophyll a concentrations or with leucine incorporation rates.

Only a small portion of molecular formulae comprising about 6% of the total signal intensity correlated positively or negatively with chlorophyll a (**Table 1**). This number is similar to the percentage of molecular formulae attributed to phytoplankton and/or biodegradation-related processes in the Amazon (Medeiros et al., 2015b). The authors report selective removal of compounds with high O/C ratios via biodegradation (Sun et al., 1997; Kim et al., 2006) and input of phytoplankton-derived material of low O/C ratio (Landa et al., 2014). This was not observed for the Delaware system, where we found that oxygen-rich polyphenols and highly unsaturated compounds increased with chlorophyll a concentration. The higher heteroelement content and occurrence of saturated compounds are consistent with phytoplankton-derived DOM (Sleighter and Hatcher, 2008). Phytoplankton classes differed significantly not only between August and November sampling, but also between the different years (Kirchman et al., in press), possibly producing different DOM compounds and thereby hampering the detection of a uniform "biological signature."

#### Abiotic Factors Influencing DOM Composition

At increasing ionic strength, abiotic processes such as flocculation and absorption to particles may influence DOM composition (Hernes and Benner, 2003). Selective removal of terrestrially derived compounds such as humic acids (Fox, 1983; Ertel et al., 1986) has been reported during estuarine mixing, as well as sorption of amino acids and dissolved polysaccharides to particles in the turbidity maximum zone of the Delaware Estuary (Mannino and Harvey, 2000a). Flocculation with metal salts, for example, has been shown to selectively remove more aromatic compounds (Riedel et al., 2012). Cifuentes (1991) proposed the release of lignin phenols from resuspended sediments in the turbid regions of the Delaware Estuary. We were not able to attribute changes in DOC concentration or molecular DOM composition to adsorption processes in this zone through our correlative approach, i.e., correlating relative FT-ICR-MS signal intensities or bulk DOC concentrations with an environmental factor (here: turbidity). Likely, extensive DOM processing before it reaches the river and estuary (Goñi and Gardner, 2003), the small relative contribution of those compound classes to the total DOC pool, the low spatial resolution of the sampling points in and around the turbidity maximum and the strong superimposing signal of the mixing of fresh- and saltwater in combination all hinder the recognition of a turbidity maximum signal. In a recent publication, Hermes and Sikes (2016) showed that POC concentrations are highest in the bottom waters of the estuarine turbidity maximum and sampling at that depth might have revealed a stronger imprint of DOM-POM interaction that are not evident at the 0.5 m sampling depth chosen for our study.

Additionally, photochemical processes are known to shape the DOM pool by removing (condensed) aromatic structures, especially in rivers with watersheds dominated by forests and grassland (Hernes and Benner, 2003; Stubbins et al., 2010; Riedel et al., 2016). This process would yield a negative concave relationship of, for example, condensed aromatics and polyphenols, vs. salinity as both sunlight exposure potential and time increase while the more colored freshwater is diluted along the estuary. As these compound classes decreased linearly along the salinity gradient, there was no evidence of a significant contribution of photochemical alteration to estuarine DOM processing in the Delaware Estuary. Further, we analyzed the relative contribution of molecular formulae identified as photolabile, photo-resistant and photo-produced by Stubbins et al. (2010) for the Congo River, but no pattern indicative of photochemical alteration emerged (data not shown).

In line with our findings, no large in situ sources or sinks of chromophoric DOM in the Delaware Estuary were reported by Del Vecchio and Blough (2004). These and other authors found photochemical processing to be more important in clearer waters offshore with higher light penetration depths for the Delaware, Amazon, and Mississippi river plumes (Hernes and Benner, 2003; Medeiros et al., 2015b; Seidel et al., 2015b). Indeed, an extensive photochemical DOM sink during August months was described also for the Delaware Estuary plume by Vodacek et al. (1997). The authors further assume oceanic and terrestrial DOM pools mixing in the estuary and on the shelf to originate from constant or very similar sources, which is consistent with a high overlap of detected molecular formulae described in this study.

### CONCLUSION

In our study, we revealed DOM molecular-level variations that are masked by looking at DOC concentrations alone. Mixing of fresh- and marine waters, hydrological conditions, and also biological processes alter the composition of the DOM discharged through the Delaware Estuary into the Atlantic Ocean. For the first time, we applied collisioninduced fragmentation to nominal masses along a salinity gradient, demonstrating that the chemical structures behind DOM molecular formulae exhibit a high similarity independent of their origin from marine or terrestrial environments.

Understanding the numerous processes occurring during land-to-ocean carbon transport is of utmost significance to the global carbon cycle as riverine discharge of DOC alone is sufficient to support the turnover of DOC throughout the entire marine environment (Williams and Druffel, 1987). Future increase in discharge due to increased precipitation and storm events projected by climate change scenarios and land use change reducing forest cover will most probably alter the signature of the DOM reaching the ocean, corroborating the need for further monitoring especially of such well-understood land-to-ocean connectors as the Delaware Estuary.

#### RESOURCE IDENTIFICATION INITIATIVE

CRAN: RRID:SCR\_003005, vegan: RRID:SCR\_011950.

#### AUTHOR CONTRIBUTIONS

DK, TD, and JN conceived the study. HO analyzed the samples and wrote the manuscript with considerable input of all coauthors.

## FUNDING

This work was supported by NSF grants OCE-1030306 and OCE-1261359 to DK. HO, JN, and TD were funded by the University of Oldenburg, Germany.

## ACKNOWLEDGMENTS

The field work was only possible with the help of Matt Cottrell, Monica Stegman, Thomas Lankiewicz, Katie Kalis, Raphaël Lami,

#### REFERENCES


and Dave Kieber while Liying Yu was essential in the laboratory analyses. We are grateful to Matthias Friebe and Ina Ulber for DOC/TDN analysis and to Katrin Klaproth for FT-ICR-MS analysis.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/feart. 2016.00095/full#supplementary-material


organic matter in marine environments. J. Geophys. Res. Oceans 108:3291. doi: 10.1029/2002JC001421


patterns. Geochim. Cosmochim. Acta 63, 2219–2235. doi: 10.1016/S0016- 7037(99)00128-3


**Conflict of Interest Statement:** The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The handling Editor declared a past co-authorship with the author TD and states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2016 Osterholz, Kirchman, Niggemann and Dittmar. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Influence of Major Storm Events on the Quantity and Composition of Particulate Organic Matter and the Phytoplankton Community in a Subtropical Estuary, Texas

#### Nicolas E. Reyna, Amber K. Hardison\* and Zhanfei Liu\*

*Marine Science Institute, University of Texas at Austin, Port Aransas, TX, USA*

#### Edited by:

*Nicholas David Ward, Pacific Northwest National Laboratory (DOE), USA*

#### Reviewed by:

*Gupta G.V.M., Ministry of Earth Sciences, India Ana Rosa Arellano, University of Florida, USA Allison Myers-Pigg, Texas A&M University, USA*

#### \*Correspondence:

*Amber K. Hardison amber.hardison@utexas.edu Zhanfei Liu zhanfei.liu@utexas.edu*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *08 November 2016* Accepted: *07 February 2017* Published: *24 February 2017*

#### Citation:

*Reyna NE, Hardison AK and Liu Z (2017) Influence of Major Storm Events on the Quantity and Composition of Particulate Organic Matter and the Phytoplankton Community in a Subtropical Estuary, Texas. Front. Mar. Sci. 4:43. doi: 10.3389/fmars.2017.00043* Variations in the freshwater inflow regimes of estuaries due to perturbations, such as storm events, alter the source, and composition of particulate organic matter (POM) and the phytoplankton community which are key links in estuarine carbon and nitrogen cycling. To evaluate the impact of varying freshwater discharge on POM quantity and composition and the phytoplankton community, monthly samples of surface water were collected at four long-term monitoring stations from 2012–2015 in a subtropical estuarine system, the Mission-Aransas Estuary, Texas (USA). This system is characterized by a semiarid climate, drought, and sporadic precipitation events. We analyzed organic carbon and nitrogen content, carbon and nitrogen stable isotopes, and chlorophyll *a* and accessory pigments of the POM samples. Following a prolonged dry spell (2012– 2015), consecutive major storm events in spring 2015 led to a dramatic freshening of the entire estuary. Large increases in particulate organic carbon (POC) concentrations and decreases in δ <sup>13</sup>C values of POC over several weeks following the storms suggest an increase in *in-situ* production at lower salinities as a result of increased freshwater inflow. These changes in bulk POM coincided with an increase in chlorophyll *a* concentration, further indicating a significant contribution by phytoplankton to the elevated POM. Concurrently, pigment biomarkers revealed a significant (10-fold) increase in the cyanobacterial pigment, zeaxanthin, and further 16S rRNA analysis showed that *Cyanobium* spp. was responsible for the observed bloom. The combination of environment conditions, including freshening, high temperature, and high nutrients, likely contributed to the cyanobacteria bloom. These results show that episodic rain events can substantially affect estuarine phytoplankton community composition, which impacts the available energy resources for secondary production and thus may have ecosystem-wide implications on productivity.

Keywords: mission aransas estuary, storms, particulate organic matter, carbon and nitrogen isotopes, accessory pigments, cyanobacteria, phytoplankton pigments, phytoplankton

## INTRODUCTION

Estuaries provide a valuable link between terrestrial and oceanic environments through the transport, regeneration, and production of particulate organic matter (POM). POM includes both living and non-living organic matter and is the base of the food web in estuarine systems. Although concentrations of POM in estuaries are typically higher than those in the open ocean, POM in both environments consists of both labile and refractory compounds (Canuel et al., 1995; Harvey and Mannino, 2001). Autochthonous POM sources, such as estuarine phytoplankton, contain a higher relative fraction of labile POM than allochthonous sources, making them a preferable food source for primary consumers (Monbet, 1992; Canuel, 2001; Volkman et al., 2008). This fraction of the POM pool is more readily assimilated by consumers and therefore has a greater impact on the metabolism of the system than the typically more refractory terrestrially-derived organic matter (Day et al., 1989; Canuel et al., 1995; Volkman et al., 2008), although in some cases terrestrial organic matter can be biologically or photochemically labile in estuaries (Hernes and Benner, 2003; Ward et al., 2012, 2013). Estuaries are characterized by strong gradients in physical, chemical, and biological forces such as salinity and nutrient cycling which produce spatiotemporally heterogeneous conditions for phytoplankton growth and affect the availability of this nutritious, labile source of POM (Harding, 1994).

Due to their fast growth rates and importance to aquatic food webs, phytoplankton communities can rapidly respond to a range of environmental perturbations such as storm events, making them a sensitive and crucial indicator for detecting ecological change in estuaries (Peierls et al., 2003; Cloern and Dufford, 2005; Paerl et al., 2010). This has led to extensive characterization of phytoplankton communities in temperate estuaries such as San Francisco and Chesapeake Bays (e.g., Cloern and Dufford, 2005; Adolf et al., 2006; Harding et al., 2015). These systems typically experience high freshwater discharge in the spring as a result of snowmelt and rainfall. Riverine export of nutrients and warming temperatures often lead to a seasonal spring diatom bloom (Cloern and Dufford, 2005; Adolf et al., 2006; Harding et al., 2015). Phytoplankton community composition has been studied extensively in temperate estuaries with high freshwater inflows, but estuaries with low inflow and salinities that vary broadly over time have been less studied (Murrell et al., 2007). Phytoplankton community composition in temperate estuaries is commonly regulated by nutrients, flushing time, temperature, and light availability, whereas environmental controls in estuaries with episodic flooding events can vary drastically from year to year (Bianchi et al., 1997; Pinckney et al., 1998; Paerl et al., 2010; Dorado et al., 2015). Increases in terrestrially derived inorganic nutrients transported by freshwater inflows often cause a shift toward larger-celled phytoplankton, such as diatoms, which have relatively faster growth rates and reduced consumption by grazers (Cloern and Dufford, 2005; Reynolds, 2006). However, this is not always the case, as high levels of nutrients and high temperatures can lead to blooms of small-celled phytoplankton, such as cyanobacteria (Juhl and Murrell, 2005; Ren et al., 2009; Riekenberg et al., 2015).

As intensity and frequency of episodic events are projected to increase (Huntington, 2006), it is important to understand how storms impact biogeochemical processes in estuaries. As an ideal model for subtropical regions, the microtidal Mission-Aransas Estuary (MAE) is located on the western edge of the Gulf of Mexico in south Texas, with a strong episodic nature of precipitation and resulting freshwater inflow events (Mooney and McClelland, 2012; Bruesewitz et al., 2013; Lebreton et al., 2016). Compared to more "traditional" estuarine models, the MAE is often nutrient limited due to the lack of inorganic nutrient influx related to low runoff and river discharge (Mooney and McClelland, 2012; Bruesewitz et al., 2013). Episodic high freshwater inflows to the MAE significantly impact POM source and concentration (Mooney and McClelland, 2012; Lebreton et al., 2016), yet it remains unclear how storm events affect phytoplankton communities in the MAE. Phytoplankton communities in the MAE have not been systematically characterized, but in this region, chromophytes, haptophytes, pelagophytes, and dinoflagellates generally are the dominant species (Qian et al., 1996).

The objective of this study was to elucidate the impact of storm events on temporal and spatial patterns of POM and phytoplankton communities in the MAE. Previous work in the MAE determined that large storm events lead to elevated POM concentrations and depleted δ <sup>13</sup>C values of particulate organic carbon (POC) (Mooney and McClelland, 2012). Specifically, we hypothesized that high salinity drought conditions would have low POM concentrations, relatively enriched δ <sup>13</sup>C-POC, and smaller-celled phytoplankton (e.g., cyanobacteria). Inversely, low salinity post-flood conditions would have elevated POM concentrations, relatively depleted δ <sup>13</sup>C-POC, and larger-celled phytoplankton (e.g., diatoms). Phytoplankton community structure was determined using established algorithms of accessory pigments, including fucoxanthin, which is mainly produced by diatoms, and zeaxanthin, which represents cyanobacteria (Qian et al., 2003).

#### MATERIALS AND METHODS

#### Site Description

The MAE is a shallow, bar-built estuary spanning approximately 200,000 acres with an average depth of 2 m (Diener, 1975; Armstrong, 1982). The system receives low base freshwater inflow and sporadic pulses from the Mission and Aransas rivers which discharge into Copano Bay (**Figure 1**). These two rivers have similar drainage areas (2,000–2,700 km<sup>2</sup> ), but the Aransas River is impacted by cultivated crops and wastewater treatment plants while the Mission River is relatively pristine (Mooney and McClelland, 2012).

Compounded with limited exchange with the Gulf of Mexico via the tidal inlet at the Port Aransas Shipping Channel, calculated water replacement times within the estuary are as long as 3 years (Armstrong, 1982; Solis and Powell, 1999). Lateral exchange with San Antonio Bay to the north and Corpus Christi Bay to the south can also have an impact on estuarine water quality. For example, during large flood events, the San Antonio

and Guadalupe rivers discharge freshwater into San Antonio Bay which can make its way into the estuary via Mesquite Bay and the Intracoastal Waterway (**Figure 1**; Armstrong, 1982; Longley, 1994).

The Mission-Aransas National Estuarine Research Reserve (MANERR) runs a system-wide monitoring program (SWMP) that continuously records water quality data and collects monthly nutrient samples at four stations along the inshore to offshore transect of our sampling region (**Figure 1**). Two of these stations are located in Copano Bay, a secondary bay more heavily influenced by freshwater inflow due to discharge via the adjacent Mission and Aransas rivers. The Copano Bay West station, located at the mouth of the Aransas River, represents the riverine endmember station in this system while the Copano Bay East station exchanges more openly with the adjacent Aransas Bay. The Aransas Bay station reflects the influence of freshwater inflows on the estuary with increased distance from the river endmember. The station located at the Port Aransas Shipping Channel (Ship Channel station hereafter), which is as the major tidal inlet with the Gulf of Mexico, represents the marine endmember in the MAE.

#### Sample Collection, Water Quality, and Nutrient Analysis

From 2012 to 2015, surface water samples (0–0.5 m) were collected twice a month from March through October, and once a month from November through February at each of the four SWMP stations. Briefly, 2–4 L of surface water were collected from each station using acid-cleaned high density polyethylene bottles, kept on ice, and transported back to the University of Texas Marine Science Institute for filtration on the day of collection. Particles were obtained by filtering 85 to 1000 mL of the water onto a pre-combusted (450◦C, 5 h) glass fiber filter (0.7 µm Whatman GF/F, 47 mm diameter) using low vacuum pressure, and triplicate filters were stored at −80◦C until elemental, pigment, and DNA analyses.

At each station, salinity and temperature (◦C) were measured continuously every 15 min by YSI 6600 V2 sondes (Yellow Springs Instruments) mounted on sampling platforms 0.5 m above the benthos. The MANERR also collects and analyzes monthly surface water nutrient samples at each station. After collection, samples are immediately stored in the dark on ice. Inorganic nutrient (NO<sup>−</sup> 3 , NO<sup>−</sup> 2 , NH<sup>+</sup> 4 , and PO3<sup>−</sup> 4 ) concentrations are determined with a QuAAtro nutrient autoanalyzer (Seal Analytical) using standard colorimetric methods (precision is ± two standard deviations for each parameter). NO<sup>−</sup> 3 , NO<sup>−</sup> 2 , and NH<sup>+</sup> <sup>4</sup> were summed together and presented as dissolved inorganic nitrogen (DIN). All National Estuarine Research Reserve (NERR) water quality data is available for download at the Centralized Data Management Office (CDMO, http://cdmo.baruch.sc.edu).

## POC and PN Concentrations and Source Characterization

Half of one 47 mm filter was used per time point for carbon and nitrogen determination. The filters were acidified in a fumigation chamber with concentrated (12 N) hydrochloric (HCl) acid for 24 h to remove inorganic carbon (Hedges and Stern, 1984). Following acid fumigation, the filters were allowed to vent off excess acid for 4–6 h and then dried at 60◦C for at least 24 h. POC and particulate nitrogen (PN) concentrations (µM) and stable isotope ratios (h) were determined using an Elementar Micro Cube elemental analyzer (Elementar Analysensysteme GmbH, Hanau, Germany) interfaced to an Isoprime VisION isotope ratio mass spectrometer (Isoprime Ltd., Stockport, UK) at the UC-Davis Stable Isotope Facility (UCD-SIF). Stable isotope values were expressed in δ notation relative to carbon in Vienna PeeDee Belemnite and nitrogen in air. Precision at the UCD-SIF for C/N content is within 5%, for δ <sup>13</sup>C is ± 0.2h, and for δ <sup>15</sup>N is ± 0.3h. It should be noted that HCl fumigation of previously frozen filters has been shown to lead to artificially lower δ <sup>15</sup>N-PN values (0.1 h), but this added uncertainty was acceptable given the natural variability in environmental POM samples (Lorrain et al., 2003).

### Pigment Analysis and Phytoplankton Community Composition

The second (whole, untreated) 47 mm GFF filter per time point was used for pigment analysis. Pigment extraction followed the protocol of Sun et al. (1991, 1994). Briefly, pigments on filters were extracted twice with 3 mL of 100% acetone by sonication in chilled water for 15 min in the dark. The extracts were combined and analyzed using high performance liquid chromatography (HPLC) coupled with a UV-vis detector. The pigment analysis followed the protocol of McTigue et al. (2015). Briefly, a binary gradient of 28 mM tetrabutylammonium acetate (TBA) in methanol (30%:70%, v:v; eluent A) and methanol (eluent B) was used. Eluent B was ramped from 5 to 95% in 22 min, and held for 7 min before decreasing to 5% within 2 min. A C<sup>8</sup> HPLC column (Agilent Eclipse XDB, 3.5 µm, 4.6-mm diameter × 150-mm length) was used, and the eluted pigments were detected by visible absorbance (wavelength = 450 nm). Concentrations were determined by comparing pigment peaks of equal retention time to those of certified commercial standards (DHI, VWR, and Sigma-Aldrich). Analyzed pigments included chlorophyll a, and the accessory pigments fucoxanthin, peridinin, zeaxanthin, prasinoxanthin, 19-hex-fucoxanthin, 19 but-fucoxanthin, alloxanthin, and chlorophyll b.

Chlorophyll a is targeted as a proxy for fresh algal POM. Specific microalgal groups can be detected using the accessory pigments: Chlorophyll b for chlorophytes; fucoxanthin in diatoms; peridinin in dinoflagellates; prasinoxanthin in prasinophytes; 19-hex-fucoxanthin in haptophytes; and zeaxanthin in cyanobacteria, which includes prochlorophytes (Wright et al., 1996; Morata and Renaud, 2008; Szymczak-Zyla et al., 2008). The relative chlorophyll a contributions of different algal groups to total phytoplankton biomass were calculated using the pigment algorithms summarized by Qian et al. (2003) (**Table 1**), which were based on previous studies and the analysis of phytoplankton cultures (Letelier et al., 1993; Andersen et al., 1996; Mackey et al., 1996; Wright and van den Enden, 2000). Negative algal group contributions were removed and reported as zero. The sum of the calculated major phytoplankton groups in the total chlorophyll a was normalized to 100% in order to TABLE 1 | Pigment equations used for separating chlorophyll a biomass between the major phytoplankton taxa in the MAE (Qian et al., 2003).


*Chl a, chlorophyll a; Chl b, chlorophyll b; cyano, cyanobacteria; prymn, prymnesiophytes; pela, pelagophytes; dino, dinoflagellates; diat, diatoms; cryp, cryptophytes; pras, prasinophytes. Note that cyanobacteria include prochlorophytes.*

compare compositional changes of phytoplankton spatially and temporally.

### DNA Sequencing

To identify specific cyanobacterial species of the phytoplankton community, DNA was extracted from the GF/F filters (untreated half filter) using the PowerMag Soil DNA isolation kit (MO BIO) according to the manufacturer's instructions. The sequencing (Illumina Miseq) was performed using 28F and 519R to amplify the 500 bp region of 16S rRNA gene by the Research and Testing Laboratory (RTL) (Lubbock, Texas). The sequence data were analyzed according to the bioinformatics pipeline of RTL. Operational taxonomic unit (OTU) clustering was performed at 97% using USEARCH before classification against a 16S database derived from GenBank.

#### Statistics

Using R version 3.3.1 (http://www.r-project.org; R Core Team, 2013), a pairwise Pearson correlation test was performed between the POM characteristics, algal group biomass, water quality parameters, and nutrient concentrations at each station throughout the study period, with a significance cutoff of p < 0.05. Principal component analysis (PCA) was performed on data from Copano Bay West, including phytoplankton composition (%), DIN, PO3<sup>−</sup> 4 , salinity, and temperature using MATLAB (Xue et al., 2011).

#### RESULTS

#### River Discharge, Salinity, and Surface Water Temperature

Freshwater is delivered to Copano Bay from the Mission and Aransas rivers, so daily mean discharge data from both rivers were combined for the study period, October 30, 2012 to October 29, 2015 (**Figure 2A**). From October 2012 to early March 2015 (28 months) there was only one combined discharge event above 100 m<sup>3</sup> s −1 , representing a prolonged "dry period." In contrast, from March 2015 to the end of October 2015 (4 months) there

were three such events, indicating a "wet period" (gray box in **Figure 2A**). The average daily discharge during this wet period was 8.7 m<sup>3</sup> s −1 , more than 10 times higher than during the dry period (0.8 m<sup>3</sup> s −1 ).

During the dry period, average daily salinity at all stations ranged from 26 to 42 (**Figure 2B**). Copano Bay West salinity decreased from 35 to 7 during late March 2015 due to a major storm event (**Figure 2B**). Over the next month, salinity at this station rose to 23 before falling to 2 in late April following a second major storm event. Over the next month, salinity at Copano Bay West rose to 10 before falling to 0 in mid-May as a consequence of a third major storm event. The recurrence of these major storm events over a period of approximately 2 months kept salinity at this station low (<10) until early September 2015. Copano Bay East salinity followed a similar pattern as Copano Bay West in response to the spring 2015 storm events, albeit with a slower, steadier decline (**Figure 2B**), presumably due to its more open location at the mouth of Copano Bay. From March to July 2015 salinity decreased from 32 to 4 and remained below 10 until late July 2015. Similarly, Aransas Bay salinity decreased from 31 to 7 over this 4-month period, but by late July 2015 had reached values of 30 due to its closer proximity to and flushing with the inlet with the Gulf of Mexico (Port Aransas Shipping Channel) (**Figure 2B**). The Ship Channel station was the least impacted by the spring 2015 storm events. However, salinity at this station did respond to these events by falling from 35 to 22 from April to late May 2015 (**Figure 2B**).

Average daily surface water temperatures at Copano Bay West and the Ship Channel displayed consistent annual patterns throughout the study period (**Figure 2C**). This is representative of surface water temperature patterns across the MAE (unpublished data, MANERR). During June temperatures typically reach ∼30◦C and are sustained until September, when they begin to fall until reaching annual lows of ∼10◦C between December and February (**Figure 2C**).

#### Spatial and Temporal Patterns of Inorganic Nutrients and Chlorophyll a

During the dry period, concentrations of DIN, dominated by NH<sup>+</sup> 4 (∼60% of DIN, data not shown), were generally low, punctuated by episodic peaks throughout the year, ranging from 0 to 5.9 µmol-N L−<sup>1</sup> at all stations (**Figure 3A**). During fall 2013, Aransas Bay and Ship Channel DIN concentrations peaked at 3.3 and 5.9 µmol-N L−<sup>1</sup> , respectively. A similar seasonal pattern was seen at both stations in 2014, along with elevated summer and fall concentrations at Copano Bay East. The major storm events during spring 2015 significantly increased DIN concentrations, most notably in Copano Bay (**Figure 3A**). At Copano Bay West, DIN increased from 0.2 to 23.3 µmol-N L−<sup>1</sup> between April 10 and May 20, 2015. Concurrently, at Copano Bay East, DIN increased from 4.5 to 14.1 µmol-N L−<sup>1</sup> . These elevated concentrations were dominated by NO<sup>−</sup> 3 , making up 99 and 67% of DIN at Copano Bay West and East, respectively (data not shown). Values at both Copano Bay stations dropped to dry

period levels by early June 2015 and remained low throughout the study period (**Figure 3A**).

PO3<sup>−</sup> 4 concentrations ranged from 0 to 1.2 µmol-P L−<sup>1</sup> at all stations throughout the dry period, remaining relatively low over time (**Figure 3B**). Copano Bay PO3<sup>−</sup> 4 concentrations began to increase concurrently with DIN in late May 2015, however in contrast to DIN, which peaked briefly in May-June, PO3<sup>−</sup> 4 concentrations continued to rise throughout the wet period. Copano Bay West PO3<sup>−</sup> 4 peaked at 7.0 µmol-P L−<sup>1</sup> in early August 2015 (**Figure 3B**). In contrast, at Copano Bay East, Aransas Bay, and the Ship Channel, maximum values were lower (4.9, 2.8, and 1.7 µmol-P L−<sup>1</sup> , respectively), and did not occur until late October 2015 (**Figure 3B**).

Chlorophyll a concentrations across the estuary remained < 5 µg L−<sup>1</sup> during the dry period, but showing a seasonal pattern of elevated concentrations during spring-summer (**Figure 3C**). Following the major storm events of spring 2015, however, chlorophyll a concentrations increased from 1 to 12 µg L−<sup>1</sup> between June 4 and 10, 2015 at Copano Bay West (**Figure 3C**). Values at this station fluctuated from 3 to 13 µg L−<sup>1</sup> for 2 months until dropping to 1 µg L−<sup>1</sup> in late August 2015. At Copano Bay East, chlorophyll a increased from 2 to 12 µg L−<sup>1</sup> between June 10 and July 9, 2015, showing a lag (∼1 month) in response to the storm events compared to Copano Bay West (**Figure 3C**). Values at Copano Bay East remained > 9 µg L−<sup>1</sup> for 2 weeks. Similarly, in Aransas Bay, chlorophyll a increased from 1 to 16 µg L−<sup>1</sup> between June 10 and 23, 2015. Although this station had the highest overall value during the study period, chlorophyll a concentrations there remained < 2 µg L−<sup>1</sup> throughout the rest of the summer (∼2 months), showing a shorter bloom than seen in Copano Bay. At the Ship Channel, chlorophyll a remained < 2 µg L−<sup>1</sup> throughout the study period, with the exception of July 2015 when it peaked at 4 µg L−<sup>1</sup> , concurrent with the summer peaks at the other stations (**Figure 3C**).

#### Spatial and Temporal Patterns in POM Concentration and Isotopic Values

During the dry period, POC concentrations at Aransas Bay, and Copano Bay West and East were elevated in spring (**Figure 4A**). However, during the wet period, POC concentrations in Copano Bay were the highest of the study period. From 2012 to 2015, POC concentrations ranged from 23 to 653 µmol-C L−<sup>1</sup> at Copano Bay West, 14 to 329 µmol-C L−<sup>1</sup> at Copano Bay East, 22 to 187 µmol-C L−<sup>1</sup> at Aransas Bay, and 10 to 89 µmol-C L−<sup>1</sup> at the Ship Channel station. The highest concentrations at Copano Bay West, Aransas Bay, and the Ship Channel during the dry period were 342, 187, and 89 µmol-C L−<sup>1</sup> , respectively, and peaked in March 2013. POC concentration at Copano Bay East during the dry period peaked at 152 µmol-C L−<sup>1</sup> in early April 2014. During the wet period, POC concentrations were higher at the two Copano Bay stations (∼3-fold) than the Aransas Bay and Ship Channel stations. Copano Bay West POC concentrations began to increase in late March 2015 and peaked at 653 µmol-C L−<sup>1</sup> in early August 2015, but did not exceed 300 µmol-C L−<sup>1</sup> until early July 2015. At Copano Bay East, POC concentrations

began to increase in early June 2015 and peaked at 329 µmol-C L −1 in early July 2015.

Aransas Bay; SC, Ship Channel.

δ <sup>13</sup>C-POC in the MAE showed similar patterns at all three bay stations (Copano Bay East, Copano Bay West, Aransas Bay) during the study period; values ranged from −28.2 to −20.8h, with a shift from relatively enriched to depleted values in June 2015 (**Figure 4B**). Copano Bay West displayed the greatest variation in δ <sup>13</sup>C-POC, accounting for the minimum and maximum values seen across the estuary throughout the study period. This may be due to the station's close proximity to the Aransas River mouth, making it more heavily influenced by freshwater inputs. The most enriched values at all three bay stations (excluding the Ship Channel) were recorded from July to September 2014. The more enriched δ <sup>13</sup>C-POC values correlated with higher salinities in Copano Bay (R <sup>2</sup> = 0.70; p < 0.001) and Aransas Bay (R <sup>2</sup> = 0.66; p < 0.001) throughout the entire study period, although no increase in δ <sup>13</sup>C was observed during summer 2013, when salinities and temperatures were nearly identical to those during summer 2014. The most depleted δ <sup>13</sup>C values at the three bay stations over the study period occurred in June 2015, showing a response to the storm-driven influx of freshwater and leading to more depleted average δ <sup>13</sup>C-POC values at Copano Bay West during the wet period (−26.3 ± 1.1h) relative to the dry period (−24.1 ± 1.2h).

PN and POC concentration patterns were similar (R 2 = 0.59; p < 0.007), with highest concentrations detected during summer 2015, but PN was more variable than POC (**Supplementary Figure 1**). PN concentrations ranged from 4 to 65 µmol-N L−<sup>1</sup> at Copano Bay West, 3 to 34 µmol-N L−<sup>1</sup> at Copano Bay East, 3 to 35 µmol-N L−<sup>1</sup> at Aransas Bay, and 3 to 33 µmol-N L−<sup>1</sup> at the Ship Channel station. δ <sup>15</sup>N-PN in the MAE, ranging from 3.6 to 12.7h, showed less noticeable patterns over time than δ <sup>13</sup>C-POC (**Supplementary Figure 1**).

Molar C/N ratios of POM ranged from 1.7 to 13.7 across the MAE and stayed relatively constant during the dry period (**Figure 4C**). During the wet period, there was a shift toward higher ratios (C/N > 10) from the end of June to September 2015 at Copano Bay East and West. This C/N ratio increase occurred as POC concentrations increased, concurrent with a depletion in POC δ <sup>13</sup>C (**Figure 4**).

### Spatial and Temporal Patterns in Phytoplankton Community Composition

The average percentage of total chlorophyll a accounted for by individual phytoplankton taxa showed that cyanobacteria and diatoms were the dominant taxa in the MAE during both the dry and the wet periods, but the community composition shifted after the major storm events of spring 2015 (**Figure 5**; **Table 2**). At Copano Bay West, both taxa displayed distinct shifts in average biomass between the dry and wet periods (**Figure 5A**). During the dry period, the average biomass of cyanobacteria and diatoms were 29 ± 3 and 43 ± 4%, respectively (**Table 2**). In contrast, during the wet period these averages were 59 ± 9 and 17± 5% for cyanobacteria and diatoms, respectively, showing a shift toward a cyanobacteria-dominated system. A similar trend was observed at the Copano Bay East and Aransas Bay stations, albeit with a more muted response in Aransas Bay (**Table 2**; **Supplementary Figure 2**). Further DNA sequencing of the Copano Bay West samples showed that the cyanobacterial community in the dry period were dominated by Prochlorococcus spp. (56–61%), followed by Cyanobium spp. (19–24%) and Synechococcus spp. (16–19%); during the wet period, Cyanobium spp. became the dominant species (75–78%) from June to August, but decreased to 35% in October 2015 (**Supplementary Figure 3**).

Unlike the three bay stations, phytoplankton community composition at the Ship Channel was similar during the dry and wet periods (**Figure 5B**; **Table 2**). The average biomass of cyanobacteria and diatoms during both periods were 22 ± 4 and 52 ± 5%, respectively. Average dinoflagellate biomass increased from 7 ± 1 to 12 ± 4% during the wet period, with an accompanying decrease in average pelagophyte biomass. DNA sequencing results showed that Synechococcus spp. dominated the Ship Channel cyanobacteria community (92%) in 2013, but the community became more diverse during the wet period with contributions from Synechococcus spp. (42%), Prochlorococcus spp. (31%), and Cyanobium spp. (25%) (**Supplementary Figure 3**).

## DISCUSSION

#### POM Source Following Storm Events

Freshwater discharge into the MAE has been variable over the past decade; multi-year droughts with very low baseflows are interspersed with large storm events occurring at high frequency for many months at a time, as seen during the study period (**Figure 2A**) and in previous studies (Mooney and McClelland, 2012; Bruesewitz et al., 2013; Lebreton et al., 2016). These sporadic freshwater inflows significantly influence system-wide salinity, with a decreasing degree of influence along the inshore to offshore transect (**Figure 2B**). During summer 2015, there was a substantial peak in POC concentration in Copano Bay, surpassing any values observed during the dry period, when salinities were more stable across the estuary (**Figures 2B**, **4A**). These results fit the pattern described previously in this system by Mooney and McClelland (2012): during 2007 (a wet year), POC concentrations in Copano Bay remained elevated for many months following major storm events in July before showing a sharp decline as salinities began to rise in spring 2008 (a dry year). The observed increase in POC during the present study can therefore be attributed to the freshening of the estuary as a result of the exponential increase in river discharge.

Sources of POC in estuaries can be inferred from the δ <sup>13</sup>C-POC values. δ <sup>13</sup>C-POC across the MAE are typically ∼ −24h under drought conditions, but following major storm events decrease to as low as −28h (**Figure 4B**; Mooney and McClelland, 2012). The depletion in δ <sup>13</sup>C-POC is consistent with an expected depletion in dissolved inorganic carbon δ <sup>13</sup>C-DIC concurrent with decreasing salinity associated with the mixing of riverine (δ <sup>13</sup>C-DIC ∼ −11h, Fogel et al., 1992) and marine (δ <sup>13</sup>C-DIC ∼ 0h, Fogel et al., 1992) water masses, with a muted effect at the Ship Channel as a result of flushing with nearshore Gulf waters (**Figure 4B**; Mooney and McClelland, 2012). Riverine δ <sup>13</sup>C-DIC can drive down the estuarine δ <sup>13</sup>C-POC through uptake and fixation by estuarine phytoplankton (**Figure 4B**; Fogel et al., 1992). Delivery of freshwater phytoplankton from the rivers to the bay may also have contributed to the lower δ <sup>13</sup>C signal that we observed (δ <sup>13</sup>C = −31.4h, Lebreton et al., 2016). While depleted δ <sup>13</sup>C values in river water are often attributed to C3 plant contributions from the surrounding terrestrial landscape, this is not a viable explanation in this system because POM in the Mission and Aransas rivers becomes enriched in δ <sup>13</sup>C-POC during storm flow conditions (δ <sup>13</sup>C = −22 to −20h, Mooney and McClelland, 2012). In any case, the fact that POC concentrations and stable isotope values in the bay changed gradually over several weeks following the storms (**Figures 4A,B**) points to a buildup of in-situ production rather than organic matter delivered to Copano Bay via the Mission and Aransas rivers. In-situ production likely played a key role in POM dynamics throughout the wet period, as the result of initial inoculation by river-borne phytoplankton followed by increased growth of phytoplankton within the bay under relatively fresh conditions.

Following major storm events during the summer of 2015, phytoplankton biomass (chlorophyll a) at all bay stations increased, confirming enhanced in-situ production (**Figure 3C**). Both chlorophyll a and POC peaked first in May (**Figures 3C**, **4A**). Then, POC reached exceptionally high levels in August, which occurred as a slow buildup of the standing stock through the summer. Chlorophyll a showed sustained high levels during the summer. These dynamics indicate two distinct peaks of phytoplankton production during the wet period. DIN concentrations also peaked in May, reflecting riverine input, then dropped precipitously, and remained low, likely supporting the initial increase in phytoplankton production (**Figure 3A**). Concurrent with the increased production in late summer, the C/N of Copano Bay POM increased (>10), most notably at Copano Bay West (**Figure 4C**). After the May phytoplankton community used the available river-derived DIN, the established phytoplankton community likely became nitrogen-limited, indicating that the rising C/N may be attributed to gradual nitrogen limitation of phytoplankton in Copano Bay (**Figures 3A,C**, **4C**; Goldman, 1980).

### Phytoplankton Community Shifts after Storm Events

Disturbances to water column hydrology, specifically drops in salinity related to hurricanes and other major storm events, have been well documented to have significant impacts on estuarine phytoplankton community composition (Qian et al., 1996; Peierls et al., 2003; Paerl et al., 2014b). For example, following the landfall of Hurricane Floyd in the fall of 1999, phytoplankton community composition in the Pamlico Sound, North Carolina, underwent numerous shifts: dinoflagellates, which typically have a strong spring bloom in the system, were rare outside of a moderate bloom the following spring, along with significant cyanobacterial blooms (exceeding pre-hurricane levels) during the subsequent two summers (Peierls et al., 2003).

TABLE 2 | Average biomass percentages (%) of each phytoplankton taxa at all stations in the MAE during both the dry and wet periods (all values are presented as average ± standard error).


*CW, Copano Bay West; CE, Copano Bay East; AB, Aransas Bay; SC, Ship Channel.*

In our study, phytoplankton community structure shifted dramatically after the 2015 spring storms with the estuarine-wide drop in salinity, particularly at Copano Bay West (**Figure 5A**). From April 30 to June 4, high percentages of diatom biomass corresponded with the initial peaks in chlorophyll a, POC, and DIN (**Figures 3A,C**, **4A**, **5A**), suggesting that the initial DIN pulse may have led to the rapid growth of diatoms. However, the continued increase in freshwater inflow associated with subsequent major storm events during the summer led to an increase and dominance by small-celled cyanobacteria, in contrast to an increase in large cell phytoplankton, as was hypothesized (**Figure 5**). The largest POC peak and sustained high levels of chlorophyll a during the summer reflected cyanobacterial production. Similarly, there was an increase in zeaxanthin (a cyanobacterial indicator) concentration in Aransas Bay during the summer of 1992 when salinities were depressed (<5) as the result of a large freshwater influx into the system (Qian et al., 1996). Prior to June 2015, the spring estuarine phytoplankton community was dominated by diatoms, followed by cryptophytes, dinoflagellates, and prasinophytes. However, during the wet period cyanobacteria became the dominant phytoplankton. For example, cyanobacteria accounted for 59% of total phytoplankton biomass at Copano Bay West during the wet period (**Table 2**; **Figure 5A**). Although slight increases in cyanobacterial biomass were noted during the two previous summers, this class of phytoplankton typically accounted for less than 29% (±3%) of the phytoplankton community prior to the wet period. Furthermore, cyanobacterial dominance of the phytoplankton community in Copano Bay continued until late September, when salinities rose to 13 and 18 at

Copano Bay West and East, respectively (**Table 2**; **Figure 5A**; **Supplementary Figure 2A**). As with POM, a muted response in phytoplankton community composition occurred at Aransas Bay (**Table 2**; **Supplementary Figure 2B**). This variation in phytoplankton community response between the two bays can be explained by the flushing gradient within the system and its exchange with the Gulf of Mexico, which is further illustrated by the more rapid increase in salinity at Aransas Bay following the flood event (**Figure 2B**). Also, although not characterized in this study, lateral water exchange with Mesquite Bay to the north and Corpus Christi Bay to the south is expected to have had an impact on flushing in Aransas Bay (Armstrong, 1982; Longley, 1994). Estuarine and nearshore water masses are exchanged more readily in Aransas Bay due to its close proximity to the Port Aransas Shipping Channel, where community composition and salinity showed a relatively small and brief response to the storm events (**Figures 2B**, **5B**). Most notably, this dominance of cyanobacteria (59% of phytoplankton biomass at Copano Bay West) was not seen in either of the previous two summers, when salinities were higher, and is therefore attributed to the formation of a niche for either a riverine species or an existing estuarine species with a tolerance for low salinities. Note that there is uncertainty in our calculated phytoplankton composition, since cellular content of pigments in phytoplankton cells can change with physiological state, such as nutrients and light (Qian et al., 2003). However, this should not significantly affect our results, as we analyzed more than 3 years of time-series data with the same cycling of light levels and similar nitrogen-limited conditions.

Environmental conditions may have played an important role in enhancing the growth of cyanobacteria in this system. The large pulses of freshwater, together with high surface temperatures (∼30◦C) from June to September, may have contributed to the cyanobacteria bloom. In the PCA biplot of relative phytoplankton biomass and various environmental parameters for Copano Bay West, cyanobacteria were particularly enriched in the samples from the wet period, well separated from the rest of the samples along PC 1 (**Figure 6**). From the PCA, the development of cyanobacteria appears to be related to temperature, PO3<sup>−</sup> 4 , and water freshening (opposite of salinity). Although cyanobacteria tend to have slow growth rates compared to other phytoplankton taxa, they have been shown to respond more strongly to temperature in field and laboratory studies (Juhl and Murrell, 2005; Reynolds, 2006; Paerl et al., 2014a). Three of the major phytoplankton taxa in the MAE, diatoms, dinoflagellates, and cyanobacteria, have growth optima ranges of 15–18◦ , 20–23◦ , and 25–30◦C, respectively (Paerl et al., 2014a; Dorado et al., 2015). In addition, zeaxanthin shows stronger temperature dependence than other accessory pigments in the New River Estuary, North Carolina (Hall et al., 2013). Therefore, higher temperature, together with lower salinity and high PO3<sup>−</sup> 4 concentrations, appear to be the major factors supporting the cyanobacterial bloom after the storms events.

In addition to lowered salinity and increased temperature, DIN and PO3<sup>−</sup> 4 concentrations were higher during the wet period and can therefore be attributed to riverine export of inorganic nutrients (**Figures 3A,B**, **6**). Copano Bay DIN peaked in May 2015, but was drawn down quickly, as discussed previously; this DIN peak corresponded to the higher percentages of diatoms, suggesting their role in DIN uptake. The rapid disappearance of DIN may have also been due to processes such as denitrification and anammox likely occurring in sediments of this shallow water system (Bruesewitz et al., 2013). The subsequent shift toward a cyanobacteria-dominated system even though DIN concentrations remained low from June to September, suggested that cyanobacteria thrived under nitrogen-limited conditions (**Figures 3C**, **5A**, **6**). Copano Bay PO3<sup>−</sup> 4 concentrations increased more gradually, and began reaching peak concentrations across the estuary as chlorophyll a concentrations began to drop, suggesting that PO3<sup>−</sup> <sup>4</sup> was rapidly recycled following the cyanobacterial bloom; this pattern was consistent with a previous study in the system (**Figures 3B,C**; Mooney and McClelland, 2012). Relative to the rapid removal of DIN, concentrations of PO3<sup>−</sup> 4 remain high throughout the wet period, indicating inefficient removal of phosphorus by the system (e.g., via microbial uptake and/or particle adorption). Cyanobacterial species that were observed in a recent study of the Breton Sound Estuary, Louisiana generally had low salinity tolerances, and it was determined that temperature, salinity, and DIN concentrations best explained annual patterns in phytoplankton community composition (Riekenberg et al., 2015). Additionally, cyanobacteria tend to be buoyant and become established in stable water columns, such as those experienced in most marine systems during the summer when the water column is stratified (Riekenberg et al., 2015). Low salinities and high temperatures created an ideal situation for the shift to a cyanobacteria-dominated phytoplankton community in the MAE during summer 2015. Additionally, it is possible that decreased filter feeding pressure contributed to the sustained increase in cyanobacterial biomass. A study on the impact of flood events on oyster populations in Copano Bay found that depressed salinities during the summer of 2007 led to reductions in oyster abundance and filtration rates (Pollack et al., 2011), which could lead to the persistence of ungrazed phytoplankton biomass in this shallow (<2 m) water column.

DNA sequencing data showed that in addition to the phytoplankton community shifting to become more cyanobacteria-dominated, the cyanobacterial community itself shifted from being dominated by Prochlorococcus spp. to Cyanobium spp. after the storm events (**Table 2**; **Supplementary Figure 3A**). Cyanobium spp. is a unicellular autotrophic coccoid picocyanobacteria (0.6–4.5 µm cell length × 0.3–1.7 µm cell width) (Komärek and Cepak, 1998; Komarek et al., 1999). Given their small size, some cells may have passed through the combusted glass fiber filter (0.7 µm pore size), leading to a potentially low estimate of Cyanobium spp. abundance. If we were actually underestimating the abundance of cyanobacteria, then this further strengthens our conclusions of cyanobacterial dominance. In a study of cyanobacterial communities in Chesapeake Bay, temperature was the most dominant environmental factor controlling picocyanobacterial populations; Cyanobium spp. occurrence was associated with lower salinities (Cai et al., 2010). In a study of cyanobacterial communities in the upper mixed layer (0–10 m) of the Baltic Sea, temperature and salinity were the potential drivers in structuring summer cyanobacterial communities (Bertos-Fortis et al., 2016). Picocyanobacteria were present in this layer year-round, and Cyanobium spp., in particular, have the ability to thrive in dynamic environments (such as the MAE) which is the reason for their year-round presence despite seasonal changes (Bertos-Fortis et al., 2016). Additionally, opportunistic picocyanobacteria occurrence was mainly explained by changes in temperature and salinity which matches the pattern seen in the present study; Cyanobium spp. dominated during the low salinity summer 2015 wet period, and Prochlorococcus spp. dominating during the high salinity summers of 2013 and 2014 (**Figures 2B**, **4A**; **Supplementary Figure 3A**; Bertos-Fortis et al., 2016).

Although Cyanobium spp. are not known nitrogen fixing bacteria, DIN assimilation can be the primary fate of DIN in the MAE during drought conditions even when measured DIN concentrations are low (**Figure 3A**; Stal, 2009; Hou et al., 2012; Bruesewitz et al., 2015). This indicates that high DIN regeneration coupled with high uptake, along with a high surface to volume ratio due to their small size, may have allowed Cyanobium spp. to continue flourishing for several months after DIN concentrations had dropped (**Figure 3A**; **Supplementary Figure 3A**). Overall, picocyanobacteria dominated the summer cyanobacterial community in the MAE throughout the study period, but lower salinities related to consecutive major storm events led to Cyanobium spp. dominating the picocyanobacterial community during the summer of 2015.

## CONCLUSIONS AND IMPLICATIONS

POM source and concentration can indicate changes in the source of estuarine production, but this may be too coarse to elucidate fine changes in the source of production in the MAE. Using photosynthetic pigments, we determined that periods of drought favor a more diverse assemblage of phytoplankton, with relatively high diatom and dinoflagellate biomass. Major flood events led to a picocyanobacteria-dominated phytoplankton community, specifically Cyanobium spp., which was present in summer picocyanobacterial communities throughout the study period. The timing of the consecutive storm events in spring 2015 had a profound impact on the ability of this species to dominate, as rising temperatures and depressed salinities opened up a niche for this species to flourish. Changes in phytoplankton community composition can drastically affect the transfer of energy to higher trophic levels, thus having major implications for ecosystem-wide productivity (Cloern and Dufford, 2005), and is a topic for further study in this system. This study also suggests that this system is very efficient at removing the riverine-derived DIN after the storm events, but the flux of PO3<sup>−</sup> 4 to the coastal ocean through the ship channel may be significant (Bhavya et al., 2016; Krishna et al., 2016).

In a region where recreational and commercial fishing are important to the local economy, impacts on ecosystem productivity should be carefully monitored. Given the importance of cultivated crops in the Mission-Aransas watershed, understanding the impact of freshwater inflows on the phytoplankton community in this estuary is key to the management of these vital sporadic freshwater inflows. Furthermore, cyanobacteria show extreme plasticity toward changes in environmental conditions and aquatic systems across the world, and cyanobacterial bloom frequency and duration are projected to increase in response to climate change (Paerl et al., 2014a; Bertos-Fortis et al., 2016).

#### AUTHOR CONTRIBUTIONS

NR, AH, and ZL conceived, designed, and wrote the paper.

#### ACKNOWLEDGMENTS

We thank J. Liu, S. Liu, K. Jackson, J. Xue, and C. Bonsell for help with pigment analyses and lab support. Thanks also to the MANERR staff: J. Tunnell, C. Hyatt, C. Croulet, and H. Samberson for help with sample collection and logistical support. We are grateful for substantive comments on previous drafts of this paper from E. Buskey, J. McClelland, and three reviewers. N. Reyna was partially supported by the University of Texas

#### REFERENCES


Graduate School Diversity Mentoring Fellowship. This research was partially funded by National Oceanic and Atmospheric Administration (NA15NOS4780185) and by operations grants to the Mission-Aransas National Estuarine Research Reserve from the Office of Coastal Management, National Oceanic and Atmospheric Administration. System-Wide Monitoring Program data was collected by the research staff of the Mission-Aransas NERR and hosted by the Central Data Management Office of the National Estuarine Research Reserve System. Publication supported in part by an Institutional Grant (NA14OAR4170102) to the Texas Sea Grant College Program from the National Sea Grant Office, National Oceanic and Atmospheric Administration, U.S. Department of Commerce.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2017.00043/full#supplementary-material

Supplementary Figure 1 | (A) Particulate nitrogen (PN) concentration and (B) δ <sup>15</sup>N-PN (h) at all stations, October 30, 2012 to October 29, 2015. In all panels, the wet period is indicated by a shaded box. CW, Copano Bay West; CE, Copano Bay East; AB, Aransas Bay; SC, Ship Channel.

Supplementary Figure 2 | Phytoplankton community composition as percent (%) of total biomass at the (A) Copano Bay East and (B) Aransas Bay stations, October 30, 2012 to October 29, 2015. In all panels, the wet period is indicated by a shaded box.

Supplementary Figure 3 | Cyanobacterial community composition as displayed by cyanobacterial operational taxonomic units (OTUs) at the (A) Copano Bay West and (B) Ship Channel stations.


inflows and phytoplankton communities in a subtropical estuary in the Gulf of Mexico. PLoS ONE 10:e0130931. doi: 10.1371/journal.pone.0130931


Australian sector of the Southern Ocean: comparisons of microscopy and size frequency data with interpretations of pigment HPLC data using the "CHEMTAX" matrix factorisation program. Mar. Ecol. Prog. Ser. 144, 285–298.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Reyna, Hardison and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Short-Term Dissolved Organic Carbon Dynamics Reflect Tidal, Water Management, and Precipitation Patterns in a Subtropical Estuary

Peter Regier 1, 2 and Rudolf Jaffé1, 2 \*

<sup>1</sup> Department of Chemistry and Biochemistry, Florida International University, Miami, FL, USA, <sup>2</sup> Southeast Environmental Research Center, Florida International University, Miami, FL, USA

Estuaries significantly impact the global carbon cycle by regulating the exchange of organic matter, primarily in the form of dissolved organic carbon (DOC), between terrestrial and marine carbon pools. Estuarine DOC dynamics are complex as tides and other hydrological and climatic drivers can affect carbon fluxes on short and long time scales. While estuarine and coastal DOC dynamics have been well-studied, variations on short time scales are less well-constrained. Recent advancements in sonde technology enable autonomous in situ collection of high frequency DOC data using fluorescent dissolved organic matter (FDOM) as a proxy, dramatically improving our capacity to characterize rapid changes in DOC, even in remote ecosystems. This study utilizes high-frequency FDOM measurements to untangle rapid and complex hydrologic drivers of DOC in the Shark River estuary, the main drainage of Everglades National Park, Florida. Non-conservative mixing of FDOM along the salinity gradient suggested mangrove inputs accounted for 6% of the total DOC pool. Average changes in FDOM concentrations through individual tidal cycles ranged from less than 10% to greater than 50% and multi-day trends >100% change in FDOM concentration were observed. Salinity and water level both inversely correlated to FDOM at sub-hourly and daily resolutions, while freshwater controls via precipitation and water management were observed at diel to monthly time-scales. In particular, the role of water management in rapidly shifting estuarine salinity gradients and DOC export regimes at sub-weekly time-scales was evident. Additionally, sub-hourly spikes in ebb tide FDOM indicated rapid exchange of DOC between mangrove sediments and the river channel. DOC fluxes calculated from high-resolution FDOM measurements were compared to monthly DOC measurements with high-resolution fluxes considerably improving accuracy of fluxes (thereby constraining carbon budgets). This study provides a better understanding of short-term DOC dynamics and associated hydrological drivers and indicates the importance of high-frequency measurements to accurately constrain coastal carbon processes and budgets, particularly in coastal systems increasingly altered by hydrologic restoration and climate change.

Keywords: dissolved organic carbon (DOC), mangrove estuary, fluorescence sensor, water management, carbon fluxes

#### Edited by:

Patricia M. Medeiros, University of Georgia, USA

#### Reviewed by:

Piotr Kowalczuk, Institute of Oceanology Polish Academy of Sciences, Poland Heather Erin Reader, Technical University of Denmark, Denmark

> \*Correspondence: Rudolf Jaffé Jaffer@fiu.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 23 September 2016 Accepted: 17 November 2016 Published: 02 December 2016

#### Citation:

Regier P and Jaffé R (2016) Short-Term Dissolved Organic Carbon Dynamics Reflect Tidal, Water Management, and Precipitation Patterns in a Subtropical Estuary. Front. Mar. Sci. 3:250. doi: 10.3389/fmars.2016.00250

## INTRODUCTION

Estuaries are the continental-oceanic interface of the global aquatic carbon (C) cycle. Terrestrially derived organic C is transported to coastal margins through these systems, primarily as dissolved organic carbon (DOC; Hedges et al., 1997). Tidal wetlands and estuaries, which occupy a small portion of world coastlines, account for an estimated third of all organic C buried in coastal sediments (Bauer et al., 2013), serving as a globally relevant C sink (Chmura et al., 2003; Bridgham et al., 2006). DOC also influences biogeochemical cycling within estuaries, including regulating the transport of nutrients and metals (Boyer et al., 1997; Qualls and Richardson, 2003; Yamashita and Jaffé, 2008) and impacts microbial communities (Tranvik, 1998; Fellman et al., 2010). Components of the DOC pool which absorb UV light, known as chromophoric dissolved organic matter (CDOM), control benthic productivity through light attenuation in the water column (Osburn et al., 2009; Ganju et al., 2014).

DOC dynamics vary dramatically across temporal scales (Spencer et al., 2007; Jollymore et al., 2012), from sub-hourly changes in tidal systems (e.g., Bergamaschi et al., 2012) to inter-annual trends (e.g., Evans et al., 2005). Characterizing the patterns and drivers of DOC variability to understand ecological impacts and constrain coastal C budgets thus requires information spanning a wide range of temporal resolutions. Although long-term investigations of coastal DOC dynamics are relatively common (e.g., Chen et al., 2013; Regier et al., 2016), equivalent studies at high temporal resolution are lacking due to the complex logistics and laboratory costs associated with DOC sampling in remote estuarine environments. However, highresolution measurements are critical for temporally dynamic systems like estuaries, where rapid water quality changes observable in high-frequency measurements may not be accurately represented by weekly or monthly sampling intervals (Kirchner et al., 2004; Jollymore et al., 2012; Sobczak and Raymond, 2015).

The solution to this dearth of high-frequency DOC information lies in recent technological advances in multiparameter sondes coupled with sensors measuring fluorescent dissolved organic matter (FDOM), a proxy for DOC (Downing et al., 2009). Sondes combine sensors, data-logging and internal power sources into a single water-proof instrument to achieve high-frequency sampling intervals in situ, eliminating the need for laboratory analysis except to calibrate proxy relationships. These sensors have been applied in a range of environments, including lakes (Watras et al., 2015), streams and rivers (Pellerin et al., 2012; Wilson et al., 2013), wetlands (Ryder et al., 2014), and coastal systems (Downing et al., 2009; Bergamaschi et al., 2012). High-frequency FDOM data provide the resolution necessary to fill in knowledge gaps beyond the scope of conventional monthly sampling frequencies. For instance, high-frequency FDOM measurements have facilitated the development of ecosystemspecific proxies for mercury (Bergamaschi et al., 2012), nutrients (Wilson et al., 2013; Etheridge et al., 2014), and DOC lability (Wilson et al., 2013). Likewise, high-resolution characterization of DOC proxies have provided new insight into processes controlling diel carbon cycles (Watras et al., 2015), rapid changes in hydrologic export of DOC (Bergamaschi et al., 2012; Wilson et al., 2013) and the role of DOC in light attenuation within optically complex waters (Ganju et al., 2014). As such, the application of instruments capable of high-frequency DOC data collection holds great potential for advancing our understanding of temporally dynamic aquatic ecosystems.

In the present study, high-resolution measurements were collected to examine patterns and trends in DOC at sub-hourly time-scales in the coastal estuaries of Everglades National Park (ENP). Landscape-scale restoration efforts underway to re-establish hydrologic connectivity along the Everglades flow-path are expected to alter seasonal timing, quantity and quality of freshwater inflows to ENP (Sklar et al., 2005), but the potential effects on ecosystem function remain uncertain (Estenoz and Bush, 2015). Since hydrology controls DOC dynamics in this system, it is expected that changes in the balance between freshwater (due to changes in management and rainfall) and saltwater intrusion (due to sea-level rise) will alter DOC cycling and export patterns (Orem et al., 2015; Regier et al., 2016). To this end, significant efforts have been placed into understanding the spatial and temporal patterns of DOC quality and quantity across the Everglades landscape (Maie et al., 2006a; Yamashita et al., 2010; Chen et al., 2013). In addition, DOC export patterns and drivers have been investigated at low and high temporal resolutions (Bergamaschi et al., 2012; Cawley et al., 2014; Regier et al., 2016). However, the relationship between rapidly changing and complex estuarine hydrology and high-resolution DOC dynamics for this system remains mostly undetermined. This study applies high-frequency measurements of FDOM to this knowledge gap, to better understand the hydrologic drivers of short-term DOC patterns in the Shark River, located in the mangrove forests of the coastal Everglades. In addition, this study builds on previous characterization of longterm DOC flux drivers (Regier et al., 2016) by linking short-term (daily) to long-term (monthly) DOC export regimes.

## METHODS

## Site Description

Data for this study was collected in the Shark River estuary, located in the coastal ecotone of southwest ENP (**Figure 1**). The Shark River connects vast upstream freshwater wetlands to coastal margins and is situated within the largest contiguous mangrove forest in the United States. Data was collected at SRS-5 (**Figure 1**), a monitoring site maintained by the Florida Coastal Everglades Long Term Ecological Research (FCE-LTER) project. The site is subject to semi-diurnal tides and experiences distinct wet and dry seasons, where the wet season delivers the majority of annual freshwater inflows to the estuary via rainfall and inputs through water management structures (Duever et al., 1994; Saha et al., 2012).

#### Instrument Calibration and Setup

High-frequency data were collected in situ (measured directly in the water column) using an EXO-2 water quality sonde (Yellow Springs Instrument Company (YSI), Ohio). The sensor measuring FDOM had excitation and emission wavelength

ranges of 365 ± 5 and 480 ± 40 nm, respectively, with a range of 0–300 ppb quinine sulfate equivalents (QSE) and resolution of 0.01 ppb QSE reported by the manufacturer. The sonde was calibrated prior to deployment according to manufacturer protocols and again immediately after instrument retrieval. Calibration solutions for FDOM fluorescence were made from quinine sulfate dihydrate diluted with 0.05 M H2SO4, with calibration errors lower than 5% for all deployments. Fluorescence of FDOM is reported in quinine sulfate units (QSU). Prior to deployment, sensors were wrapped in copper tape and surrounded by an anti-fouling sensor guard to minimize impacts of biofouling. Additionally, a wiper was programmed to clean sensor heads every 3 h. In spite of these precautions, sensor drift was detected in two time-series, which were not included in this study. For the data presented here, no drift was detected based on pre and post-calibration. Evidence of erratic turbidity spikes during preliminary deployments indicated interference due to crab activity, known to alter particulate and nutrient dynamics in mangrove sediments (Kristensen and Alongi, 2006). In order to protect against this, mesh was installed over openings in the sensor guard so that water flowed through uninhibited but crabs were unable to interfere with data collection.

#### Data Collection

A total of five time-series were collected between November 2014 and November 2015 averaging 27 days in duration (**Table 1**). Two time-series were collected during the wet-dry seasonal transition (T1 and T2), two during the dry season (D1 and D2) and one during the wet season (W1). The sonde was equipped with six external sensors (temperature/conductivity, pH, dissolved oxygen (DO), turbidity, FDOM, and chlorophyll) and one internal pressure sensor for measuring depth. The sonde was secured to the boardwalk at SRS-5 and situated adjacent to the riverbank with sensor heads located ∼0.5 meters below the lowest low tide mark. Ancillary hydrologic data was retrieved from the USGS gage at Gunboat Island (#252230081021300), located <1 km upstream from SRS-5 (**Figure 1**). As the sonde was not deployed relative to a standard vertical datum, stage values collected by the sonde were not directly comparable between time-series. Therefore, stage data from the USGS gage at Gunboat Island was matched to sonde stage using timelagged correlations. After matching, sonde stage and USGS stage showed strong linear relationships (r > 0.93, p < 0.0001 for all time-series).

Data collected at 15 minute intervals from sonde and USGS sources are denoted as 15-min. Tidally filtered discharge was reported at hourly resolution after applying the PL33 low-pass filter for removal of tidal fluctuations (Flagg et al., 1976). Daily averages of high temporal resolution measurements were also reported, with the first and last day of each time-series excluded since these dates did not capture 24 h of continuous data. Longterm daily salinity data for calculations of δ-salinity was retrieved from the Gulf of Mexico Coastal Ocean Observing System (GCOOS) station GBIF1 (http://data.gcoos.org/index.php) as hourly data and binned to daily averages, where days with <12


TABLE 1 | Average values of selected hydrologic variables for each time-series.

\*Stage and discharge calculated from 15-min data, filtered discharge calculated from hourly data, all other variables calculated from daily data (first and last day of each time-series are clipped).

measurements were excluded. Rainfall data was collected at the Shark River station of the Everglades Depth Estimation Network (EDEN, http://sofia.usgs.gov/eden/). Water management inflows were calculated as the sum of daily average discharge for S12A-D structures (USGS gages, http://waterdata.usgs.gov/).

#### Fluorescence Interference and Corrections

Corrections for thermal quenching of fluorescence were conducted following methods of Watras et al. (2011). Briefly, temperature and FDOM fluorescence were measured simultaneously for water samples collected at SRS-4, SRS-5, and SRS-6 (**Figure 1**) over a broad temperature range (5– 30◦C) exceeding the expected seasonal temperature range for SRS-5. Based on linear regression, a correction factor (ρ = −0.0064 ± 0.0004) was established. A reference temperature (Tr) of 25◦C was used instead of 20◦C to reflect sub-tropical temperature regimes.

Interference from dissolved and suspended materials that absorb or scatter light can alter in situ fluorescence measurements, and impact instruments differently based on geometry of the sensor optics (Downing et al., 2012). For dissolved materials, absorbance of UV radiation by CDOM can be presented as absorbance at 254 nm (A254, unitless) or absorption coefficients (a254) (Hu et al., 2002; Kowalczuk et al., 2010). As a quantitative measurement, a<sup>254</sup> is preferred to A<sup>254</sup> as a<sup>254</sup> values are comparable between studies regardless of path length. However, A<sup>254</sup> values are also reported here to facilitate comparison with values presented by Downing et al. (2012) and Cawley et al. (2014). Conversion of A<sup>254</sup> to a<sup>254</sup> was performed using Equation 2 in Kowalczuk et al. (2010) where path length was 1 cm for both Downing et al. (2012) and Cawley et al. (2014).

The Shark River is characterized by relatively high CDOM absorbance (A254) ranging from 0.1 to 0.6 (a254: 23–138 m−<sup>1</sup> ) based on Cawley et al. (2014). Figure 1 in Downing et al. (2012) indicates this range of absorbance values equates to FDOM attenuation up to ∼15%. Monthly absorbance data collected upstream (SRS-4) and downstream (SRS-6) of the study site (**Figure 1**) spanning 11/2014–05/2015 were used to estimate SRS-5 absorbance as the average of SRS-4 and SRS-6 (Figure S1). Estimated A<sup>254</sup> values for SRS-5 (average A254: 0.44, average a254: 101 m−<sup>1</sup> ) indicated that, at least for the portion of the study period represented in **Figure 1**, attenuation due to dissolved materials accounted of <10% of the FDOM signal based on a closed path sensor with optics at 90◦ (Downing et al., 2012).

To assess fluorescence interference due to suspended solids (measured as turbidity), a turbid solution containing 15 g of mangrove sediment collected at SRS-5 (dried, ground, and sieved through a 30-mesh screen) and ∼4 L of filtered low-absorbance (A254: 0.09, <sup>a</sup>254: 21 m−<sup>1</sup> ) mangrove creek water was mixed (water was filtered using combusted 0.7 µm GF/F filters). The solution was equilibrated for 48 h to negate potential FDOM inputs leached from the sediments (Downing et al., 2012; Lee et al., 2015). The solution was then placed in a constantly stirred vessel connected to a YSI flow-cell containing the FDOM and turbidity sensors with previously acid-washed tubing and a peristaltic pump. Sequential removal of turbidity was achieved by siphoning 250 mL aliquots from the stirred vessel, filtering and then returning the filtrate to solution (Figure S2). As previously observed, highly turbid waters attenuated FDOM fluorescence considerably (Downing et al., 2012; Lee et al., 2015). Less turbid conditions (values < 100 FNU) showed a strong linear relationship (r = 0.99, p < 0.0001, n = 216) to FDOM (Figure S2, values <100 FNU). For this study, turbidity values were generally low with 0.5% of turbidity values exceeding 100 FNU and 1.9% exceeding 25 FNU. Based on linear regression of turbidity values lower than 100 FNU in Figure S2, it is estimated that 25 FNU attenuated <5% of the total FDOM fluorescence signal. Furthermore, spikes in turbidity did not appear to elicit anomalous FDOM signals (e.g., Figure S3). This suggests that, while attenuation of fluorescence by suspended solids likely influences FDOM signals, a simple correction for the effects of turbidity on FDOM signals may not be appropriate (Lee et al., 2015). As such, turbidity corrections were not applied to FDOM measurements (See Section Statistics and FDOM Errors for estimates of error).

#### Establishing the FDOM-DOC Relationship

The relationship between FDOM and DOC was calibrated by collecting grab samples concurrent with in situ FDOM measurements. Due to high seasonal and spatial variability of DOC optical properties in the Shark River (Cawley et al., 2014), the FDOM:DOC relationship was established both with spatial transects and temporal sampling. Four seven-point spatial transects were collected from the marine end member (Ponce De Leon Bay) to the freshwater end member at SRS-4 (**Figure 1**). Transects were collected during wet and dry seasons and seasonal transitions and temporal data were collected through portions of two consecutive tidal cycles. DOC samples were filtered through 0.7 µm GF/F filters and quantified on a Shimadzu TOC analyzer after acidification and purging to remove inorganic C. A total of 53 paired measurements of DOC and FDOM collected from 02/2015 to 04/2016 exhibited strong linear correlation (r = 0.96, p < 0.0001, Figure S4) yielding a conversion equation of FDOM (QSU) = 0.11∗DOC (µM) + 19.06. The root mean square error (RMSE) for this regression was ±9.8 QSU or ±7.5% of the average FDOM value (See Section Statistics and FDOM Errors). Due to the error component associated with converting FDOM fluorescence to DOC concentrations, FDOM measurements were primarily reported in QSU throughout the manuscript while DOC concentrations were reported in µM or ppm for comparison with literature values and to facilitate calculating DOC fluxes. DOC fluxes were calculated as the product of DOC concentration and discharge.

#### Statistics and FDOM Errors

All statistical analyses were performed using the JMP <sup>R</sup> Version 12 statistical package (SAS Institute Inc., Cary, NC, 1989–2007). Results of linear regressions are reported as correlations (r). Changes in FDOM presented as percent were calculated using the average FDOM value for the study (130.8 QSU) for consistency. Non-conservative mixing lines were calculated for regression plots of FDOM vs. salinity (e.g., **Figure 3**), based on previously published methods (Cawley et al., 2014).

Mangrove inputs during FDOM spikes were quantified for three consecutive tidal cycles in T2 with clear ebb tide spikes (**Figure 5**). For each tidal cycle, consecutive data points were connected with lines. Next, a baseline was drawn by connecting the lowest FDOM value at the start of the tidal cycle to the lowest value at the end of the tidal cycle. The baseline was then set to zero, so integrations only included the change in FDOM during the tidal cycle rather than the entire FDOM signature. Next, spikes were removed by connecting the points immediately preceding and following the spike. Finally, each tidal cycle was integrated with and without FDOM spikes using the "AUC" package in R (Ballings and Van den Poel, 2013). The integration process is represented graphically in Figure S5. FDOM contributions from the three tidal cycles ranged from 5.7 to 14.8% and were estimated to account for up to 24% in other tidal cycles.

Error terms associated with each of the three potential interferences to in situ FDOM measurements are identified in Section Fluorescence Interference and Corrections. The correction factor for temperature was based on linear regression of r > 0.99 (p < 0.0001) with and RMSE <1 QSU. For dissolved materials, attenuation was estimated at <10% based on A<sup>254</sup> values for the site. For suspended solids, attenuation/scattering effects on FDOM were estimated at less than 5% for >98% of the data-set. Assuming attenuation of dissolved and suspended solids is additive, FDOM measurements may be ∼15% lower than actual FDOM values. These conditions represent the upper limits of FDOM attenuation based on high levels of dissolved matter (A254: 0.44, <sup>a</sup>254: 101 m−<sup>1</sup> ) and high turbidity (25.0 FNU). As attenuation from dissolved and suspended solids is not corrected for in FDOM values reported in this study, it is important to remember these potential sources of under-linearity when interpreting QSU values.

#### RESULTS

#### Environmental Conditions

This study covers a period of limited freshwater delivery to the Shark River estuary due to below-average rainfall and consequent reductions in water management inputs across the northern border of ENP into Shark River Slough (**Table 1**). This limited rainfall led to moderate to severe drought for the region stretching from the early dry season through the early wet season (droughtmonitor.unl.edu). The combination of these factors resulted in lower than normal discharge rates from the Shark River, with net negative flows (marine inflows to the estuary) from 12/2014 to 09/2015, the longest period of negative discharge for the 2002–2015 data record (USGS gage #252230081021300, tidally filtered discharge). As such, time-series D1, D2, and W1 (early dry, late dry and early wet seasons, respectively) all received minimal freshwater inputs from water management (although W1 received increased rainfall). In contrast, T1 and T2 timeperiods exhibited high discharge rates for the estuary linked to high inflows from the S12A-D structures (**Figure 1**). Sea level acted as an additional hydrologic control and increased from D2 through T2, with the highest daily mean sea level value recorded at Key West between 1913 and 2013 occurring on 10/01/2015, the start-date for T2 (CO-OPS<sup>1</sup> ). Basic statistics for in situ water quality data collected for this study are displayed in **Table 2** by time-series. Salinity and freshwater hydrology were inversely related, with lowest salinity during periods of high freshwater discharge and low sea-level (T1). Salinity increased through D1 and peaked during D2 due to minimal freshwater inputs. Salinity values decreased throughout W1 due to increased rainfall and during T2 due to increased water management inflows (**Table 1**). Dissolved oxygen (DO) inversely co-varied with

TABLE 2 | Averages and standard deviations for water quality variables by time-series.


<sup>1</sup>Center for Operational Oceanographic Products and Services (CO-OPS) (2016-07-14): NOAA Monthly Mean Sea Level Summary Data for the Key West, Florida, Water Level Station (FCE) (NOAA/NOS Co-OPS ID 8724580) from 01-Jan-1913 to Present. Center for Operational Oceanographic Products and Services (CO-OPS); Long Term Ecological Research Network. doi: 10.6073/pasta/4fca540ab6a8146f26b97c4eb1186a80.

seasonal patterns in temperature and on diel time-scales, with higher values during the day. In general, turbidity increased with channel velocity (both positive and negative), suggesting control by tidal or wind-induced erosion rather than allochthonous delivery to the estuary, matching previously reported decoupling of particulate and dissolved carbon pools for this system (He et al., 2014).

Values for FDOM ranged from 55.2 to 208.6 QSU, matching previously reported QSU values for maximum fluorescence intensity in coastal mangrove rivers of the Everglades (Jaffé et al., 2004). High tidal variability was clearly evident for the high resolution data, with variation of the FDOM signal in excess of 100% between consecutive low and high tides (e.g., 1/26/15 and 3/27/15, **Figure 2**). Highest FDOM values occurred during the beginning of the dry season, similar to previously observed longer-term seasonal trends upstream of SRS-5, which have been attributed to a combination of evaporation-mediated concentration and water management inputs (Chen et al., 2013). In contrast, FDOM values were considerably lower during the late dry season (D2) when DOC concentrations are historically higher in the upstream freshwater marshes (Chen et al., 2013).

#### Relationships between FDOM, Water Quality, and Hydrology

To quantitatively assess the relationships between water quality, hydrology and FDOM, linear regression of FDOM vs. parameters in **Table 2** is presented in **Table 3** as correlations, both for 15 min data and daily averages. Of these parameters, salinity had the strongest relationship to FDOM at both temporal scales. The relationship between FDOM and stage had the second highest r-values, with a higher correlation for daily data. At daily resolution, filtered discharge exhibited the third strongest correlation (though weak: r = 0.28). Temperature, pH, and filtered discharge showed weak correlations to FDOM for the high-resolution data (r = −0.20, −0.24, and 0.24, respectively) as did temperature, discharge and turbidity for daily resolution (r = −0.22, 0.17, and −0.22, respectively). Based on the above (see **Table 3**), salinity, stage, and tidally filtered discharge (at daily resolution) were investigated as potential drivers of highfrequency FDOM dynamics.

#### Salinity and FDOM

Measurements of FDOM at 15-min intervals showed distinct semi-diurnal patterns related to tidal influence and inversely tracked salinity patterns where highest FDOM values corresponded to lowest salinity for each tidal cycle (**Figure 2**), as observed in other mangrove systems (Dittmar and Lara, 2001). Correlations to FDOM were higher for 15-min than daily measurements (**Table 3**), indicating tidal control. The relationship between salinity and FDOM was examined using the ratio of the two variables (FDOM:salinity) shown in **Figure 3A**. A clear divide in the data-set was observed through this relationship, where T1 and T2 had lower FDOM:salinity



15-min data measured at 15 minute increments except filtered discharge (hourly); daily data averages of 15-min data.

FIGURE 2 | FDOM, salinity, stage and tidally filtered discharge for the 5 time-series: T1, D1, D2, W1, and T2 (see Table 1 for dates). Gray dots are 15-min interval measurements for FDOM, salinity and stage. Black lines for FDOM, salinity and stage are smoothing lines to highlight multi-day trends. For filtered discharge, black lines is hourly tidally filtered discharge measured at Gunboat Island. Gray filtered discharge lines represent data from the adjacent Harney River drainage to serve as reference for gaps in the Gunboat discharge record.

ratios compared to the remaining three time-series. This divide matched closely with management-driven inflows as well as tidally filtered discharge (both considerably higher for T1 and T2, see **Table 1**). A similar divide in the DOC:salinity ratio observed by (Cawley et al., 2014) indicated that T1 and T2 most closely resembled wet season characteristics while the remaining time-series matched the dry season. As outlined in Section Environmental Conditions, the study period experienced abnormal hydrology, and the divide in FDOM:salinity confirmed that conventional seasonal wet/dry timing did not fit this data-set. Rather, seasonal transition time-series (T1 and T2) resembled classic wet season hydrology and salinity gradients while the remaining three time-series resembled dry season conditions due to reduced rainfall and consequent drought. Based on this observation, the data-set was divided into two sub-groups: the first for time-series during periods of low freshwater inflow (D1, D2, and W1, **Figure 3B**), the second for time-series during periods of high freshwater inflow (T1 and T2, **Figure 3C**). Dividing the data-set in this manner dramatically improved linear regression fits (r = 0.94 for D&W, r = 0.93 for T) in comparison to the full data-set (r = 0.83). An additional split in **Figure 3C** during T2 was observed, where FDOM:salinity ratios transitioned from high salinity and low freshwater inputs to dramatically increased freshwater inflows and low salinity. To examine this, T2 was sub-divided into two sections: before and during the start of high outflows (positive discharge from the estuary, **Figure 2**) and during and after high outflows. The first sub-group clustered more with D&W time-series due to enhanced saltwater presence, higher stage and low or negative filtered discharge (**Figure 3C**, dark gray points). The second sub-grouping fell in the same FDOM:salinity region as T1: characteristic of high water management inflows in spite of high sea levels.

The data presented in **Figure 3** suggest a non-conservative relationship between FDOM and salinity, where FDOM values were higher at the mid-estuary than would be expected for conservative (linear) mixing behavior of freshwater and marine endmembers. A similar relationship was previously reported for this location by Cawley et al. (2014), who determined that non-conservative inputs were primarily due to DOC contributions from the riparian mangrove forests. Following this hypothesis, FDOM:salinity plots were fitted with quadratic regression lines which improved fits over linear regression (full data-set: r = 0.85, D&W: r = 0.94, T: r = 0.94). Non-conservative inputs were calculated for individual neap and spring tides for each time-series (n = 20) as the difference between integrations of quadratic and linear fits (Cawley et al., 2014). An average nonconservative input percentage of 6 ± 2% was estimated with a maximum of 9% (neap tide, D2) and a minimum of 3% for the consecutive spring tide.

#### FDOM and Tidal Dynamics

Tidal FDOM patterns were inversely related to stage with a temporal lag component, where highest tidal values for FDOM lagged lowest stage (low tide) by 1–2 h. Stage correlations to FDOM (FDOM:stage) were considerably higher for daily data (**Table 3**), likely due to the temporal lag observed at tidal timescales. When tidal FDOM patterns were time-matched to stage (lowest FDOM values time-lagged to coincide with peak stage), a much stronger regression was achieved (r = −0.77), similar to FDOM:salinity. As with salinity, the FDOM:stage data was divided into T and D&W groupings, with a better fit for D&W (r = −0.79) than T (r = −0.72), indicating stronger tidal control of FDOM:stage dynamics during periods of low freshwater inflows. Values of FDOM were consistently higher for neap tides (average 136.4 QSU) compared to spring tides (average 124.6 QSU). Neap tides also exhibited lower salinity, lower stage and higher tidally filtered discharge, indicating reduced tidal influence. The difference in FDOM between neap and spring tides was considerably larger for T time-series (+17.1 QSU, average of spring and neap: 119.8 QSU) compared to D&W time-series (+10.2 QSU, average of spring and neap: 136.5 QSU). In general, decreasing FDOM trends were observed during spring tides, while increasing trends accompanied neap tides. Differences in neap/spring FDOM patterns provided evidence that greater freshwater influence during neap tides led to higher allochthonous DOC inputs to the estuary. In addition, this suggested a temporal decoupling between weekly (spring/neap tides) and monthly (time-series) resolutions for the relationship between freshwater discharge and FDOM where neap tides

increased both filtered discharge rates and FDOM signals while **Figure 3** indicated dilution of FDOM (lower signal) during periods of increased freshwater delivery.

Rapid changes in FDOM during ebb tides were consistently observed near the tidal FDOM minimum which did not fit the semi-diurnal sinusoidal tidal signature. Examples of these FDOM patterns along with additional parameters are shown in **Figure 4**, with the unexpected spikes highlighted in black. These FDOM spikes consistently coincided with a knee observed in ebb tide stage patterns which could not be fully explained by tidal flow asymmetry typically observed in mangrove channels (Mazda et al., 1995). The knee denoted a deceleration of falling water levels attributed to geomorphological control of tidal hydrodynamics by riparian mangrove forests. Specifically, it was hypothesized that these patterns correspond to tidally regulated porewater exchange between mangrove sediments and the river channel. The location where the sonde was deployed is immediately adjacent to mangrove sediments perforated with crab burrows, a feature shown to enhance hydraulic connectivity of mangrove sediments to the water column [up to 20% in some systems, Stieglitz et al. (2013)]. Reduced dissolved oxygen (DO) levels, often associated with porewater hypoxia in mangrove ecosystems (Bouillon et al., 2007), were generally present during FDOM spikes supporting the hypothesis of porewater contributions. However, these patterns were somewhat variable, with examples in **Figure 4** showing consistently depleted DO concurrent with FDOM spikes during T2 but not T1.

#### filtered discharge. Correlation of daily FDOM values to filtered discharge was relatively weak for the whole data set (r = 0.30, n = 145, p = 0.0009) and for D&W (r = 0.29, n = 77, p = 0.0083) but stronger for T (r = 0.60, n = 45, p < 0.0001), likely due to substantial water management inputs during T1 and T2. To quantify the impact of filtered discharge on changes in FDOM, the difference in FDOM between consecutive days was calculated as δ-FDOM = FDOM(x+1)−FDOM(x) where x is the date. Based on δ-FDOM calculations for time-series, daily changes in FDOM of +40.0 QSU to −28.6 QSU indicate equivalent shifts of +377.4 µM (+31%) to −274.7 µM (−22%) for DOC concentrations. The regression of δ-FDOM:filtered discharge (r = 0.65, n = 115, p < 0.0001) showed a much stronger relationship than FDOM:filtered discharge (r = 0.30, n = 120, p = 0.0009). Interestingly, correlations between filtered discharge and δ-FDOM were stronger for D&W (r = 0.74, n = 72, p < 0.0001) than T (r = 0.71, n = 43, p < 0.0001), which was unexpected since T time-series were more strongly linked to filtered discharge patterns. Regression slopes for T and D&W (0.01 and 0.03, respectively) indicated that equivalent changes in filtered discharge yielded a change in the FDOM signal three times greater for dry season time-series compared to wet season. This was attributed to lower stage during D&W time-series (**Table 1**), leading to smaller volume in the Shark River. The corresponding evapo-concentration of the Shark River DOC pool would be more sensitive to dilution from changes in freshwater flow.

#### Daily FDOM Dynamics

Multi-day patterns in FDOM (see smoothing lines in **Figure 2**) corresponded to freshwater input patterns observed as tidally

#### DOC Fluxes

Previously, long-term DOC fluxes were calculated for this site based on monthly DOC concentrations and tidally filtered discharge (Regier et al., 2016), with modeling of monthly flux

Regier and Jaffé Short-Term Dissolved Organic Carbon Dynamics

values predicted by salinity, rainfall, water management inflows, and the Atlantic Multidecadal Oscillation (AMO), a long-term climatic index. Short-term fluxes based on DOC (using FDOM as a proxy) and hourly tidally filtered discharge were calculated for this study to compare and contrast patterns and drivers of DOC fluxes in the Shark River across temporal scales. Both T time-series exhibited net positive flux values (DOC exported from the estuary), averaging 3.4 ∗ 10<sup>9</sup> mg C d−<sup>1</sup> . In contrast, all D&W time-series had negative flux values averaging −4.5 ∗ 10<sup>9</sup> mg C d−<sup>1</sup> , indicating considerably longer DOC residence times between periods of positive filtered discharge (average movement calculated from velocities of 2432 m d−<sup>1</sup> for D&W in comparison to 8843 m d−<sup>1</sup> for T time-series in relation to ∼9500 m from the study site to the river mouth). For hourly flux values, no correlations for water quality or hydrology variables (except discharge) were observed, including salinity and stage, which both correlated to long-term fluxes (Regier et al., 2016). As discharge was higher during neap tides, DOC fluxes were also considerably higher, with averages of 2.8 ∗ 10<sup>9</sup> mg C d−<sup>1</sup> and −5.9 ∗ 10<sup>9</sup> mg C d−<sup>1</sup> for neap and spring, respectively. This pattern held true when fluxes were sub-divided into D&W and T groupings with neap tide fluxes 6.2 ∗ 10<sup>9</sup> mg C d−<sup>1</sup> and 1.2 ∗ 10<sup>10</sup> mg C d−<sup>1</sup> higher than spring tide fluxes for D&W and T subsets, respectively.

To validate the extrapolation of high-frequency FDOM measurements to DOC values and improve the accuracy of DOC flux determinations, flux values were calculated from daily DOC data (based on FDOM as a proxy) and monthly grab samples where the DOC value at the beginning of each time-series was extrapolated to all days in the time-series (Gaiser and Childers, 2016). As the two sets of fluxes were determined for the same days using the same filtered discharge data, the only difference was the application of low-resolution vs. high-resolution DOC concentrations in the estimates. The resulting flux calculations were compared using linear regression (**Figure 6**), with stronger correlations for D&W time-series (r = 0.95, n = 73, p < 0.0001) in comparison to T time-series (r = 0.93, n = 43, p < 0.0001). To interpret if fluxes based on monthly grab samples were over-estimated or under-estimated in comparison to high-frequency flux calculations, a 1:1 line was included. In general, linear regression for D&W fell slightly below the 1:1 line (slope = 0.80), indicating monthly fluxes were underestimated (**Figure 6A**) while T time-series were on or above the line (slope = 1.25), particularly for high fluxes during T2 (**Figure 6B**).

#### DISCUSSION

High-resolution measurements of FDOM showed highly dynamic DOC behavior controlled at tidal and daily time-scales by a mixture of tidal influence, freshwater inputs, and exchange with mangrove forests. Whereas tidal control was consistent for all time-series, freshwater control of FDOM was more sporadic and varied seasonally. The study period was unique as the two major sources of freshwater to the Shark River (rainfall and water management inflows), which are usually coupled (though time-lagged, Saha et al., 2012) were separated, allowing for a comparison of the influence of each freshwater source on high-resolution FDOM dynamics. In addition, previously unobserved spikes during ebb tide were identified as likely originating from porewater exchange with the mangrove forests. While tidal control is evident at hourly resolution and freshwater influence is well-characterized at daily resolution, the abovementioned mangrove-derived spikes would not be observable, even at hourly resolution.

#### Tidal-Scale FDOM Dynamics

The observed semi-diurnal variability in FDOM patterns in contrast to those of salinity and stage suggest strong tidal control of DOC leading to significant inverse relationships for all timeseries with the former and weaker correlations with the latter. These relationships have been well-established in this region (Bergamaschi et al., 2012; Cawley et al., 2014) with salinity identified as a significant predictor of long-term DOC fluxes (Regier et al., 2016). However, untangling the influences of salinity and stage on estuarine DOC dynamics remains difficult (requiring controlled experimental conditions, e.g., Chambers et al., 2014) as these factors tightly co-vary. Data presented here shows that FDOM is temporally coupled to salinity and lags stage, suggesting that high-frequency DOC dynamics are primarily controlled by tidal drivers rather than simply by physical hydrology (stage).

A recent mesocosm study of C response to salinity and inundation in mangrove peat soils from the region found that higher salinity decreased porewater DOC by 2.8 ± 3.3 ppm (23 ± 27% of average DOC for this study), while increased inundation enhanced DOC concentrations by 3.6 ± 4.5 ppm [30 ± 37%; based on a 8 cm increase in stage; (Chambers et al., 2014)]. Findings from the current study of an inverse relationship between DOC and salinity at tidal time-scales match findings from the mesocosm study (Chambers et al., 2014). However, increased salinity conditions equivalent to those reported by Chambers et al. (2014) yielded a much larger decrease in DOC (−7.5 ppm or −62% for D&W time-series in **Figure 3B**, −5.0 ppm or −41% for T time-series in **Figure 3C**).

In contrast, the inverse relationship between tidal stage and DOC (attributed to dilution) does not match with higher DOC in inundated mesocosm experiments (Chambers et al., 2014). In the latter, an equivalent increase in stage of 8 cm yields decreased DOC of considerably smaller magnitudes (−0.7 ppm or −6% for D&W time-series, −0.5 ppm or −4% for T time-series).

Decoupled response of DOC to inundation between mesocosm and in situ measurements may stem from the estimated small contributions of mangrove-derived DOC at this location (∼6% of the DOC pool), whereas the mesocosm results present changes in DOC exclusively from mangrove peats. Based on Chambers et al. (2014), increased inundation is expected to decrease the concentration of the overall DOC pool through dilution while simultaneously enhancing mangrove inputs. As sea level rise drives higher saltwater intrusion and increased inundation in this region, higher C loss from the mangrove forest (See also Section Mangrove Contributions to the DOC Pool) is expected to account for a larger portion of the Shark River DOC pool and increase outwelling of mangrovederived DOC (Dittmar and Lara, 2001). Such processes are expected to increase light attenuation in the water column (Ganju et al., 2014) of adjacent light-sensitive seagrass and coral reef communities (Shank et al., 2010; McPherson et al., 2011) through long-range DOC transport (Maie et al., 2012; Yamashita et al., 2013. However, DOC export from this region in general is predicted to decrease in the future due to shifting hydrology due to climate change (Regier et al., 2016). Thus, the potential balance between enhanced mangrove forest inundation and predictions based on long-term DOC datasets remains to be determined. Better understanding of the potentially complex relationship between inundation and DOC export is therefore crucial for low-relief coastal systems like mangrove forests, where ecosystem sustainability is based on accretion keeping pace with sea-level rise (McKee, 2011). Mangrove accretion is primarily dependent on primary productivity (Cahoon and Lynch, 1997) and correlates to organic C burial rates, which are below global averages for the Shark River (Breithaupt et al., 2014). Thus, potential DOC losses from mangrove peats due to the combined effects of increased salinity and inundation could increase the vulnerability of the coastal mangrove fringe to submergence.

#### Mangrove Contributions to the DOC Pool

The DOC pool in the Shark River is strongly influenced by upper watershed freshwater wetland sources, combined with mangrove inputs (Yamashita et al., 2010; Bergamaschi et al., 2012; Cawley et al., 2014). However, it is not clear from previous studies if this mangrove contribution is related to porewater inputs, litter leachates, or combinations of these. As such, these contributions could respond differently to inundation patterns, tidal action, changes in primary productivity, and associated litter-fall. Bouillon et al. (2007) reported considerable porewater contributions of DOC (estimated at 30% of the total DOC pool) draining a mangrove forest in Tanzania during ebb tide, with porewaters exhibiting considerably higher salinity than the creek water column. In contrast, the notably lower percentage of DOC attributed to mangrove inputs in the Shark River indicates that porewater DOC may not account for a considerable component of the DOC pool. However, this is likely due to the large allochthonous DOC inputs received from upstream freshwater marshes in the Everglades, whereas the Tanzania creek system received minimal freshwater inputs (Bouillon et al., 2007).

However, the abovementioned spikes in FDOM during ebb tide provide new information on the role of mangrove inputs to this system. Although non-conservative mixing calculations indicate that mangrove DOC comprises a relatively minor component of the total DOC pool, FDOM spikes indicate seasonally variable contributions exceeding the 6% mangrove input estimation. For three consecutive tidal cycles, FDOM signals were compared by including or excluding the FDOM spikes (**Figure 5**). Based on this comparison, between 5.7 and 14.8% of the total change in FDOM during these tidal cycles was attributed to mangrove porewater inputs. The percent contribution of FDOM spikes varied with season, with higher and more consistent mangrove DOC contributions observed during the T time-series compared to D&W time-series. This supports previous research reporting higher CDOM production during wet periods compared to dry periods (Shank et al., 2010). However, the magnitude of these FDOM spikes infer that the majority of mangrove DOC export into the river during ebb tide (estimated at 65% for SRS-6, Romigh et al., 2006) occurs at rapid time-scales (1–2 h). Thus, accurately characterizing this highly dynamic exchange process to constrain aquatic C fluxes in mangrove forests requires high-frequency measurements.

Seasonal variation in FDOM spikes also gives insight into the potential role of mangrove porewaters and primary production in DOC exchange between mangrove forests and the river channel. If primary production preferentially dictated mangrove DOC inputs over hydrology, it would be expected that the highest inputs (largest spikes) would coincide with peak primary productivity during the wet-season (Castañeda-Moya et al., 2013). However, spikes are smaller and less consistent during W1 than T1 or T2, suggesting that the extent and duration of tidal

FIGURE 5 | Quantification of rapid spikes in FDOM during ebb tide attributed to mangrove porewater inputs were calculated for three consecutive tidal cycles during T2. Dark gray areas outlined in black represent FDOM spikes associated with ebb tide mangrove porewater inputs. Light gray areas outlined in gray represent the rest of the change in FDOM during each tidal cycle, where change is defined as FDOM above a baseline running from the lowest FDOM value at the start of the tidal cycle to the lowest FDOM value at the end of the tidal cycle. Percentages indicate the percent of FDOM change attributed to ebb tide mangrove porewater inputs.

inundation rather than patterns in primary productivity controls the magnitude of DOC porewater exchange. Furthermore, this may indicate that DOC spikes are more strongly associated with leaching from mangrove sediments and porewater exchange, rather than direct leaching from mangrove leaves and stems (Maie et al., 2006b; Romigh et al., 2006).

### Water Management and Rainfall Control Daily FDOM Patterns

Daily and multi-day patterns in FDOM showed clear links to freshwater inputs (**Figure 2**, Section Daily FDOM Dynamics), indicating the essential role of upstream hydrology in regulating DOC concentrations in the Shark River. In particular, inflows from water management structures drive a division in FDOM:salinity ratios between time-series (**Figure 3**). Of the hydrologic parameters influencing the estuarine salinity gradient, water management structures are the only means of directly regulating freshwater inflows to this system, with natural delivery of freshwater from rainfall and potential groundwater inputs and seepage through barriers to flow (Saha et al., 2012). A comparison of T1 and T2 time-series demonstrates the control of FDOM temporal dynamics exerted by managed freshwater inputs. Both T1 and T2 exhibit similar FDOM:salinity ratios (**Figure 3**), suggesting similar DOC source and mixing dynamics. However, neither the salinity gradient (dramatically different sea levels, **Table 1**, and salinity, **Table 2**, between the two time-series) nor rainfall (similar for all time-series except W1) can explain the similar FDOM:salinity distributions for T1 and T2 in contrast to the other three time-series. Instead, freshwater control by water management inflows appears to link these time-series (high for T1 and T2, low or absent for the other time-series). This is evident in the shift of FDOM:salinity observed for T2 (**Figure 3C** and Section Salinity and FDOM) from ratios characteristic of the dry-season to wet-season which occurred in less than a day. Based on this rapid change along with the divide in **Figure 2**, it is clear that water management inflows (either directly or indirectly) control Shark River hydrology and DOC dynamics and export at short time-scales.

Both rainfall and water management inputs were identified as significant freshwater drivers of long-term DOC fluxes in this system, with rainfall contributing almost double the modeling power as managed inflows (Regier et al., 2016). For the highfrequency data collected during this study, the influence of rainfall is less obvious than water management, likely due to unsusually low precipitation rates for all time-series except for W1 (**Table 1**). The clearest evidence of rainfall driving changes in DOC occurred with onset of the rainy season (between D2 and W1). Low FDOM values for D2 (relative to D1) were attributed to enhanced saltwater intrusion into upstream freshwater marshes observed through semi-diurnal tidal signatures in stage data at the Rookery Branch gage 20 km upstream of SRS-5 (#022908295, http://waterdata.usgs.gov/). Subsequently increased FDOM during W1 was attributed to the enhanced hydrologic connectivity of the estuary to upstream marsh regions as a result of increased rainfall. Decreased DOC concentration and exports due to drought conditions (e.g., D2) have been reported for other coastal wetland systems (e.g., Ardón et al., 2016). Likewise, rainfall events have been linked to increased DOC concentrations in both marsh and mangrove systems (Bergamaschi et al., 2012; Ryder et al., 2014), although shifts in other drivers between D2 and W1 like soil temperature may be partially responsible for the changes in DOC concentration (Davidson and Janssens, 2006; Ryder et al., 2014).

During the study period, the influences of water management and rainfall on daily FDOM dynamics were evident. However, whereas water management inflows dramatically shifted estuarine hydrology from negative to positive discharge, rainfall did not. The importance of water management to maintain seasonal freshwater delivery has been well-established for this region (Sklar et al., 2005; Obeysekera et al., 2014). High-frequency measurements presented here indicate that, particularly during periods of below-average freshwater flow (e.g., drought), water management inputs are capable of rapidly altering salinity and associated water quality gradients in the coastal Everglades. Based on this, corrective actions by water management (e.g., emergency water releases to slow saltwater intrusion during periods of high drought) guided by high-frequency monitoring provides managers with new tools to rapidly respond to observed or predicted environmental problems. Likewise, as this region is clearly sensitive to managed inflows, high-frequency measurements are increasingly critical in the monitoring of estuarine response to changes in freshwater inputs.

#### Linking Short-Term and Long-Term Fluxes

Hydrology is the primary driver of DOC fluxes, and the strong linear fits in **Figure 6** confirm this. However, differences in fluxes based on monitoring frequency of DOC concentrations are evident through the deviations of fluxes from the 1:1 line. In general, D&W time-series cluster together close to 1:1 (**Figure 6A**), suggesting that during periods of lower freshwater inflows and therefore less variable filtered discharge (**Figure 2**), interpolating daily DOC fluxes from monthly DOC values is relatively accurate. In contrast, T time-series show more variability (**Figure 6B**), where T1 flux calculations closely follow the 1:1 line but T2 does not. In particular, high flux values during T2 are dramatically over-estimated (up to 180%) by interpolation of monthly DOC to daily resolution. This is likely due to large and rapid fluctuations in discharge (**Figure 2**) driving rapid changes in DOC concentrations. As such, interpolation of daily fluxes from monthly DOC sampling frequencies may be applicable during periods of relatively static discharge, but is not sufficiently accurate to constrain DOC export rates during periods of variable freshwater flows. This supports previous findings that weekly grab sampling was temporally inadequate to characterize DOC fluxes during rapid shifts in hydrology (Jollymore et al., 2012; Pellerin et al., 2012). Thus, current long-term monitoring projects seeking to better constrain C budgets in dynamic aquatic systems like estuaries could greatly benefit from the simultaneous application of high-frequency measurements.

## CONCLUSIONS

A suite of environmental drivers, including salinity, stage and tidally filtered discharge, were identified as regulators of high-frequency FDOM patterns in the Shark River estuary. Values for FDOM were dynamic at tidal time-scales, changing in excess of 100% between consecutive low and high tides during periods of high freshwater discharge. The drivers of FDOM manifested in multiple distinct patterns, with salinity and stage varying semi-diurnally with tides, and knees in ebb tide stage corresponding to spikes in FDOM potentially attributed to mangrove porewater inputs accounting for up to 24% the tidal FDOM signal, particularly during periods of higher inundation. Freshwater discharge also controlled daily and multi-day changes in FDOM (up to ±25%), indicating hydrologically driven shifts in DOC concentration, and potentially source. The influence of water management on DOC dynamics in the mangrove estuary demonstrated the capability of managed inflows to restore the salinity gradient and accompanying estuarine C dynamics following pronounced drought and elevated sea levels. Rainfall was observed to alter DOC concentrations but had less influence over salinity and FDOM than water management (potentially related to drought conditions during much of this study). In general, managed inflows most strongly affected FDOM dynamics during seasonal transitions while precipitation preferentially and expectedly, controlled FDOM during the onset of the wet season. While management strategies currently focus on longer term hydrologic response (e.g., seasonal to inter-annual) in the context of Everglades restoration, rapid response of the coastal zone to freshwater inflows suggests the ability of water management to control hydrology in the mangrove ecotone at much shorter time-scales. Since response of DOC mixing and source dynamics to environmental drivers changes across temporal scales, integration of high-frequency data into current long-term monitoring programs is essential to understand temporally complex relationships between water quality and hydrology.

Hydrology, climate and management have all been identified as drivers of long-term DOC export in the Shark River (Regier et al., 2016), and findings here generally support that these drivers apply to short-term (hours to days) variations in DOC (based on FDOM as a proxy). Adaptive restoration management relies on constantly improving our understanding of ecosystem function and response to climatic and management factors. Further investigating key ecological indicators like biogeochemistry and nutrient transport spanning spatial and temporal scales will advance modeling capabilities and inform management decision-making. High-resolution measurements of FDOM provide new insights into organic matter cycling in coastal systems and indicate the potential value of high-frequency data to complement and expand existing long-term monitoring programs through a temporally integrated understanding of biogeochemical processes and nutrient budgets in dynamic coastal ecosystems.

#### AUTHOR CONTRIBUTIONS

PR performed field work, data analysis, interpretations, and contributed to writing the manuscript. RJ obtained the funding to support this study, designed the study, assisted in data analyses and interpretation and writing the manuscript.

#### ACKNOWLEDGMENTS

This work was primarily supported through the George Barley Endowment and is associated with the NSF-funded Florida Coastal Everglades LTER (DBI-0620409) and the SFWSC programs. The authors thank John Kominoski for helpful comments on the original manuscript, Mike Rugge for assistance with **Figure 1** and Rafael Travieso for help in the field. The authors also thank Piotr Kowalczuk and Heather Erin Reader for comments that greatly improved this manuscript. This is contribution number 816 from the Southeast Environmental Research Center.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2016.00250/full#supplementary-material


peat catchment. Sci. Total Environ. 490, 28–36. doi: 10.1016/j.scitotenv.2014. 04.091


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Regier and Jaffé. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Surface Gradients in Dissolved Organic Matter Absorption and Fluorescence Properties along the New Zealand Sector of the Southern Ocean

#### Eurico J. D'Sa<sup>1</sup> \* and Hyun-cheol Kim<sup>2</sup>

<sup>1</sup> Department of Oceanography and Coastal Sciences, Louisiana State University, Baton Rouge, LA, USA, <sup>2</sup> Korea Polar Research Institute, Incheon, South Korea

#### Edited by:

Michael Seidel, University of Oldenburg, Germany

#### Reviewed by:

X. Antón Álvarez-Salgado, Spanish National Research Council, Spain Isabel Reche, University of Granada, Spain Leanne C. Powers, University of Maryland Center for Environmental Science, USA

> \*Correspondence: Eurico J. D'Sa ejdsa@lsu.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 07 November 2016 Accepted: 17 January 2017 Published: 28 February 2017

#### Citation:

D'Sa EJ and Kim H-c (2017) Surface Gradients in Dissolved Organic Matter Absorption and Fluorescence Properties along the New Zealand Sector of the Southern Ocean. Front. Mar. Sci. 4:21. doi: 10.3389/fmars.2017.00021 The Southern Ocean plays a critical role in the global carbon cycle; dissolved organic matter (DOM), a component in the carbon cycling, can be characterized optically. Sea surface chromophoric dissolved organic matter (CDOM) absorption and fluorescence properties were examined in the New Zealand sector of the Southern Ocean (NZSSO) along a transect encompassing various hydrographic fronts associated with the Antarctic Circumpolar Current (ACC) during summer. Phytoplankton chlorophyll, dissolved organic carbon (DOC) and CDOM absorption were observed to be most elevated off the New Zealand shore and then decreased to low values [chlorophyll: 0.21 ± 0.06 mg m−<sup>3</sup> ; DOC: 54.19 ± 4.02 µM, and CDOM absorption coefficient at 325 nm (ag325): 0.097 ± 0.061 m−<sup>1</sup> ] between the Subtropical (STF) and Antarctic Polar (APF) Fronts. Increases in phytoplankton biomass and DOC concentrations between the fronts were associated with meanders or eddies observed in satellite sea surface salinity and chlorophyll imagery. Overall, CDOM absorption was the dominant contributor to total absorption at 443 nm with implications for ocean color. Beyond the APF in the Antarctic Zone, an elevated chlorophyll band likely associated with upwelled waters transitioned to low chlorophyll in the summer ice edge zone that influenced DOM optical properties. A latitudinal increase in ag325 and corresponding decrease in spectral slope S (µm−<sup>1</sup> ) poleward from the STF could be due to a combination of factors including, decreasing CDOM photooxidation, upwelling of high-CDOM waters or bacterial CDOM production in the Antarctic Zone. Parallel factor analysis (PARAFAC) of fluorescence spectra identified two protein-like (C1 and C2) and two humic-like (C3 and C4) components common in the global ocean. ag325 and the humic-like C4 fluorescent component were positively correlated to chlorophyll, indicating biological control. Surface distribution of the protein-like C1 and C2 and the marine humic-like C3 components showed patterns that appeared to be influenced by both physical and biological processes. This study provides insights into surface CDOM optical properties and its transformation along a complex topographically influenced sector of the Southern Ocean that could be used to trace changes linked to the meridional overturning circulation.

Keywords: CDOM, DOC, EEMs, PARAFAC, chlorophyll, absorption, Southern Ocean

## INTRODUCTION

The Southern Ocean, comprising a region from the Subtropical Front (STF) to the Antarctic continent connects the three main oceanic basins through the easterly flowing Antarctic Circumpolar Current (ACC) that plays a significant role in the global biogeochemical processes and carbon cycling (Sarmiento and LeQuere, 1996). Driven by the strong westerly winds, the ACC consists of several narrow and persistent fronts that contribute to the ocean overturning circulation and the upwelling of carbon and nutrients to the surface ocean that supports much of the ocean's productivity (Orsi et al., 1995; Belkin and Gordon, 1996; Daly et al., 2001; Budillon and Rintoul, 2003; Marshall and Speer, 2012). While the Southern Ocean is considered a high-nutrient low-chlorophyll (HNLC) region due mainly to iron limitation (Boyd et al., 1999), many productive regions especially during the spring-summer months contribute significantly to the global primary production, carbon cycling, and biogeochemical processes (Arrigo et al., 1998; Daly et al., 2001; Hiscock et al., 2003; Honjo, 2004; McNeil and Tilbrook, 2009). The chromophoric dissolved organic matter (CDOM) is the fraction of the dissolved organic matter (DOM) pool that absorbs light in the UV and visible region of the spectrum. It influences ocean color and light propagation in oceanic waters, and provides useful information on DOM sources and sinks (Blough and Del Vecchio, 2002; Nelson and Siegel, 2002, 2013; Swan et al., 2009).

Photosynthetic plankton in the surface ocean is the main source of oceanic DOM and is principally exported to deeper waters through mixing and downwelling of water parcels (Hansell and Carlson, 2001; Jiao et al., 2010; Hansell, 2013). Components of this DOM pool in the form of humic, fulvic, and amino acids impart characteristic absorption and fluorescence properties that can be used to characterize CDOM composition and the diagenetic state (Mopper and Schultz, 1993; Coble, 1996; Stedmon and Markager, 2005). CDOM can be produced in situ by biological production (authochthonous), primarily through microbial remineralization of organic matter or transported from terrestrial sources (Blough and Del Vecchio, 2002; D'Sa, 2008; D'Sa and DiMarco, 2009; Romera-Castillo et al., 2010). It is removed by photochemical degradation and microbial consumption and is influenced by physical processes such as circulation, upwelling, or mixing (Blough and Del Vecchio, 2002; Nelson et al., 2010). The spectral absorption property of CDOM especially over the narrow 275–295 nm wavelength interval (S or S275−295) has been used to gain insights into source, molecular size and reactivity with elevated slope values linked to photo-oxidative degradation (Helms et al., 2008; D'Sa et al., 2014). While CDOM optical properties have been studied in various oceanic regions (Nelson and Siegel, 2002, 2013; Kitidis et al., 2006; Swan et al., 2009), only limited studies have been reported for the Southern Ocean (Kieber et al., 2009; Ortega-Retuerta et al., 2009; Del Castillo and Miller, 2011; Nelson et al., 2010; Nelson and Gauglitz, 2016).

A subfraction of CDOM that fluoresces when excited with UV light (fluorescent DOM or FDOM) is used in the characterization of CDOM using excitation-emission matrix spectroscopy (EEMs) wherein a three-dimensional fluorescence intensity landscape is obtained across a range of excitation (e.g., 250–450 nm) and emission (e.g., 290–550 nm) wavelengths (Coble, 1996). EEM spectra of different fractions of FDOM such as humic-like and protein-like fluorophores display characteristic signatures in the EEM spectra. For example, humic-like material with greater aromaticity ("A" and "C" type compounds typically of terrestrial origin) generally display broad fluorescence peaks with emission typically at longer wavelengths in comparison to marine humics or "M" type compounds attributed to biological activity. In contrast, protein-like fluorescent material display narrower peaks typically in the UV spectral region with characteristics similar to tryptophan and tyrosine-like amino acids likely derived from marine planktonic organisms or bacterial activity (Coble, 1996; Yamashita and Tanoue, 2003; Stedmon and Markager, 2005). Due to the overlapping contribution of the various fluorophores in the EEMs, statistical approaches such as parallel factor analysis (PARAFAC) are commonly used to decompose CDOM EEMs data into their individual fluorescent components that have provided a more quantitative and qualitative assessment of its composition (Stedmon et al., 2003; Stedmon and Bro, 2008) in various oceanic regions (Murphy et al., 2008; Jorgensen et al., 2011; Catalá et al., 2015, 2016; D'Sa et al., 2016), including the Southern Ocean (Wedborg et al., 2007; Nelson and Gauglitz, 2016).

A characteristic feature of the Southern Ocean is the presence of fronts (STF, SAF, APF, and SACCF) that transition the Southern Ocean from the warm subtropical waters into cold polar waters thus separating zones of uniform water properties in a series of step-like changes in temperature, salinity, and biogeochemical processes (Belkin and Gordon, 1996; Budillon and Rintoul, 2003; Honjo, 2004). In the New Zealand sector of the Southern Ocean (NZSSO), however, the location of these fronts are strongly influenced by prominent topographic features that include the Campbell Plateau, the Pacific Australian Ridge (PAR) and the highly productive Ross Sea to the south (Gordon, 1975). Surface CDOM absorption properties have previously been reported for the NZSSO during early spring (Kieber et al., 2009) while DOM fluorescence at a single wavelength has been reported for a transect on the western Pacific Ocean east of New Zealand (Yamashita et al., 2007). Knowledge on CDOM absorption contribution to the total absorption has been useful in improving satellite estimates of chlorophyll and primary production in the Southern Ocean (Reynolds et al., 2001; Siegel et al., 2002; Lee et al., 2011).

In this study, CDOM absorption and fluorescence properties were examined from surface measurements conducted during a transect of the Korean Research vessel Araon along the ∼172◦E meridian between New Zealand and Terra Nova Bay (TNB) in the Ross Sea. Additionally, surface absorption budgets were obtained to assess CDOM contribution to the total absorption field while satellite-derived salinity and chlorophyll estimates were used to assess synoptic large-scale features (e.g., meanders and eddies) to aid the interpretation of CDOM optical measurements in the NZSSO.

## METHODS

#### Sampling and Study Area

Seawater samples (27) were collected from a flow-through system obtained from a depth of ∼2 m as the Korean ice breaker and research vessel Araon transected from Christchurch, NZ to TNB in the western Ross Sea from 30 January to 6 February 2014 (**Figure 1**). Sub-samples were then filtered through GF/F filters and the filtrate stored in acid cleaned pre-combusted amber bottles with teflon-lined caps at 4◦C in a refrigerator for later laboratory processing for CDOM absorption and fluorescence. Filtered GF/F samples were also preserved in a freezer for measurements of chlorophyll concentrations and particulate absorption. Sampling from the flow-through system was however discontinued between 68 and 74◦ S due to the presence of summer sea-ice. Seawater temperature was measured at the outflow of the flow-through system; however, differences in temperature between the inlet and outlet of the flow-through likely existed

FIGURE 1 | Study site. Sea surface temperature (SST) shown at sampling stations obtained between 30 Jan and 6 Feb, 2014 along a transect between Christchurch, New Zealand (NZ) and Terra Nova Bay (TNB) in the western Ross Sea. Thick arrow indicates the direction of the Antarctic Circumpolar Current (ACC) and the thinner arrows indicate the approximate locations of the Subtropical Front (STF), the Subantarctic Front (SAF) and its northern (NSAF) and southern (SSAF) branches, the Antarctic Polar Front (APF), and the Southern Antarctic Circumpolar Current Front (SACCF). CR and PAR indicate the locations of the Chatham Rise and the Pacific Australian Ridge, respectively. Also shown are the locations of three sampling stations (S3, S26, S31) discussed in the text.

and the data has been mainly used to examine broad temperature trends along the transect. The main topographic features such as the Chatham Rise (CR), Campbell Plateau (CP), and the PAR system are shown (**Figure 1**), along with the locations of fronts such as the Subtropical (STF), Subantarctic (SAF) and its northern (NSAF) and southern (SSAF) branches, Antarctic Polar (APF), and the Southern Antarctic Circumpolar Current (SACCF) Fronts in the NZSSO (**Figure 1**).

## Chlorophyll Concentrations

Seawater samples (500 ml) were filtered through Whatman GF/F filters (25 mm diameter) for the fluorometric determination of chlorophyll (Holm-Hansen et al., 1965) using a Turner Designs Model 10AU fluorometer calibrated with pure chlorophyll a (Sigma Chemical Company).

#### DOC Measurements

Samples for dissolved organic carbon (DOC) were filtered through pre-rinsed Whatman GF/F filters and stored in acid cleaned, pre-combusted amber bottles with Teflon-lined caps in a freezer. Laboratory measurements of DOC were made on a Shimadzu TOC 5000A (with ASI-5000A autosampler) using a high temperature combustion method to convert carbon compounds to carbon dioxide, including using the Consensus Reference Material (CRM) for QA/QC (Benner and Strom, 1993; Hansell and Carlson, 2001).

#### CDOM and Particulate Absorption Measurements

Water samples collected from the flow-through system were filtered immediately through pre-rinsed Whatman GF/F filters under low vacuum and stored in acid cleaned, pre-combusted amber bottles at 4◦C in the dark before laboratory analysis. After the filtered samples were allowed to reach ambient room temperature, absorbance measurements of CDOM (A(λ)) were obtained on a WPI UltrapathTM system from 190 to 722 nm on either 0.5 m or 2.0 waveguide cells. To minimize differences in refractive index between sample and reference which cause offsets in absorbance measurements (D'Sa et al., 1999; D'Sa and DiMarco, 2009), a reference salt solution was prepared using granular NaCl (Mallinckrodt) and Milli-Q water to closely match the seawater samples. Absorbance data were corrected by subtracting the mean value over a 10 nm interval (695–705 nm) of the measured absorbance at 700 nm from each wavelength and the ag(λ) (m−<sup>1</sup> ) was calculated using the equation:

$$\log(\lambda) = 2.303 \times \frac{A(\lambda)}{l} \tag{1}$$

Absorption coefficient at 325 nm (ag325) was used as a quantitative parameter of CDOM. The spectral slopes for the intervals of 275–295 nm (S or S275−295, <sup>µ</sup>m−<sup>1</sup> ) were calculated according to Helms et al. (2008). Due to very low CDOM absorption in the visible, absorption especially at 443 nm was below the detection limit of the instrument for some samples.

Particulate absorption measurements were obtained using the quantitative filter technique (QFT) procedure (Mitchell et al., 2003). To measure the particulate absorption, 500 ml of surface seawater samples were filtered onto 25 mm GF/F filters under low vacuum and preserved at −80◦F before laboratory analysis. Optical density of particles on the filter paper were measured in the transmission mode on a Perkin-Elmer Lambda 850 with an integrating sphere and also on a fiber-optic based spectrophotometer (Naik and D'Sa, 2012) as it provided better sensitivity in comparison to the Lambda 850. Details on the laboratory analysis of particulate absorption on the two systems have been previously described (Naik and D'Sa, 2012; Naik et al., 2013). As both CDOM and particulate absorption were very low, measurements on both the spectrophotometer and waveguide were sensitive to instrument noise and baseline variability resulting in uncertainties for some samples in the visible band of the spectrum. Thus, due to the very low absorption values, the relative absorption budget is reported only at 443 nm, a commonly used wavelength.

#### EEMs Measurements and PARAFAC Analysis

Filtered seawater samples used for CDOM absorption measurements were also used to record EEMs using a FluoroMax-4 (Jobin Yvon Horiba) fluorometer by scanning the emission spectra from 290 to 550 nm at 5 nm intervals over excitation wavelengths between 250 and 450 nm at 5 nm increments. The EEMs spectra were obtained after correction of the fluorescence spectra for instrument bias, and the water Raman normalization of the fluorescence intensity (Singh et al., 2010; D'Sa et al., 2014). Due to low absorbance of the samples, there was no requirement for inner filter correction of the fluorescence data. The resulting EEM fluorescence observations were evaluated by PARAFAC analysis using the DOM-Fluor toolbox (Stedmon and Bro, 2008), with the model constrained by non-negativity, and run with three to seven components. Model validation was carried out using split-half analysis, residual analysis and random starts (Stedmon and Bro, 2008; D'Sa et al., 2016).

#### Satellite Data

MODIS Aqua chlorophyll imagery at 4 km were obtained from NASA Ocean Color Giovanni website (giovanni.gsfc.nasa.gov). Due to extensive cloud cover over the study area, imagery averaged between 26 Jan and 10 Feb, 2014 were used that included the sampling period (30 Jan–6 Feb, 2014). Sea surface salinity data (7-day average) from the Aquarius microwave instrument onboard the Aquarius/SAC-D satellite was obtained from NASA Jet Propulsion Laboratory corresponding to the sampling period. A comparison study of the 1◦ gridded Aquarius sea surface salinity with Argo array salinity product showed standard deviation in salinity values ranging between 0.15 and 0.6 for high-latitude oceans (Lee, 2016).

## RESULTS

## Surface Physical Properties

Sea surface temperature (∼2 m depth) measured on board the Araon during its 8-day transect from New Zealand to Terra Nova Bay (TNB) show gradients and a decreasing meridional trend in temperature (**Figure 1**). Poleward decrease and gradients in sea surface salinity are also observed in the Aquarius satellite salinity imagery (**Figure 2A**) and at matching station locations extracted from the satellite data (**Figure 3**). While frontal positions are best identified from hydrographic data across the ACC, in this study we use underway data and satellite imagery combined with historical observations (Budillon and Rintoul, 2003) to infer latitudinal locations of the fronts during the study period (**Figure 3**).

The STF, which separates the warmer and saltier subtropical surface water from the cooler and fresher subantarctic surface water, marks the northernmost extent of the ACC and remains close to the coast south and east of NZ and then moves offshore. Large gradients in salinity and temperature observed in the convergence zone (**Figures 1**, **2**) have been associated with considerable eddy activity southeast of NZ (Williams, 2004; Fernandez et al., 2014). South of NZ, the Sub Antarctic Front (SAF), a major front of the ACC system bifurcates into northern (NSAF) and southern (SSAF) branches with the NSAF (∼53◦ S) closely following the steep southeastern flank of the Campbell Plateau (CP) and the SSAF (∼58◦ S) found over the abyssal plain of the south-west Pacific Basin (Budillon and Rintoul, 2003; Fernandez et al., 2014). Large variability in sea surface salinity south of NZ (**Figure 2A**) indicates eddies and meanders associated with these fronts in the subantarctic zone lying between the STF and SAF (Budillon and Rintoul, 2003). The APF located at ∼62◦ S and generally found between 60.2 and 62.8◦ S, is characterized by large gradients in temperature and salinity (**Figure 3**) and delineates the cold Antarctic Surface Water (ASW) to the south (Antarctic Zone) from the warmer subantarctic surface waters to the north. The Antarctic Zone contains the Antarctic divergence (Daly et al., 2001) and the southern boundary of the ACC (SACCF) which in this region follows closely the Pacific-Antarctic Ridge and is located at ∼63.8◦ S (Budillon and Rintoul, 2003). The seasonal ice zone, generally located south of the SACCF, was ice-covered south of 68◦ S during the transect. In the western Ross Sea, the Terra Nova Bay polynia is an ice-free coastal region with cold ASW (**Figures 1**, **3**; temperature up to +2 ◦C, S < 34.5; Rivaro et al., 2011).

## Surface Chlorophyll Distribution

Surface chlorophyll estimates from MODIS reveal elevated concentrations southeast of NZ and Chatham Rise and the STF region (**Figure 2B**). Along the rest of the transect from STF to APF, chlorophyll values were low with few elevated bands that appear related to eddies and meanders associated with the interaction of the STF and the SAF (**Figures 2A,B**). An elevated chlorophyll band south of the APF is observed in the Antarctic Zone while high chlorophyll values were observed further south in the TNB region of the western Ross Sea. A similar pattern was observed in surface chlorophyll from field observations along the transect (**Figure 3**) with low chlorophyll values in the core of the ACC between STF and APF (0.21 ± 0.06 mg m−<sup>3</sup> ) that varied in the range 0.12–0.38 mg

FIGURE 2 | Satellite imagery. (A) Aquarius satellite-derived sea surface salinity for the period 29 Jan–6 Feb, 2014 along the New Zealand sector of the Southern Ocean, and (B) MODIS-derived chlorophyll (25 Jan–10 Feb, 2014) with approximate locations of the Southern Ocean fronts (red squares).

FIGURE 3 | Surface biophysical properties. Sea surface temperature, satellite-derived salinity, chlorophyll, and DOC concentrations along the transect. Dashed vertical lines represent the locations of the Southern Ocean fronts.

m−<sup>3</sup> , and the more elevated values associated with peaks observed due to meanders and eddies (**Figure 2A**). Higher chlorophyll values were however observed off NZ shore and STF region (0.66 ± 0.43 mg m−<sup>3</sup> ), the elevated band south of the APF (0.64 ± 0.15 mg m−<sup>3</sup> ), with highest values in TNB (2.42 mg m−<sup>3</sup> ).

## Surface Dissolved and Particulate Matter Absorption along the Transect

Surface DOC concentrations were high off the NZ shore (66.71±21.88 µM) and TNB (62.11 µM), with slightly elevated values observed coincident with chlorophyll peaks at 50◦ S, 55◦ S, and around the APF. In the ACC core (STF to APF) values ranged between 48.07 and 60.01 µM with a mean of 54.19 ± 4.02 µM. ag325 between the STF and SACCF ranged between 0.024–0.145 m−<sup>1</sup> (mean of 0.097 ± 0.061 m−<sup>1</sup> ) and showed an increasing trend poleward (**Figure 4**). ag325 was however elevated in the NZ shelf waters (0.179 ± 0.110 m−<sup>1</sup> ) and in TNB (0.188 m−<sup>1</sup> ). The spectral slope S values (29.45 ± 3.71 µm−<sup>1</sup> ) generally showed an inverse relationship to ag325 along the transect (**Figure 4**). S was low closest to the NZ shore (station 1) but increased toward STF and generally decreased poleward into the Antarctic Zone (**Figure 4**). ag325 was positively correlated with chlorophyll (**Figure 5B**; r <sup>2</sup> = 0.48) but was uncorrelated to DOC concentrations (**Figure 5A**).

The relative absorption coefficients in the visible (443 nm) of CDOM, particulate matter (phytoplankton aphy and nonalgal anap) varied across the transect (**Figure 6**) ranging for ag443 from 0.0043 to 0.14 m−<sup>1</sup> , aphy(443) from 0.0043 to 0.14 m−<sup>1</sup> , and anap(443) from detection limit to 0.0033 m−<sup>1</sup> , respectively. Although anap(443) contributed the least to the total absorption (<10%), its relative contribution increased closer to the NZ coast and south of the APF, coinciding with the band of elevated chlorophyll (**Figure 6A**). Variability in phytoplankton absorption aphy(443) was similar to chlorophyll along the transect with elevated values off the NZ shore and south of APF (**Figures 3**, **6A**). However, its contribution relative to CDOM at 443 nm was variable across the transect being generally greater in regions of high chlorophyll while CDOM dominated at other locations as also indicated by the ternary plot (**Figure 6B**).

#### Surface FDOM Properties

Typical EEMs spectra obtained from surface samples in the STF, APF, and TNB reveal the presence of prominent peaks in the spectra (**Figure 7A**). These peaks have traditionally been used to characterize CDOM composition using terminology such as A, C, M, B, and T peaks (Coble, 1996). The presence of the marine humic-like M peak (ex/em, ∼300/380–420 nm) is clearly observed, while the fluorescence peaks associated with the protein-like material, namely, the tyrosine-like and tryptophan-like aromatic amino acids (in the UV spectral range) appear combined visually. In contrast, the typical humic-like A and C peaks though not visually observable, were present at low background levels. PARAFAC analysis identified four major fluorescent components in the surface waters between NZ and TNB (**Figure 7B**; **Table 1**). The excitation and emission loadings of each of the four components (**Figure 7B**, bottom right; **Table 1**) identified components 1 (C1) and C2 with excitation/emission (ex/em) wavelengths of 260/300 nm and 270/330 nm associated with tyrosine- and tryptophan-like material that is microbially produced (Yamashita and Tanoue, 2003; Murphy et al., 2008). Component 3 with ex/em wavelength of 295/400 nm has been previously identified as the marine or "M" humic-like component of microbial origin (Coble, 1996; Yamashita and Tanoue, 2003; Stedmon and Markager, 2005). The broad fluorescence peak of component 4 (C4) with primary (and secondary) ex/em wavelength peaks at <260(360)/460 nm is similar to previously identified terrestrial humic-like A and C like material and generally ubiquitous in the marine environment (Coble, 1996; Murphy et al., 2008, **Table 1**).

The distribution pattern of the four FDOM components differed along the transect (**Figure 8**). The humic-like C4 component was elevated off NZ shore and TNB and uniformly low between the STF and APF (**Figure 8A**) and was positively correlated to chlorophyll (**Figure 9A**). The C3 marine M-like humic component was found to be elevated off NZ and in regions

with elevated phytoplankton chlorophyll (**Figure 8A**) but showed no clear trends with chlorophyll (not shown). The two proteinlike components C1 and C2 were elevated off NZ and TNB, as well as at/near the fronts (**Figure 8B**), with C1 showing a weak negative correlation to salinity (**Figure 9B**).

### DISCUSSION

#### Biophysical Environment

The STF located south and east off NZ forms the boundary between the warmer/salty, nutrient-poor subtropical waters and the colder/fresher, nutrient rich subantarctic waters. Off the NZ shore, the subtropical surface water, generally located 15–40 km offshore, is modified by mixing with freshwater runoff from land and with the subantarctic water further offshore (Sander et al., 2015; **Figure 2A**). The MODIS imagery (**Figure 2B**) shows a broad band of elevated chlorophyll in the subtropical frontal region around NZ that are thought to be related to the mixing of warm, macronutrient-poor, relatively micronutrientrich (especially iron) subtropical water with cold, macronutrientrich and micronutrient-poor subantarctic water (Orsi et al., 1995; Belkin and Gordon, 1996; Boyd et al., 1999). Similar patterns of chlorophyll distribution were previously described with elevated chlorophyll occurring in the STF to the west and east of NZ throughout the year (Murphy et al., 2001) and along a narrow band c. 43◦ S on the south side of the Chatham Rise and extending eastwards to the Chatham Islands (**Figure 2B**). South of New Zealand, the complex marine topography associated with the Macquarie Ridge and the Campbell Plateau strongly influences the circulation pattern with the ACC undergoing intense wavelike migrations and energetic eddies (Gordon, 1975; Bryden and Heath, 1985). Surface expression of such circulation patterns and associated meanders and eddies observed in the satellite sea surface salinity imagery south of New Zealand (**Figure 2A**) also appear associated with more elevated chlorophyll concentrations observed in the MODIS imagery and field measurements along the transect (∼50 and 55◦ S; **Figures 2B**, **3**). Similar trends were previously observed linking meanders and eddies to the interaction of the two SAF branches and the complex topography (Campanelli et al., 2011). Nonetheless, chlorophyll concentrations were low in the subantarctic waters (0.2–0.3 mg m−<sup>3</sup> ) and comparable to levels reported in other subantarctic regions (Boyd et al., 1999; Han and Takahashi, 2001; Murphy et al., 2001; Hunt and Hosie, 2006). South of the APF, an elevated chlorophyll band (0.64 ± 0.15 mg m−<sup>3</sup> ; **Figures 2B**, **3**) showed similarity to earlier observations that appear persistent in the region (Moore and Abbott, 2000; Campanelli et al., 2011). This significant increase in phytoplankton biomass has been attributed to iron replenishment in the mixed layer of the Antarctic zone by the Upper Circumpolar Deep Water (UCDW; Hiscock et al., 2003) or to the wind-induced increase of the mixed layer depth (Carranza and Gille, 2015). As in previous observations (Daly et al., 2001; Han and Takahashi, 2001), elevated surface chlorophyll were not observed at fronts, but near fronts or in eddies and meanders (**Figures 2**, **3**). Further south in the summer ice edge zone, relatively low chlorophyll values (**Figure 3**) were similar to values reported in the Australian sector of the Southern Ocean in early austral summer that were attributed to the short time for phytoplankton growth following ice melt or to reduced water column stability due to strong prevailing winds (Han and Takahashi, 2001). In the TNB region of the western Ross Sea, chlorophyll biomass was relatively high (**Figures 2**, **3**) and likely associated with a late summer bloom with values in the range previously reported for surface waters in the area (Rivaro et al., 2011). Overall, surface summer chlorophyll distribution along the NZ transect supported the Southern Ocean as a HNLC region with productive sites linked to meanders and eddies and other local processes.

### CDOM Distribution and Absorption Budgets

Surface DOC concentrations in the NZSSO between the STF and SACCF (54.19 ± 4.02 µM) were comparable to those reported in the Australian sector of the Southern Ocean (45–55 µM; Ogawa et al., 1999) and summer surface concentrations in the Ross Sea (55 ± 5µM; Carlson et al., 2000). In comparison to the refractory DOC pool (∼42 µM) observed in deeper waters of the Ross Sea,

transect. (B) ternary plot of absorption coefficient at 443 nm.

and the Southern Ocean (Ogawa et al., 1999; Carlson et al., 2000), surface DOC in excess of that in deep water is likely associated with a labile or semi-labile fraction of DOM with turnover on varying time scales (hours, days, months to years) that supports heterotrophic microbial production and is potentially important for export (Carlson, 2002; Hansell, 2013).

ag325 was highest in the NZ shore station and decreased to its lowest values near the STF indicating terrestrial influences on CDOM in the northern sector of the transect. Surface patterns and range of ag325 distribution in the ACC were generally consistent with earlier reported values in the Pacific region of the Southern Ocean (Nelson et al., 2010). A general poleward increase in ag325 (0.024–0.145 m−<sup>1</sup> ) and a corresponding decreasing trend in spectral slope S between the STF and APF were likely due to a combination of factors such as (i) decreasing rates of CDOM loss poleward due to solar photodegradation,

#### TABLE 1 | Description of the four components identified by PARAFAC analysis of EEMs fluorescence data and their comparisons with previously identified components.


<sup>a</sup>Coble (1996), <sup>b</sup>Stedmon and Markager (2005), <sup>c</sup>Wedborg et al. (2007), <sup>d</sup>Murphy et al. (2008), <sup>e</sup>Catalá et al. (2015).

(ii) upwelling of the CDOM-rich subsurface waters in the Antarctic Zone and its convective transport associated with the wind-driven overturning circulation (Nelson et al., 2010; Marshall and Speer, 2012) or, (iii) to biogeneration of CDOM by bacteria (Ortega-Retuerta et al., 2009). Elevated values of the spectral slope S observed at a meander south of the STF and at fronts (e.g., NSAF, APF, SACCF; **Figures 2**, **4**) suggest that this CDOM optical property could potentially be used to identify such features in the Southern Ocean. CDOM was uncorrelated with DOC (**Figure 5A**) but positively correlated to chlorophyll (**Figure 5B**) suggesting strong biological control on CDOM in the surface waters of the study region.

The light absorption properties of CDOM and particulate matter in the visible (e.g., 443 nm) influences the light field

and ocean color. ag(443) was similar to those reported in other regions of the Southern Ocean (Clementson et al., 2001; Kieber et al., 2009; Del Castillo and Miller, 2011). aphy(443) and anap(443) between the STF and APF were comparable to those reported in the same latitudinal bands in the Australia sector of the Southern Ocean (Clementson et al., 2001), and other oceanic waters (Bricaud et al., 1998; Naik et al., 2013). anap(443) was generally <10% of the total absorption but its contribution increased in the elevated chlorophyll band south of the APF (**Figures 6A,B**) likely due to the persistence of this band and the multiple pathways for decay and regeneration of the phytoplankton biomass. Overall, CDOM contribution was greater (50–60%) than phytoplankton absorption (30–40%). However, in the high chlorophyll band south of the APF and in the northern stations off NZ, phytoplankton absorption was dominant. Terrestrial influences off NZ (station 1) and at TNB stations likely contributed to the dominant CDOM contribution (∼70 and 80%, respectively; **Figure 6B**). Thus, while the general surface characteristics of the absorption field within the ACC region appear consistent with other regions of the Southern Ocean, effects of deviations in regions such as off NZ on ocean color should be considered in its interpretation.

## FDOM Distribution along the Transect

The four PARAFAC fluorescent components identified in the surface waters of the NZSSO have also been reported in other oceanic regions (Murphy et al., 2008; Jorgensen et al., 2011; D'Sa et al., 2014; Catalá et al., 2015, 2016) including the Southern Ocean (Wedborg et al., 2007; Nelson and Gauglitz, 2016). The relative concentrations and latitudinal distribution patterns of the four components however, appeared to be influenced by physical and biological processes along the NZSSO. A positive correlation of the humic-like C4 component to chlorophyll (**Figure 9A**) suggests a strong biological control consistent with previous results documented for the global epipelagic ocean (Catalá et al., 2016). A decreasing trend of the C4 component

with increasing distance from the NZ shore as well as its high intensity in TNB suggest a strong terrestrial influence on this component in these two regions (**Figure 8A**). However, in the isolated region of the Antarctic Zone, elevated C4 values (**Figure 8A**) suggest likely contribution by upwelling associated with the UCDW and is consistent with higher intensity of this humic-like component in deeper waters of the ocean (Catalá et al., 2016; Nelson and Gauglitz, 2016). Furthermore, uniformly elevated C4 intensity in contrast to highly variable chlorophyll distribution in the Antarctic Zone suggests no direct linkage of this component to local productivity. A decreasing trend of C4 to uniformly low background values north of the APF likely due to photobleaching (Catalá et al., 2016) suggests a transformation of this humic-like material in surface waters as it flows equatorward associated with the Southern Ocean's meridional overturning circulation. The transformation of this DOM fluorescent material as the upwelled waters of the UCDW continue equatorward to form the Antarctic Intermediate Water (AAIW) and the Subantarctic Mode Water (SAMW) (Abernathey et al., 2016; Armour et al., 2016) has potential to influence DOM fluorescent composition in the global ocean (Marshall and Speer, 2012). The marine humic-like C3 component in contrast, revealed large variability across the surface waters of the NZSSO. Although a previous study indicated biological control on the distribution of this component in the global epipelagic waters (Catalá et al., 2016), such direct linkage was not observed in this study; rather, trends of elevated C3 in the productive region offshore of NZ and the STF, and the productive band in the Antarctic Zone south of the APF were observed (**Figure 8A**). This component was, however, low in the highly productive surface waters of the TNB, the summer ice edge region, and between the fronts. This pattern likely suggests elevated production of this fraction associated with microbial activity at more productive sites or at sites such as the STF with predominance of regenerated production (Kopcsynska et al., 2001). In contrast, although the TNB site was highly productive (**Figure 3**), the short-term cycling of the cold and high saline waters associated with the TNB polynia may not provide the conditions for the microbial production of this fluorescent fraction of CDOM.

The tyrosine- and tryptophan-like components C1 and C2 revealed broadly similar patterns of distribution with C1 having greater intensity and variability across the transect (**Figure 8B**). Off NZ, C1 showed a steep decline in intensity up to the STF suggesting strong terrestrial influence. It then increased significantly at the STF and thereafter declined toward the APF in concert with decreasing salinity (**Figure 8B**) suggesting some physical control (Catalá et al., 2016) as indicated by a weak negative relationship to coarse estimates of satellite-derived sea surface salinity (**Figure 9B**). However, anomalous increase of this component at or near hydrographic fronts (e.g., STF, SSAF, APF ad SACCF) were observed (**Figure 8B**) and are similar to those reported in the South Africa sector of the Southern Ocean, wherein, pronounced concentrations of the proteinlike fluorescent fractions were detected even at great depths (Wedborg et al., 2007), likely indicating linkage to more resistant DOM fractions in the deeper ocean waters (Catalá et al., 2015). Both C1 and C2 showed increases at fronts and low values between the fronts (e.g., NSAF and APF) suggesting some photooxidative loss of these components. Large variability in these two protein-like components in the Antarctic Zone could be due to both physical and biological processes associated with upwelling, elevated phytoplankton biomass and seasonal sea ice effects.

In summary, surface field data combined with satellite (salinity and chlorophyll) and historical hydrographic observations were used to understand the physical regime (specifically, the locations of fronts) and the presence of meanders along the NZSSO, a topographically complex region of the Southern Ocean. CDOM absorption and fluorescence variability along the NZSSO were then examined in relation to frontal positions and biochemical properties (chlorophyll and DOC concentrations) in addition to assessing CDOM absorption contribution to total absorption in the visible band that influences ocean color. In the STF region south of NZ, the interaction of the subtropical and subantarctic waters generally resulted in decreasing levels of chlorophyll and DOC concentrations, and CDOM optical properties that suggested terrestrial influences. However, in the core of the ACC, between the STF and the APF, surface chlorophyll and DOC concentrations and CDOM absorption were low and consistent with other studies. Fluorescence fractions identified using PARAFAC analysis were consistent with those in other oceanic regions, but revealed patterns influenced by fronts and water properties typical of the NZSSO. In the Antarctic Zone, upwelling of CDOM-rich waters likely associated with the Southern Ocean's overturning circulation and its equatorward transport and transformation could contribute to modified CDOM and FDOM properties that could serve as a tracer of changes in biochemistry associated with the overturning circulation.

#### AUTHOR CONTRIBUTIONS

ED and HK designed the study and contributed to the manuscript.

#### REFERENCES


#### FUNDING

Funding for this work was provided by a Korea Polar Research Institute (KOPRI) grant PE16040 (Satellite remote sensing on west Antarctic ocean research: STAR) and partly supported by PE17120.

#### ACKNOWLEDGMENTS

The authors would like to thank the crew of the KOPRI icebreaker and research vessel Araon and the technicians and students who participated in the Southern Ocean cruise and helped in the laboratory. The authors would like to thank NASA GSFC and JPL for making available the satellite data. The use of the Ocean Data Viewer (http://odv.awi.de) is also gratefully acknowledged. The authors would also like to thank the three reviewers for helpful comments on the manuscript.


Sci. Total Environ. 408, 3211–3222. doi: 10.1016/j.scitotenv.2010. 03.044


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 D'Sa and Kim. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.