1 EFFECTS OF GREEN TURTLE GRAZING ON CARBON DYNAMICS AND INFAUNAL COMMUNITIES IN T HALASSIA TESTUDINUM SEAGRASS MEADOWS By ROBERT AARON JOHNSON A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2019
2 Â© 2019 Robert Aaron Johnson
3 To Grace
4 ACKNOWLEDGMENTS First and foremost, I want to thank my wife, Grace Wilkinson, for all the love and support she has shown me over the course of this dissertation. The many poolside and late fo llow my dreams across numerous islands over the past six years, have helped make the data, sometimes you just need to take a step back and look at the bigger pictur e to make the story become clear. I am forever grateful to my advisors, Karen Bjorndal and Alan Bolten, for all of their advice, mentorship, and support throughout my dissertation. They have always allowed, and encouraged, me to pursue the science about w hich I am most passionate. this in my advice or mentorship to future studen ts, scientists, and individuals. Alan has taught me the importance of thinking big when planning a project, but also the importance of thinking small. His ability to consider the smallest details of a project have saved me numerous headaches in the field. I thank my parents, Greg and Gail Johnson, my sister and brother, Meg and Alec Johnson, and my in laws, Jeff and Wendy Wilkinson and Becky and Brad Hauser, for all of their love, support, and turtle related questions over the last six years. My schedule h through it all. There are many members of the ACCSTR family to whom I owe thanks, but most of all to Alexandra Gulick. From joining me for field work across numerous Caribb ean
5 incredibly grateful to have you as a friend and colleague. I hope your thumb has recovered after clipping all of that seagrass. Many thanks also to Nerine Constant for assisting me with field work and introducing me to new field sites. I would like to further thank current and former ACCSTR members Hannah Vander Zanden, Melania L Ã³ pez Castro, Mariela Pajuelo, Marco Garcia, Joe Pfaller, Luciano Soares, Cynthia Lagueux, Cat hi Campbell, and George Glen. Also a huge thanks to Kate Hanes, who conducted the field work in The Bahamas for my Chapter 5. Though no longer with us, I owe a huge debt of thanks to Peter Eliazar. From taking many days to help me in the field, to helping me ship all my equipment in Miami, my projects would not have been as successful without him. I would like to thank all of my committee members Jeremy Lichstein, Tom Frazer, and Todd Osborne for all of the support and advice they have given me. I have alw ays enjoyed the conversations we have as a group (and individually) and am grateful for the unique perspective that each of these individuals has brought to my dissertation. I, and the science I produce, am better off for it. I would like to thank Savanna Barry for all of the advice and wisdom she gave me at the beginning of my graduate career, from Gainesville, to seagrass, to research in Little Cayman. Thanks also to Jason Curtis for giving so much of his time to help me with preparing and running my many samples for carbon and nitrogen analysis. Finally, I would like to thank the numerous funding organizations, and private donors including the Melnick family, the Yoder family, Lalita Shastry, and Jeff and Monette Fitzsimmons for their generous donations w hich have made this research possible.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 10 LIST OF ABBREVIATIONS ................................ ................................ ........................... 12 ABSTRACT ................................ ................................ ................................ ................... 13 CHAPTER 1 GREEN TURTLES AND THE ROLE OF GRAZING WITHIN SEAGRASS MEADOWS ................................ ................................ ................................ ............. 15 Green Turtle Grazing ................................ ................................ .............................. 15 Present and Past R oles of Green Turtles in Seagrass Meadows ........................... 16 The Importance of Seagrass Ecosystems ................................ .............................. 17 A Return to Grazing ................................ ................................ ................................ 19 2 BLUE CARBON STORES IN TROPICAL SEAGRASS MEADOWS MAINTAINED UNDER GREEN TURTLE GRAZING ................................ .............. 20 Introduction ................................ ................................ ................................ ............. 20 Methods ................................ ................................ ................................ .................. 22 Site Description ................................ ................................ ................................ 22 Experimental Design ................................ ................................ ........................ 22 Seagrass Measureme nts ................................ ................................ .................. 23 Ecosystem Metabolism Measurements ................................ ............................ 24 Data and Statistical Analyses ................................ ................................ ........... 27 Literature Collection of Seagrass Metabolic Rates ................................ ........... 28 Results ................................ ................................ ................................ .................... 30 Effect of Grazing on S eagrass Ecosystem Metabolism ................................ .... 30 Role of Biomass in Carbon Uptake ................................ ................................ .. 32 Grazing in a Global Context ................................ ................................ ............. 32 Discussion ................................ ................................ ................................ .............. 33 Conclusions ................................ ................................ ................................ ............ 37 3 RATES OF SEDIMENT RESUSPENSION AND EROSION FOLLOWING GREEN TURTLE GRAZING IN A SHALLOW CARIBBEAN THALASSIA TESTUDINUM MEADOW ................................ ................................ ....................... 45 Introduction ................................ ................................ ................................ ............. 45 Methods ................................ ................................ ................................ .................. 48
7 Site Description ................................ ................................ ................................ 48 Study Desi gn ................................ ................................ ................................ .... 49 Seagrass and Sediment Characteristics ................................ .......................... 50 Sediment Erosion ................................ ................................ ............................. 52 Particle Deposition and Resuspension ................................ ............................. 53 Da ta Analysis ................................ ................................ ................................ ... 55 Results ................................ ................................ ................................ .................... 57 Seagrass and Sediment Characteristics ................................ .......................... 57 Sediment Erosion ................................ ................................ ............................. 58 Particle Deposition and Resuspension ................................ ............................. 59 Carbon and Organic Matter Content of Sediment Fluxes ................................ . 60 Discussion ................................ ................................ ................................ .............. 61 4 SEAGRASS ECOSYSTEM METABOLIC CARBON CAPTURE IN RESPONSE TO GREEN TURTLE GRAZING ACROSS CARIBBEAN MEADOWS ................... 78 Introduction ................................ ................................ ................................ ............. 78 Methods ................................ ................................ ................................ .................. 80 Study Sites ................................ ................................ ................................ ....... 80 Sampling Seagrass Meadow Characteristics ................................ ................... 81 Ecosystem Metabolism Measurements ................................ ............................ 83 Data Analyses ................................ ................................ ................................ .. 85 Results ................................ ................................ ................................ .................... 86 Seagrass Meadow Characteristics ................................ ................................ ... 86 Ecosys tem Metabolic Rates ................................ ................................ ............. 88 Drivers of Metabolic Rates ................................ ................................ ............... 89 Discussion ................................ ................................ ................................ .............. 90 Metabolic Dynamics of Halophila stipulacea Compared to Native Seagrass ... 92 Greater Global Assessment of Grazed Meadows Needed ............................... 94 Conclusion ................................ ................................ ................................ ........ 95 5 SIMULATED GREEN TURTLE GRAZING AFFECTS BENTHIC INF AUNA ABUNDANCE AND COMMUNITY COMPOSITION BUT NOT DIVERSITY IN A THALASSIA TESTUDINUM SEAGRASS MEADOW ................................ ........... 106 Introduction ................................ ................................ ................................ ........... 106 Methods ................................ ................................ ................................ ................ 109 Site Description and Experimental Design ................................ ..................... 109 Infaunal Sample Collection and Analysis ................................ ........................ 110 Seagrass and Sediment Sample Collection and Analyses ............................. 111 Data Analyses ................................ ................................ ................................ 112 Results ................................ ................................ ................................ .................. 114 Effects of Simulated Grazing on the Infaunal Community .............................. 115 Effects of Long Term Simulated Grazing on the Infaunal Community ............ 116 Effects of Simulated Grazing on Meadow Characteristics .............................. 117 Relationships Between Infauna Abundance and Meadow Characteristics ..... 118 Discussion ................................ ................................ ................................ ............ 119
8 6 FINAL THOUGHTS AND FUTURE DIRECTIONS ................................ ............... 136 What Have We Learned? ................................ ................................ ...................... 136 Implications for Green Turtle Grazing ................................ ................................ ... 140 Looking Forward ................................ ................................ ................................ ... 141 A PPENDIX: CHAPTER 4 SUPPLEMENTAL INFORMATION ................................ .... 143 LIST OF REFERENCES ................................ ................................ ............................. 147 BIOGRAPHICAL SKETC H ................................ ................................ .......................... 161
9 LIST OF TABLES Table page 2 1 Means and standard deviations of metabolic rates and seagrass parameters for clipped plots, reference plots, naturally g razed areas, and ungrazed areas .. 39 3 1 Seagrass meadow parameters at the beginning and end of the experiment ...... 69 3 2 Surface sediment parameters at the begi nning and end of the experiment ........ 70 3 3 Monthly rates of sediment fluxes and percent resuspension .............................. 71 3 4 Carbon and organic ma tter content of sediment fluxes ................................ ...... 72 4 1 Coordinates of each sampling site and environmental parameters .................... 97 4 2 Seagrass characteristics of grazed and ungrazed Thalassia testudinum meadows and Halophila stipulacea meadows ................................ .................... 98 4 3 Percent difference in meta bolic rates and aboveground seagrass biomass between grazed and adjacent ungrazed Thalassia testudinum meadows .......... 99 5 1 Results of linear mixed effects models evaluating effects of experimental clipping on individual infaunal groups over the course of the experiment ......... 125 5 2 Results of the PERMANOVA evaluating changes in infaunal community composition ................................ ................................ ................................ ...... 127 5 3 Results of linear mixed effect s models evaluating significant differences in abundance after 16 months of cli pping for all infaunal groups .......................... 128 5 4 Seagrass meadow characteristics in experimentally clipped and unclipped reference plots ................................ ................................ ................................ .. 129 5 5 Sediment characteristics in experimentally clipp ed and unclipped reference plots ................................ ................................ ................................ .................. 130 A 1 Shoot densities for different seagrass species, total seagrass shoot densit y, and total macroa lgae density ................................ ................................ ............ 143 A 2 Metabolic rates from grazed and ungrazed Thalassia testudinum meadows and Halophila stipulacea meadows ................................ ................................ .. 144
10 LIST OF FIGURES Figure page 2 1 A naturally grazed green turtle feeding plot and an adjacent ungrazed area in a Thalassia testudinum s eagrass meadow in Little Cayman .............................. 40 2 2 Daily metabolic rates followi ng simulated or natural grazing .............................. 41 2 3 P roduction to respiration ratios ................................ ................................ ........... 42 2 4 Metabolic rates as a function of aboveground seagrass biomass ...................... 43 2 5 Seagrass ecosystem metabolism valu es compiled from the literature ............... 44 3 1 A border between an area grazed by green turtles and an ungrazed area in a Thalassia testudinum s eagrass meadow in Little Cayman ................................ . 73 3 2 Sediment resuspension and erosion measurement methods ............................. 74 3 3 Depth of the unconsolidated surface la yer and cumulative change in sediment elevation over time ................................ ................................ .............. 75 3 4 Total downward sediment flux profiles measured with sediment traps ............... 76 3 5 Total and organic downward carbon flux profiles measured from sediment traps ................................ ................................ ................................ ................... 77 4 1 A Thalassia testudinum seagrass meadow grazed by green turtles and an adjacent ungrazed meadow and a meadow dominated by the invasive seagrass Halophila stipu lacea ................................ ................................ .......... 100 4 2 Map of sea grass meadow sampling locations ................................ .................. 101 4 3 Rates of net ecosystem production , gross primary production, and ecosystem respiration in grazed and ungrazed meadows ................................ 102 4 4 Rates of net ecosystem production in meadows dominated by the invasive seagrass Halophila stipulacea compared to nearby grazed and ungrazed Thalassia testudin um meadows ................................ ................................ ....... 103 4 5 Relationship between net ecosystem production and aboveground seagrass biomass and total seagrass shoot density ................................ ........................ 104 4 6 Rates of net ecosystem production from meadows of various seagrass specie s collected from the literature ................................ ................................ . 105 5 1 Total infaunal abundance over the course of the clipping experiment .............. 131
11 5 2 Abundance of individual infaunal groups over the course of the clipping experiment ................................ ................................ ................................ ........ 132 5 3 reference plots over the course of the clipping experiment .............................. 133 5 4 Infaunal community composition ................................ ................................ ...... 134 5 5 Relationships between total infaunal abundance an d aboveground seagrass biomass and the organic matter co ntent of the surface sediments ................... 135 A 1 Relationship between net ecosystem production and environmental temperature and irradiance ................................ ................................ .............. 145 A 2 Relationship between rates of net ecosystem production after accounting for differences in aboveground seagrass biom ass and environmental temperature and irradiance ................................ ................................ .............. 146
12 LIST OF ABBREVIATIONS C Carbon C inorg Inorganic carbon C org Organic carbon DO Dissolved oxygen Fp Primary flux Ft Total flux Fr Resuspended flux GPP Gross primary production LME Linear mixed effects model MEM Mixed effects model NEP Net ecosystem production OM Organic matter R E Ecosystem respiration
13 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EFFECTS OF GREEN TURTLE GRAZING ON CARBON DYNAMICS AND INFAUNAL COMMUNITIES IN T HALASSIA TESTUDINUM SEAGRASS MEADOWS By Robert Aaron Johnson May 2019 Chair: Karen A. Bjorndal Major: Zoology Green turtles ( Chelonia mydas ) are marine megaherbivores that consume seagrass as a part of their diet across their global range. Their distinct foraging strategy, in whic h they establish and maintain grazing patches in which all blades are cropped to a short height, structurally alters the seagrass meadows in which they graze. This can affect ecosystem processes and species relationships within these grazed areas with impo rtant implications for seagrass meadow functioning. By measuring ecosystem metabolic rates, I tested the hypothesis that green turtle grazing would reduce carbon capture rates and increase rates of ecosystem respiration within seagrass meadows. I found th at rates of metabolic carbon capture were lower in areas grazed by green turtles than areas of seagrass left ungrazed. However, rates of respiration did not increase proportionally within grazed areas relative to ungrazed seagrass, and grazed areas maintai ned positive rates of carbon capture. The effects of grazing on seagrass metabolic dynamics were tested in Little Cayman, Cayman Islands, a site with high seagrass biomass and some of the highest recorded rates of production in seagrasses. I measured ecosy stem metabolic rates at four additional locations around the Greater Caribbean and Gulf of Mexico to test the
14 hypothesis that the effects of grazing remain consistent across meadows. While the strength of the metabolic response to grazing varied among loca tions, the general response was the same: metabolic carbon capture is lower, but remains positive, in grazed meadows. Seagrass sediments are important for carbon storage, and I tested the hypothesis that sediments are more vulnerable to erosion in grazed a reas by measuring sediment erosion and resuspension in experimentally clipped areas and a naturally grazed meadow. Neither sediment erosion nor resuspension differed between grazed and ungrazed areas, demonstrating that grazed meadows may protect sediments just as well as ungrazed seagrass. Finally, I investigated the effects of grazing on the infaunal communities of seagrass meadows. I found that total infaunal abundance was reduced following experimental grazing, but abundance dynamics within individual g roups were more variable over time. I also found that composition of the infaunal community changed following grazing compared to ungrazed areas.
15 CHAPTER 1 GREEN TURTLES AND THE ROLE OF GRAZING WITHIN SEAGRASS MEADOWS Green Turtle Grazing Green turtles ( Chelonia mydas ) are marine megaherbivores that consume seagrass as a part of their diet across their circumglobal distribution (Bjorndal 1997) . Of the seven species of sea turtles worldwide, green turtl es are the only species that consumes seagrass as a part of its diet. Green turtles exhibit a distinct foraging re graze. Green turtles establish grazing patches by cro pping (biting) all seagrass blades within an area just above the blade sheath junction (where the seagrass blades grow out from a protective, sheath like covering growing vertically from the belowground horizontal rhizomes) near the sediment surface. They then continually return to and re graze these same patches, creating areas within a meadow in which seagrass blades are kept cropped to a uniform, short height, akin to a mowed lawn. Initially, when green turtles crop seagrass blades in an area (those not yet grazed), they let the upper portions of the blades float away (Bjorndal 1980) . They then consume only new seagrass tissue which has grown in since the previous grazing. The new tissue growth is proportionally higher in nitrogen content (Moran and Bjorndal 2007) and free from calcareous epiphytes that grow on older portions of blades (Mortimer 1981) , resulting in a food source that is of higher nutritional quality for the turtles (Zieman et al. 1984) . This grazing strategy results in an area that is structurally altered compared to areas of a meadow left ungrazed. In addition to the direct effects that grazing has on the seagrass plants, there are also numerous ecosystem processes that are affected when turtle create and maintain their grazing patches.
16 Present and Past Roles of Green Turtles i n Seagrass Meadows Green turtle grazing plays an important ro le in the ecology and functioning of seagrass meadows. Removing the seagrass canopy within grazing patches opens up the benthic surface in a meadow. This may have various effects, such as altering the strength of physical processes (e.g. wave turbulence), or changing predator prey dynamics. Many species use seagrass meadow canopies as a source of refugia (Hemminga and Duarte 2000) , a function which may no longer be provided once the canopy has been removed through grazing. Cropping away seagrass blades also removes a source of organic matter (senescent seagrass blades and associated epiphytes) for the benthic detrital pool (Kennedy et al. 2010; Oreska et al. 2017b) . Green turtles may return some o f the organic matter and nutrients they consume to the meadows through excretion and defecation (Thayer et al. 1982; personal observation ) ; however, this is not likely a large contribution (Bjorndal 1980; Thayer et al. 1984; Lee and Dunton 1999) . Grazing may also play a role in meadow disease dynamics. It had been thought that by removing older (longer), pathogen ridden blades, green turtles grazing may aid in the overall he alth of meadows (Jackson et al. 2001) . However, experimental evidence has shown that seagrass wasting disease (genus Labyrinthula ) may preferentially colonize the recently grazed ends of Thalassia testudinum seagrass blades (Bowles and Bell 2004) , spreading disease more rapidly in some grazed areas. Historically, green turtle population sizes were much greater than those observed t oday (Jackson 2001; Jackson et al. 2001) . While the true sizes of these once great populations in areas such as the Caribbean are not known, from historical records (Jackson 1997) we know that current numbers of turtles (and those grazing in seagrass meadows) are a fraction of what they were prior to overexploitation by humans (Table 2
17 in Jackson 1997) . Estimates of these historical p opulations sizes range from tens (from fishing data) to hundreds (based on carrying capacity from seagrass area) of millions of green turtles in the Caribbean alone (Jackson 1997) . Regardless of whether green turtle populations existe d at carrying capacity, the difference in these historical abundance estimates from current abundances implies that green turtle grazing certainly played a different, and more prominent, role in seagrass meadows of the past. Green turtle abundances are cur rently increasing (Chaloupka et al. 2008; Mazaris et al. 2017) , which will lead to more seagrass area being returned to a natural grazed state. Given the importance of seagrass ecosystems, it is critical to better understand how grazing affects these meadows. The Importance o f Seagrass Ecosystems Seag rass meadows provide important ecosystem services. Many animal species, from economically important fishes to invertebrate infauna (Hemminga and Duarte 2000) rely on seagrass meadows for a place to forage or as a place of refuge. Aboveground seagrass blades also baffle currents and attenuate wave energy, creating a calmer physic al environment within the seagrass canopy (Fonseca et al. 1982; Gacia et al. 1999; Reidenbach and Thomas 2018) . This allows particles to settle out of the water column and onto the sediment surface where they are subsequently more protected from resuspension and loss (Gacia and Duarte 2001; Hendriks et al. 2008) , important processes for organic mat ter and sediment accretion (Gacia et al. 2002, 2003) . Belowground seagrass rhizomes can also play an impor tant role in sediment stabilization and seem to be particularly important for shoreline stabilization. Even a nearshore seagrass meadow that had been grazed by green turtles in Indonesia, but which still had an intact rhizosphere, was able to prevent shore line erosion (Christianen
18 et al. 2013) , which can help prevent costly beach re nourishment programs. Seagrasses also filter excess nutrients and pathogens from the water, with benefits not only to humans, but also nearby habitats, such by reducing disease incidence in adjacent coral reefs (Lamb et al. 2017) . Seagrasses are recognized as hotspots of carbon sequestration and storage within the oceans (Fourqurean et al. 2012) , and their protection and conservation has been suggested as a climate change mitigation strategy (Murdiyarso et al. 2015; Macreadie et al. 2017) . Through high rates of metabolic carbon capture (Duarte et al. 2010) , biomass growth (Duarte and Chiscano 1999) , and particulate organic matter capture (Gacia et al. 2002; Kennedy et al. 2010) , seagrass meadows have some of the highest carbon sequestration rates (per unit area) on the planet. Once captured, this carbon may b ecome buried and incorporated into the long term carbon storage pool. The majority of carbon stored within seagrass meadows is stored in the belowground sediment (Fourqurean et al. 2012) . The sediments underneath seagrass meadows are conducive to long term storage on the order of millennia due to low rates of organic matter remineralization as a result of anoxia beginning a few centimeters below the sedimen t surface (Mateo et al. 1997) . Disturbances to meadows, or the loss of seagrass, may make these carbon stocks vulnerable to loss, however. Globally, we are losing about 1.5% of the total seagrass area each year, with anthropogenic activities, coastal development, and eutrophication as major drivers of this loss (Waycott et al. 2009) . There is some concern however , that increased grazing pressure may lead to overgrazing and a loss of seagrass, as has occurred in isolated cases in Bermuda
19 (Fourqurean et al. 2010) and Indonesia (Christianen et al. 2014) , leading to a loss of carbon sequestration and storage (Atwood et al. 2015) . A Return t o Grazing As we prepare for a future with recovering green turtle populations, leading to more grazed seagrass, it is critical to understand how grazing affe cts these meadows, so that management strategies and protections can be applied most effectively to green turtle populations and seagrass ecosystems simultaneously. Throughout my dissertation research, I have focused on investigating how green turtle grazi ng effects the ecology of seagrass ecosystems with an emphasis on carbon dynamics and infaunal communities. My main goal in pursuing these types of research questions has been to further our understanding of green turtle grazing in tropical seagrass meadow s, so as to better prepare for a future in which we have more green turtles in our oceans and more (beautiful!) grazed seagrass habitats. In the following four chapters I detail my findings on how grazing affects both ecosystem metabolic and sediment carbo n dynamics, and the invertebrate infauna populations within seagrass meadows, and what these mean for the ecology of seagrass meadows in which green turtles are foraging.
20 CHAPTER 2 BLUE CARBON STORES IN TROPICAL SEAGRASS MEADOWS MAINTAINED UNDER GREEN T URTLE GRAZING Introduction Seagrass meadows form some of the most productive ecosystems in the world (Duarte and Chiscano 1999) . carbon buried by vegetated marine systems each year through high rates of production and organic matter burial (Duarte et al. 2005; McLeod et al. 2011; Fourqurean et al. 2012) . The majority of this carbon is stored belowground in the sediments, where anoxic conditions can result in storage for millennia (Mateo et al. 1997) . This suggests that conserv ation and restoration of seagrass systems could be used as a climate change mitigation strategy (Duarte et al. 2013b; a; MarbÃ et al. 2015) . Green turtles ( Chelonia mydas ) are megaherbivores that consume seagrass as a large part of their diet across much of their global range. Green turtles establish feeding plots in which they forage by cropping seagrass blades at or near the sediment surface (Fig. 2 1) and repeatedly re grazing new growth within these plots (Bjorndal 1980) , the reby structurally altering the meadow (Moran and Bjorndal 2005) . With successful conservation leading to increasing green turtle populations in some areas (Chaloupka et al. 2008) , seagrasses will increasingly be subjected to grazing pressure in addition to anthropogenic disturbances (Short and Wyllie Echeverria 1996; Duarte 2002; Orth et al. 2006; Christianen et al. 201 4) . Supporting abundant green turtle populations and sustaining a carbon sink are both important conservation aims for global seagrass ecosystems. Grazed meadows with intact grazer populations is the natural state of seagrass ecosystems. Historically, when green turtle abundance was much higher than today (Jackson 2001;
21 McClenachan et al. 2006) , Caribbean seagrass meadows supported extensive grazing (Jackson 1997) , with the majority of meadows likely in a grazed state before humans disrupted these coevolved systems through overexploitation of green turtles (Jackson 2001; Jackson et al. 2001) . While green turtle grazing helps maintain meadow health by removing older, pathogen susceptible seagrass blades (Bjorndal 1980; Jackson 2001) , grazing also reduces the size of the photosynthetic seagrass canopy that is capable of allocating production belowground for storage (Bjorndal 1980; Alcoverro et al. 2001; Mateo et al. 2006) . It has been hypothesized that grazing may have negative effects on seagrass meadow carbon sequestration and storage, and tha t the conservation of both green turtles and seagrass carbon stores are incompatible (Atwood et al. 2015) . To better conserve these ecosystems, it is necessary to under stand how ecosystem processes, such as carbon sequestration, operate within seagrass meadows in their naturally grazed state. We hypothesized that: (1) metabolic carbon uptake rates (net ecosystem production) would be lower in grazed compared to ungrazed a reas as a result of reduced aboveground biomass, and (2) carbon remineralization rates (ecosystem respiration) would be proportionally higher in grazed than in ungrazed areas as a result of increased heterotrophic respiration due to aeration of surface sed iments following removal of the seagrass canopy. We conducted an experimental manipulation in Little Cayman, Cayman Islands, by clipping seagrass ( Thalassia testudinum ) to simulate green turtle grazing. We measured areal rates of gross primary production ( GPP), ecosystem respiration (R E ), and net ecosystem production (NEP = GPP R E ) weekly with benthic incubation chambers in five experimentally clipped plots and five unclipped reference plots to
22 investigate changes in carbon dynamics following the onset of simulated grazing. Metabolism (GPP, R E , NEP) was similarly measured in nearby areas that were naturally, actively grazed for at least a year by juvenile green turtles and adjacent ungrazed areas, and these results were compared to those from the experimen tal and reference plots. To evaluate our measured rates and the effects of green turtle grazing in a broader geographical context, we also compiled published estimates of seagrass ecosystem metabolism. Methods Site Description This experiment was conducte d in seagrass meadows within Grape Tree Bay on Central Caribbean Marine Institute during May through August, 2016. Turtle grass ( Thalassia testudinum ) was the dominant se agrass comprising the meadows with interspersed manatee grass ( Syringodium filiforme ) and small amounts of shoal grass ( Halodule wrightii ) in some areas. The benthic habitat was comprised of carbonate sediments. The meadow was located roughly 40 m from sho re in shallow water with a mean depth of 1.0 m and small tidal variation (Â± 0.2 m). Mean height of the seagrass canopy was 15.8 cm with a mean T. testudinum density of 840 shoots m 2 . Areas that were naturally grazed by green turtles for at least a year we re present nearby in Grape Tree Bay. Naturally grazed areas had a mean T. testudinum blade length (canopy height) of 1.9 cm and a mean density of 776 shoots m 2 . Experimental Design We conducted an in situ clipping experiment to simulate grazing by green turtles. Ten 2 x 2 m plots were set up in an ungrazed area of the seagrass meadow. Five plots
23 were experimentally clipped to simulate green turtle grazing, and five remained unclipped to serve as reference plots. All variables were measured in both clippe d and reference plots prior to the onset of clipping to ensure that any changes measured were due to simulated grazing, and not previous differences. Prior to any measurements, the seagrass rhizomes were severed around the edges of all plots using a flat b laded shovel to prevent the translocation of nutrients into the experimental plots from the surrounding unclipped meadow (Moran and Bjorndal 2005) . This was done to simulate a grazed area larger than 2 x 2 m, which is at the smaller end of the size range of natural grazing plots (Ogden et al. 1983; Williams 1988a; Kuiper Linley et al. 2007; Holzer and McGlathery 2016) . Reference plots were also severed to ensure that any effec ts seen in the clipped plots were the result of clipping, and not from severing rhizomes. All measurements were made and samples collected from the central 1.5 x 1.5 m of each plot so as to leave a 25 cm buffer zone around plot edges to avoid edge effects. Clipping was initiated in May 2016 and maintained for twelve weeks. Clipped plots were initially established by clipping all blades within a plot just above the blade/sheath junction using scissors and collecting all clipped portions of the blades. Blades were re clipped when blade length in the plot was ~5 cm above the blade/sheath junction to mimic natural turtle grazing (Williams 1988a; Moran and Bjorndal 2005) . This method resulted in clipping every ~14 days (range 12 15). S eagrass Measurements Seagrass species composition and shoot density were measured bi weekly in all plots. Data were collected from three randomly placed 25 x 25 cm (0.0625 m 2 ) quadrats within each plot. Aboveground biomass samples were collected bi weekly in clipped
24 plots and monthly in reference plots from three 10 x 10 cm (0.01 m 2 ) quadrats in each plot by clipping all shoots within the quadrat at the sediment surface. In the lab, blades were measured for length and width, gently scraped with a razor blad e to remove any epiphytes, and rinsed in seawater. Samples were then dried at 60 Â°C for at least 24 hours before weighing for dry mass. Belowground biomass samples were collected at the beginning and end of the experiment. Samples at the beginning were co llected adjacent to the ten experimental plots to avoid destructive sampling within plots, and samples at the end were collected from the middle of each plot. Samples were collected using cylindrical PVC sediment cores (7.7 cm diameter) sharpened at one en d and hammered into the sediment. All cores were deep enough to completely penetrate through the root/rhizome mat, resulting in a complete belowground biomass sample. Samples were refrigerated following collection and then processed within 48 hours. All ab oveground biomass was removed from samples. Samples were cleaned of sediments using running seawater and then dried at 60 Â°C for at least 48 hours or until completely dry. Any remaining sediments were then removed, and samples were re dried (60 Â°C) before weighing for dry mass. These same methods were used for measuring aboveground biomass, belowground biomass, and seagrass parameters in the naturally grazed and ungrazed areas. Rhizomes were not severed in the naturally grazed or ungrazed areas as had been done around the edges of our experimental clipped and reference plots. Ecosystem Metabolism Measurements Seagrass NEP and R E measurements were made weekly using benthic incubation chambers. Sampling was prevented during weeks two and ten due to
25 hazardous weather from tropical storms Colin and Earl. Chambers were comprised of a gas tight, polyethylene bag with a sampling port attached to a rigid PVC cylinder (Hansen et al. 2000; Bar rÃ³n et al. 2004; Calleja et al. 2006) . PVC cylinders (16 cm inner diameter, encompassing 0.02 m 2 of bottom area) were sharpened at one end and inserted roughly 7.5 cm into the sediment (not severing rhizomes) (BarrÃ³n et al. 2004; Calleja et al. 2006) . These were inserted in the sediment one day prior to the incubations to allow effects of the disturbance to dissipate prior to the incubation (Ziegler and Benner 1999) . The polyethylene bags were then attached to the PVC cylinders with elastic bands and hose clamps. The use of a flexible bag allowed the propagation of turbulence to the interior of the chamber allowing i nternal mixing to more accurately mimic natural conditions (BarrÃ³n and Duarte 2009) . Chamber volume was measured in the lab to be 5.5 6.0 L. NEP and R E were calculated from the change in dissolved oxygen (DO) concentration following an incubation period using light and dark (opaque) incubation chambers, respectively. Three water samples were collected in 60 ml plastic syringes (light or dark depending on the chamber) at the beginning and end of the incubation period from each chamber. Syringe samples were capped and returned to the surface immediately following collection, and DO measurements were taken directly in the syringe using a handheld optical DO meter (YSI ProODO), which was calibrated in w ater saturated air on the morning of each sampling day (Vadeboncoeur 2011) . We tested this method in the laboratory under various conditions prior to the field study to test and confirm the reliability of the method and the pre cision of YSI ProODO meters for this application. Mean incubation length in the field was 2.3 hours (range 1.5 3.3
26 hrs), as longer incubation periods may underestimate metabolic rates (OlivÃ© et al. 2015) . Incubations were always started by 1130 hours in order to encompass solar maximum. There were two instances in which a dark chamber gained oxygen during the incubation, suggesting an error. The gain in O 2 during both of these instances was s mall, and within the margin of error of the DO probe, so these rates of respiration were assumed to be zero. Water column metabolism was measured in a similar manner using clear and dark 300 ml BOD bottles. Three clear and three dark bottles were filled un derwater at seagrass canopy height, capped, and attached to a PVC rod to incubate under in situ conditions. Three water column samples were collected in 60 ml plastic syringes at canopy height at the same time to measure initial water column DO concentrati on for the bottle incubations. One 60 ml plastic syringe was then collected from each BOD bottle following the incubation to measure final DO concentration. During instances when a dark bottle gained oxygen during the incubation (suggesting sampling error) DO changes were assumed to be zero. BOD bottle metabolic rates were subtracted from chamber rates to correct for water column metabolism and ensure that only metabolic rates of the benthic community were measured (Stutes et al. 2007) . Water column metabolism played a minor role in this seagrass system. On average it contributed <5% of total reference plot metabolism within benthic chambers; however, it p layed a larger role on the two overcast days. In addition to the experimentally clipped and reference plots, NEP and R E were also measured using these same methods in nearby areas that were naturally grazed by green turtles (>1 year) and adjacent ungrazed areas of meadow. Three light and
27 three dark incubation chambers were used on three occasions to measure NEP and R E, respectively, in both naturally grazed and ungrazed areas. Data and Statistical Analyses Hourly rates of NEP and R E were calculated from c hanges in DO concentration measured in light and dark incubation chambers, respectively. Hourly GPP was calculated as the sum of NEP and the absolute value of R E . Daily rates were calculated by multiplying GPP by the photoperiod (10 hours) and R E by 24 hou rs, and daily NEP was calculated as the difference between daily GPP and R E . Since all incubations were conducted during the middle of the day, we corrected length of daylight (13 hours) for dawn and dusk hours by assuming minimal production during the 1.5 hours on either end of the daylight period. We based estimates of GPP on the central 10 hours of daylight when solar irradiance values (HOBO Pendant data loggers) were comparable to those measured during our incubations. We assumed daytime and nighttime R E to be equal for our calculations, and a deviation from this could result in a slight change in our calculated daily metabolic rates. Oxygen units were converted to molar units, and then converted to carbon units using photosynthetic and respiratory quoti ents of one (BarrÃ³n and Duarte 2009) . While we measured ecosystem metabolism during the summer season, previous studies have shown that seagrass ecosystems are typically net autotrophic (NEP > 0) ac ross the annual cycle, possibly becoming heterotrophic for only one to a few months per year (Stutes et al. 2007; Anton et al. 2009; Apostolaki et al. 2010) . Additionally, our study was conducted in a shallow, tropical location where water temperatures and incident sunlight do not vary greatly across seasons. We therefore do not feel that the conclusions of this study would be qualitatively different had we conducted the experiment for a full annual cyc le.
28 The amount of aboveground seagrass biomass contained within incubation chambers was calculated by interpolating between clipping events. Assuming aboveground biomass to be zero immediately following clipping, we calculated daily biomass production rate s using the aboveground biomass measured at time of clipping and the number of days since the previous clipping event. Using this calculated daily rate of biomass production, we estimated what mean areal aboveground biomass was for each plot during incubat ions based on how many days had elapsed between the incubation and the previous clipping event. All calculations and statistical analyses were performed in R version 3.3.2 (R Core Team 2018) , (Wickham 2007) (Pinheiro et al. 2018) packages. The effect of simula ted grazing on seagrass meadow GPP, R E , and NEP over time was evaluated using a mixed effects model with treatment and time as fixed factors and individual plot as a random factor. Linear regression was used to evaluate the relationship between GPP and R E (production to respiration ratio) as well as seagrass meadow metabolic rates (GPP, R E , and NEP) with aboveground seagrass biomass. Unpaired t tests (two tailed), were used to evaluate differences in seagrass characteristics and aboveground biomass between clipped plots and naturally grazed areas. A one way ANOVA was used to evaluate differences in the production to respiration ratio among treatments. A two way ANOVA was used to evaluate differences in belowground seagrass biomass among treatments, with tre atment and time as factors. Literature Collection of Seagrass Metabolic Rates We searched the scientific literature for available estimates of seagrass metabolic rates to compare to those from this study by searching ISI Web of Science in
29 October 2016 usi Only results from in situ measurement of whole system seagrass metabolic rates were extracted. Mesocosm studies and metabolic measurements from individual seagrass shoots were excluded. Additionally, only values from unmanipulated seagrass were used if a manipulation was condu cted (e.g. experimental shading), then only results from the control/reference treatment were extracted. In cases when multiple measurements were made within the same site (spatially or temporally) in a study, the mean was calculated for these replicates t o obtain a single set of metabolic rates (GPP, R E , NEP) for each site. Duarte et al. (2010) compiled a database of global seagrass me tabolic rates, and data from this database were also used to supplement the values we extracted from the literature. All values obtained from the Duarte et al. (2010) database were confirmed in their original publication. In a few cases we were unable to obtain the original publication, and since we were unable to confirm these values or their method of collection they were excluded. Unp ublished data, and data from studies that did not meet our requirements above were excluded. This resulted in 58 unique estimates of seagrass metabolic rates 54 from the literature, and four new values from our study. Thirty eight of these estimates come from studies included in the Duarte et al. (2010) database, and we have compiled an additional 20 estimates of seagrass metabolism her e. All data extracted from literature sources or the Duarte et al. (2010) database were recorded in their reported units, and
30 then con verted to units of mmol C m 2 d 1 if needed. Values reported in oxygen units were converted to carbon units using photosynthetic and respiratory quotients of one. Not all studies from which values were obtained used the same methods for measuring metaboli c rates. While the vast majority of studies used in situ incubation chambers, bell jars, or the diel O 2 curve method to measure seagrass metabolic rates, the length of time used for incubations or measurement varied among studies. It has been demonstrated that measured metabolic rates can be strongly affected by incubation length, and that longer incubations (e.g., 12 or 24 hour) tend to underestimate rates due to oxygen saturation or depletion within the chamber (OlivÃ© et al. 2015) . It is therefore possible that some seagrass metabolic rates reported in the literature are low due to methodological reasons, and the true rates in these systems may be higher than those reported. Results Effect o f Grazing on Seagrass Ecosystem Metabolism Ecosystem metabolism (GPP, R E , NEP) was significantly lower in experimentally clipped plots compared to reference plots (mixed effects model; GPP, n = 83, F 1,8 = 106.4, p < 0.0001; R E , n = 83, F 1,8 = 75.3, p < 0.0 001; NEP, n = 83, F 1,8 = 34.8, p = 0.0004; Table 2 1). This difference persisted for the duration of the experimental manipulation (Fig. 2 2). Prior to the onset of clipping there were no differences in measured metabolic rates between clipped and unclipped reference plots (t t est; GPP, n = 8, t 6 = 1.3, p = 0.23; R E , n = 8, t 6 = 0.6, p = 0.55; NEP, n = 8, t 6 = 1.2, p = 0.27), and all plots were autotrophic (NEP > 0). During the experiment, GPP was 77% lower, R E 74% lower, and NEP 79% lower in clipped plots on average compared to reference plots (Table 2 1). Clipping reduced aboveground seagrass biomass by an average of
31 80%, but this was not enough to shift clipped plots from positive to negative NEP, and they remained metabolic carbon sinks (NEP > 0) for the duration of the 12 week experiment. The two occasions (weeks one and nine; Fig. 2 2) when clipped plots were slightly heterotrophic (NEP < 0) and reference plots were reduced to near metabolic balance (NEP = 0) was a result of low GPP when incubations were conducted on over cast days. We measured ecosystem metabolism using the same methods in naturally grazed areas and adjacent ungrazed areas of seagrass located near our experimental plots. These naturally grazed areas were actively maintained by juvenile green turtles and h ad been grazed continuously for at least one year. Net ecosystem production in naturally grazed areas was 92% lower than adjacent ungrazed areas on average, and we compared results from these naturally grazed areas, representing long term effects of grazin g, to those from our experimentally clipped plots, representing short term effects of grazing. Metabolic rates in the naturally grazed areas were similar to those in our clipped plots (Fig. 2 2). This was unexpected given that aboveground biomass was signi ficantly lower in the naturally grazed areas than our clipped plots (t test; n = 9, t 7 = 8.5, p < 0.0001; Table 2 1). While seagrass shoot density was not significantly different between naturally grazed areas and clipped plots, seagrass blades in the natu rally grazed areas were significantly shorter (t test; n = 9, t 7 = 5.2, p = 0.0012) and narrower (t test; n = 9, t 7 = 6.3, p = 0.0004) than in the clipped plots yielding less photosynthetic leaf area (Table 2 1). Unlike certain seagrass parameters, such as shoot density and blade width, which may take months to become reduced following the onset of grazing (Moran and Bjorndal 2005) , seagrass ecosystem metabolism (GPP, R E , NEP)
32 experienced a rapid reduction following the onset of simulated grazing, a fter which rates remained relatively stable (Fig. 2 2). The relative contribution of GPP and R E to the total metabolism of the ecosystem (measured by the production to respiration ratio, P:R) did not differ among treatments (clipped, reference, grazed, un grazed; ANOVA; n = 99, F 3 = 1.8, p = 0.16; Table 2 1), even though rates of GPP and R E were lower in clipped plots and naturally grazed areas than unclipped reference plots and ungrazed areas. GPP was also strongly, positively correlated with R E (linear re gression; n = 101, R 2 = 0.66, p < 0.0001; Fig. 2 3). High P:R ratios (range 1.9 2.4; Table 2 1) show the meadow was strongly autotrophic (NEP > 0) irrespective of green turtle grazing natural or simulated. Role of Biomass in Carbon Uptake Aboveground se agrass biomass (dry mass; DM) was strongly and positively correlated with measures of ecosystem metabolism (linear regression; GPP, n = 96, p < 0.0001; R E , n = 96, p < 0.0001; NEP, n = 96, p < 0.0001; Fig. 2 4) and explained 69%, 58%, and 37% of the variab ility in GPP, R E , and NEP, respectively. Aboveground biomass fluctuated in clipped plots (range 0.0 72.9 g DM m 2 ) with the clipping regime while biomass in the reference plots remained relatively high (range 205.9 307.7 g DM m 2 ) during the experiment . Belowground biomass did not differ between clipped plots, reference plots, naturally grazed areas, or ungrazed areas (ANOVA; n = 34, F 2 = 0.5, p = 0.611; Table 2 1). Grazing in a Global Context To evaluate the results of our experiment from Little Cayman in a broader geographical context, we compiled estimates of seagrass metabolic rates (n = 58) from the published literature. Reported rates of NEP from various seagrass species and
33 areas around t he world (including this study) ranged from 62.5 to 209.5 mmol C m 2 d 1 with a median of 20.6 (Fig. 2 5c). A majority of systems (81%) had net positive NEP, including those measured either from single sampling events or over the annual cycle. NEP in ungr azed T. testudinum meadows in Little Cayman (209.5 mmol C m 2 d 1 ) was higher than rates measured elsewhere, and NEP in our unclipped reference plots was also high, due to the high GPP relative to R E in this system. Higher rates of both GPP and R E have bee n measured in other seagrass systems than the rates we measured in Little Cayman (Fig. 2 5 a and b), but the P:R ratio was always closer to one, resulting in lower rates of NEP for those systems (Lindeboom and Sandee 1989; Pollard and Moriarty 1991; Reyes and Merino 1991; Ziegler and Benner 1998; Koch and Madden 2001; Plus et al. 2001; Nagel et al. 2009; Champenois and Borges 2012; Rheuban et al. 2014) . Discussion Despite significantly lower rates of carbon uptake (inferred from NEP), our results show that tropical T. testudinum meadows remained active metabolic carbon sinks, even under long term sustained grazing pressure. Though green turtle grazing re moves much of the aboveground seagrass biomass, the belowground biomass is normally left intact in T. testudinum meadows (Table 2 1). Production may still be allocated to belowground tissues for long term storage, thus allowing the meadow to remain a metab olic carbon sink in the presence of green turtle grazing. An example in Indonesia is an exception to this, in which green turtles became hyper abundant (20 individuals ha 1 ) within a marine protected area and began digging and consuming belowground seagras s tissues ( Halodule uninervis ) when aboveground biomass alone was not enough to sustain the population (Christianen et al. 2014) . The turtles may have been
34 able to uproot and consume H. uninervis due to its shallow rhizosphere (MarbÃ and Duarte 1998) as compared to other seagrasses such as T. testudinum th at form stronger and denser rhizome mats. To our knowledge, this behavior has not been observed elsewhere. In addition, green turtles often do not graze an entire meadow, but rather discrete patches (B jorndal 1980; Lal et al. 2010; HernÃ¡ndez and van Tussenbroek 2014) . Though NEP was 92% lower in areas that had been grazed long term by turtles, the reduction in whole meadow NEP would be less than this, as NEP in ungrazed areas remained unaffected. Cha nges in the strength of the ecosystem metabolic carbon sink from grazing are therefore dependent upon the proportion of grazed to ungrazed areas within the meadow. The strong relationship between metabolic variables (GPP, R E , NEP) and aboveground seagras s biomass indicates that metabolic carbon dynamics in this system were largely driven by the seagrass rather than other potential producers such as epiphytes or microphytobenthos. This relationship also demonstrates that under a sustained green turtle graz ing regime, rates of metabolic carbon uptake will remain lower than ungrazed areas. If turtles abandon or are excluded from an area, rates of carbon uptake will increase concomitantly with seagrass regrowth. This is further evidenced by fluctuations in NEP in our experimentally clipped plots. NEP was lower (15.8 Â± 7.2 mmol C m 2 d 1 ) when measured 2 3 days after plots were clipped and aboveground biomass was low (10.5 Â± 0.8 g DM m 2 ; even numbered weeks; Fig. 2 2c), and higher (57.4 Â± 19.9 mmol C m 2 d 1 ) w hen measured 8 10 days post clipping and biomass had increased (33.9 Â± 9.9 g DM m 2 ; odd numbered weeks).
35 Our results show that meadows grazed by green turtles can maintain net positive metabolic carbon uptake, but there is concern as to what effect recov ering green turtle populations and increased grazing will have on current seagrass blue carbon stocks (Atwood et al. 2015) . It has been suggested that following extreme seagrass degradation or loss, such as from overgrazing, carbon stored in the top meter of sediment may be vulnerable to remineralization and loss from the system (Fourqurean et al. 2012; Pendleton et al. 2012) . The producti on to respiration ratio (P:R) of the system can be used to investigate this indirectly, where increased microbial activity and organic carbon remineralization can be inferred from a decrease in the ratio. Ecosystem respiration includes both respiration by the autotroph community (R A ) and respiration by the heterotroph community (R H ). Sediments in seagrass meadows are often anoxic below the first few millimeters to centimeters (Terrados et al. 1999; Duarte et al. 2013a) , and we had predicted microbial activity and R H would increase following removal of the seagrass canopy due to increased water flow and aeration of the surface sediments. An increase in R H relative to R A would increase R E relative to GPP and therefore decrease the P:R ratio. That the P:R ratio was not affected by short term experimental clipping or long term natural grazing indicates that rates of R E were largely driven by the amount of aboveground seagrass biomass present rather than by R H in the benthos. Anaerobic metabolism and subsurface carbon dynamics can also play a role in seagrass meadow carbon cycling. We were unable to measure these processes in our study; however, we feel that these were likely to play a small role in total carbon dynamics in thi s system. Following a decrease in primary production and oxygen
36 translocation to the rhizosphere, organic matter remineralization typically switches from aerobic respiration to sulfate reduction (Call eja et al. 2006) ; however, rates of sulfate reduction are known to be low in carbonate based sediments (MarbÃ et al. 2006) , such as those in Little Cayman. Additionally, carbon dioxide produced from belowground remineralization of organic carb on may be consumed via carbonate dissolution, rather than released from the system (Burdige and Zimmerman 2002; MarbÃ et al. 2006) . Grazing does lead to a loss of potential future blue carbon sequestration (carbon that may have become sequestered) as a result of lower NEP in grazed areas c ompared to ungrazed areas, but it did not affect the ecosystem P:R ratio. The meadow maintained net positive carbon uptake and high GPP relative to R E , indicating that grazing is not likely to lead to the metabolic release of blue carbon already stored in tropical T. testudinum meadows, and these carbon stocks may remain intact in the face of increasing grazing pressure. A similar relationship between grazing and sediment carbon has been shown in a Canadian salt marsh areas that had been grazed long term by sheep exhibited higher soil organic carbon content as well as higher belowground biomass production than ungrazed areas (Yu and Chmura 2010) . While seagrass meadows are resilient to long term grazing (Moran and Bjorndal 2005) , the resilience of meadows to external stressors (e.g. decreased light availability from coastal ru noff and sedimentation/eutrophication) may differ between grazed and ungrazed areas experiencing high stress levels. Future research into the effects of grazing in meadows experiencing high stress levels, such as from reduced light availability, would be b eneficial in furthering our understanding of the ecological effects of green turtle grazing.
37 In comparison with seagrass metabolic values from the literature (Fig. 2 5), rates of GPP in our experimentally clipped plots and the naturally grazed areas are l ow (among studies that met our criteria for inclusion, Methods; Fig. 2 5a). However, the high rates of GPP relative to R E in our system resulted in higher NEP than that reported for many other seagrass systems (Fig. 2 5c). Some of these previously publishe d rates could be low due to methodological reasons, as incubation time has been shown to influence measured metabolic rates (OlivÃ© et al. 2015) (Methods). NEP in both our clipped plots and nat urally grazed areas (24.7 and 17.1 mmol C m 2 d 1 , respectively) was near the median (20.6 mmol C m 2 d 1 ) of seagrass ecosystem metabolism estimates compiled from the literature. This suggests that even under a scenario of increasing green turtle grazing pressure, Caribbean T. testudinum meadows could still function as a stronger metabolic carbon sink than some ungrazed seagrass meadows in other areas. Conclusions Overexploitation of green turtle populations over the past several centuries has led to a sh ifting in the baseline of what is considered natural for seagrass ecosystems. Green turtle grazing in seagrass pastures is the natural condition. With successful conservation leading to increasing green turtle abundance, albeit still below historical numbe rs (Jackson 2001; McClenachan et al. 2006) , it is important to understand how seagrass ecosystem carbon d ynamics will be affected. Seagrass meadows are important sites of blue carbon sequestration and storage (Duarte et al. 2010; Fourqurean et al. 2012) , and there is concern these functions may be affected as increased grazing pressure returns more seagrass meadows to their natural grazed state (Heithaus et al. 2014; Atwood et al. 2015) . Here we show, through an in situ
38 seagrass manipulation experiment and measurement of areas naturally grazed by green turtles, that rates of metaboli c carbon uptake are lower in grazed areas than ungrazed areas. These differences in NEP correspond to a reduction in the potential of the meadow to sequester blue carbon in the future. However, grazing did not affect the P:R ratio of the meadow on short o r long term time scales, suggesting that even sustained green turtle grazing is not likely to lead to a loss of sediment carbon through remineralization. These findings indicate that as more tropical seagrass habitats are returned to a natural grazed state , rates of carbon uptake and contribution to the metabolic carbon sink will be lower, but there will not be a large metabolic release of current blue carbon stocks.
39 Table 2 1. Means and standard deviations of metabolic rates and seagrass parameters for clipped plots, reference plots, naturally grazed areas, and ungrazed areas. Variable Clipped Reference Grazed Ungrazed Test P value GPP (mmol C m 2 d 1 ) 64.4 Â± 40.4 a 275.5 Â± 69.9 b 36.1 Â± 5.4 370.0 Â± 15.0 MEM < 0.0001 R E (mmol C m 2 d 1 ) 39.7 Â± 27.4 a 154.6 Â± 41.2 b 19.1 Â± 9.9 160.4 Â± 24.3 MEM < 0.0001 NEP (mmol C m 2 d 1 ) 24.7 Â± 37.6 a 119.5 Â± 66.2 b 17.1 Â± 6.4 209.5 Â± 24.3 MEM 0.0004 P:R 2.3 Â± 1.6 a 1.9 Â± 0.6 a 2.4 Â± 1.3 a 2.3 Â± 0.4 a ANOVA 0.16 Density (shoots m 2 ) 917.8 Â± 85.1 a 840.0 Â± 40.0 776.4 Â± 31.8 a 785.8 Â± 40.1 t test 0.1408 Blade Length (cm) 4.8 Â± 0.9 a 15.8 Â± 1.5 1.9 Â± 0.4 b 14.9 Â± 2.8 t test 0.0012 Blade Width (cm) 0.9 Â± 0.1 a 0.9 Â± 0.04 0.5 Â± 0.1 b 1.0 Â± 0.1 t test 0.0004 AG Biomass (g DM m 2 ) 52.1 Â± 16.5 a 259.4 Â± 44.6 8.6 Â± 4.3 b 193.8 Â± 21.5 t test < 0.0001 BG Biomass (g DM m 2 ) 3880.7 Â± 1256.3 a 3899.7 Â± 1166.4 a 3562.4 Â± 1002.6 a 4143.8 Â± 961.7 a ANOVA 0.611 Within a row, values that share a letter superscript are not significantly different. GPP: gross primary production; R E : ecosystem respiration; NEP: net ecosystem production; P:R: production to respiration ratio; AG: abovegrou nd; BG: belowground; MEM: mixed effects model
40 Figure 2 1. A naturally grazed green turtle feeding plot (right) and an adjacent ungrazed area (left) in a Thalassia testudinum seagrass meadow in Little Cayman. Photo by Robert A. Johnson.
41 Figure 2 2. Daily metabolic rates following simulated or natural grazin g. ( a ) Gross primary production. ( b ) Ecosystem respiration. ( c ) Net ecosystem production. Data are means (Â±SD). Open black squares are clipped plots, closed black squares are reference plots, open red circles are naturally grazed areas, and closed red circles are ungrazed areas. Week 0 began on 15 May 2016, and week 1 1 began on 31 July 2016. Vertical dashed line denotes initiation of clipping. Horizontal dashed line in ( c ) denotes metabolic balance (NEP = 0). See text for description of fluctuating clipped plot values.
42 Figure 2 3. Production to respiration ratios . Data are from all plots on all sampling days. Open black squares are clipped plots, closed black squares are reference plots, open red circles are naturally grazed areas, and closed red circles are ungrazed areas. Solid line is the significant linear reg ression (R 2 = 0.66, p < 0.0001). Dashed line is the 1:1 ratio (NEP = 0).
43 Figure 2 4. Metabolic rates as a function of aboveground seagrass biomass. ( a ) Gross primary production. ( b ) Ecosystem respiration. ( c ) Net ecosystem production. Open black squa res are clipped plots, closed black squares are reference plots, open red circles are naturally grazed areas, and closed red circles are ungrazed areas. Solid lines are the significant linear regressions (GPP, R 2 = 0.69, p < 0.0001; R E , R 2 = 0.58, p < 0.00 01; NEP, R 2 = 0.37, p < 0.0001). Horizontal dashed line in ( c ) denotes metabolic balance (NEP = 0).
44 Figure 2 5. Seagrass ecosystem metabolism values compiled from the literature. ( a ) Gross primary production. ( b ) Ecosystem respiration. ( c ) Net ecosyste m production. Data are means (Â±SD). Data from all studies (n = 58) are ranked by NEP. Open circles (this study) from left to right are: naturally grazed areas, clipped plots, reference plots, ungrazed areas (all panels). Horizontal dashed line in ( c ) denot es metabolic balance (NEP = 0).
45 CHAPTER 3 RATES OF SEDIMENT RESUSPENSION AND EROSION FOLLOWING GREEN TURTLE GRAZING IN A SHALLOW CARIBBEAN THALASSIA TESTUDINUM MEADOW Introduction Seagrasses are important sites of carbon sequestration (Duarte et al. 2005) . While plant biomass production through high rates of metabolic carbon capture (Duarte et al. 2010; Johnson et al. 2017) is an important mechanism for accumulating carbon in seagrass meadows, the majority of carbon in these habitats is stored belowground in the sediments (Fourqurean et al. 2012) . Seagrass meadows facilitate the capture of autochthonous and allochthonous sources of organic matter (Agawin and Duarte 2002; Gacia et al. 2002; Kennedy et al. 2010; Greiner et al. 2016) that, once deposited on the sediment surface, may eventually be buried and incorporated into the sedimentary carbon pool. The depos ition of particles from the water column, and subsequent protection of those particles once deposited, are important processes contributing to high carbon sequestration rates in seagrass meadows (Gacia et al. 1999; Duarte et al. 2005) . Se agrass canopies buffer deposited sediments against resuspension back into the water column, where they are more vulnerable to export (Terrados and Duarte 2000) , by reducing water velocity (Fonseca et a l. 1982; Gacia et al. 1999) and attenuating wave energy (Fonseca and Cahalan 1992; Hansen and Reidenbach 2013) . Newly deposited particles are often partially comprised of falling seagrass material (leaf tissue and/or associated epibiota) and/or allochthonous material originating in nearshore habitats (e.g. mangroves) that are high in organic matter and carbon content (Kennedy et al. 2004; Chen et al. 2017) . These loose surficial sediments are an
46 important source of organic carbon for sequestration within the meadow that are vulnerable to resuspension or erosiona l processes prior to burial. While seagrass presence may support a modest increase in particle deposition (Gacia et al. 19 99; Hendriks et al. 2008) , rates of particle resuspension may be three to ten times lower within seagrass meadows compared to unvegetated sediments (Gacia et al. 1999; Gacia and Duarte 2001) . Given globally declining seagrass area (Waycott et al. 2009) , there is concern as to what a loss of these p rotective services provided by meadows will mean for sediment retention and carbon storage in these systems. Grazing by green turtles ( Chelonia mydas ) can significantly reduce the size of the aboveground seagrass canopy (Bjorndal 1980) . Green turtles are prominent m egaherbivores in meadows in many areas (Hei thaus et al. 2014) and consume seagrass throughout their circumglobal range (Bjorndal 1997) . They graze aboveground seagrass biomass by cropping blades near the sediment surface and repeatedly re grazing the same areas (Bjorndal 1980; Ogden 1980) , thereby maintaining a short meadow canopy. Green turtles have been recorded consuming belowground rhizome biomass in addition to aboveground tissues when they reached hyper abundance within a protected area i n Indonesia (Christianen et al. 2014) ; however, to our knowledge, th is grazing behavior has not been reported elsewhere. In the Caribbean, green turtles typically consume the seagrass Thalassia testudinum . This seagrass species grows a robust belowground rhizome mat, and while green turtles consume the aboveground plant ti ssues, the belowground rhizomes are left alive and intact (Bjorndal 1980) . With declining seagrass area (Waycott et al. 2009) , and increasing green turtle abundance (Chaloupka et al. 2008) , more seagrass area will be returned to a naturally
47 grazed state in ecologically important foraging regions such as the Caribbean (Jackson 1997, 2001; Jackson et al. 2001) . Many studies to date on seagrass meadow sediment dynamics have focused on the benefits provided by ungrazed seagrass beds compared to unvegetated areas (Koch 1999; Terrados and Duarte 2000; Gacia et al. 2003; Potouroglou et al. 2017) , but have not given adequate consideration to the effects grazing may have. Extreme cases o f overgrazing and meadow collapse can lead to erosion and loss of seagrass sediments and associated carbon (Christianen et al. 2014; Atwood et al. 2015) , but this is not likely to be the normal effect of grazing. Grazing of the aboveground ca nopy by green turtles may leave organic matter and carbon rich surface sediments potentially vulnerable to resuspension and erosion. Empirical tests of this are lacking, however, and it is not currently known how these processes operate in seagrass meadows under normal green turtle grazing regimes. It is critical to understand how sediment and carbon dynamics in meadows may be affected by increased future grazing. We hypothesized that (1) rates of particle resuspension would be higher in grazed areas where surface sediments are no longer protected by a tall seagrass canopy, and (2) following grazing, sediment elevation would decrease as loose surface sediments become vulnerable to erosion. To evaluate effects of grazing on these processes, we conducted an e xperiment in a shallow Thalassia testudinum dominated seagrass meadow in Little Cayman, Cayman Islands. A large area of this meadow had been actively grazed by juvenile green turtles for longer than a year (personal observation). We measured rates of parti cle deposition and resuspension in this grazed area compared to an adjacent ungrazed area, providing the first measures of particle
48 flux processes in a naturally grazed seagrass meadow. Erosion was measured from changes in sediment elevation in a series of experimentally clipped plots compared to unclipped reference plots to evaluate effects following the initial removal of the seagrass canopy. Coupled with measured organic matter and carbon content of particle fluxes and surface sediments, we evaluated sho rt term effects of grazing on sediment carbon dynamics in the meadow. Methods Site Description This study was conducted in a seagrass meadow within Grape Tree Bay on the Cen tral Caribbean Marine Institute from May to August 2016. Grape Tree Bay normally experiences relatively calm hydrodynamic conditions, but can experience periods of inclement weather and high hydrodynamic conditions during the months of June to November. Tw o tropical storms, Colin and Earl, passed near Little Cayman during our sampling period, bringing high winds and stronger hydrodynamic conditions to our study site. Grape Tree Bay is roughly 80 100 m wide and 1.7 km long, and is bounded to the north by a fringing coral reef and the shore to the south. The substrate of the bay was comprised of carbonate sediments. The meadow was located in an area with an average water depth of 1.0 1.2 m with a tidal fluctuation of 0.2 m. Juvenile green turtles were pres ent within the bay and had established foraging areas of cropped seagrass within the meadows (Fig. 3 1). Green turtle foraging areas are easily identified by the distinctive grazing marks on the seagrass blade tips, and turtles were observed feeding in the se areas daily. Average canopy height was 16.2 cm in the ungrazed area of seagrass and 1.9 cm in the large area naturally grazed by turtles. Thalassia
49 testudinum was the dominant seagrass comprising the meadows, with average shoot densities of 786 shoots m 2 in the ungrazed area and 776 in the grazed area. Syringodium filiforme seagrass was present at lower densities and interspersed among the T. testudinum ; Halodule wrightii seagrass was present in only one small part of the grazed area. Thalassia testudin um accounted for 93% of the aboveground biomass on average (combined grazed and ungrazed areas) in the meadow. Study Design We evaluated sediment dynamics in areas that were newly grazed (experimental clipping to simulate grazing) and those that had been grazed long term (natural grazing). Clipped plots were used to measure the effects that may follow the onset of grazing in an area, such as increased sediment erosion following removal of the protective seagrass canopy. Established feeding areas of green t urtles within the meadows had been grazed for more than one year (personal observation), and in one these areas we measured differences in rates of particle deposition and resuspension compared to an adjacent ungrazed area. Particle deposition and resuspen sion were measured in the large grazed and ungrazed areas, rather than in experimental plots, to avoid potential edge effects that could affect these processes in clipped plots, which were surrounded by ungrazed seagrass. It is not known why particular are as were grazed by the turtles and other areas of the seagrass were left ungrazed (this is an ongoing area of research). Five pairs of 2 x 2 m plots were established in an ungrazed meadow adjacent to a natural green turtle grazing area, so that all sampling areas for this study experienced the same environmental and hydrodynamic conditions (e.g., depth, currents, distance to shore). Within each pair, one plot was randomly selected and experimentally clipped to
50 simulate grazing, and the other served as an unc lipped reference plot. All plots were separated by at least two meters from other plots. Seagrass blades were clipped with scissors (Moran and Bjorndal 2005) . Simulated grazing was initiated in mid May by clipping all blades within clipped plots t o just above the blade sheath junction and collecting the upper portions of the blades for later analyses. Blades were re clipped when new blade growth reached ~5 cm above the blade sheath junction to mimic natural grazing (Willia ms 1988a; Moran and Bjorndal 2005) , resulting in the plots being re clipped every ~14 days (range 12 15 days). Seagrass a nd Sediment Characteristics Seagrass shoot density, leaf morphometry, aboveground biomass, and belowground biomass were measured in all locations (clipped plots, reference plots, naturally grazed area, ungrazed area). Shoot density was measured prior to the onset of clipping, and bi weekly thereafter by counting all shoots within three 25 x 25 cm quadrats (0.0625 m 2 ) randomly place d within each clipped and reference plot. Shoot density in the naturally grazed area was counted from three 25 x 25 cm quadrats Particle deposition and resuspension each deployment. Aboveground seagrass biomass samples were collected prior to the onset of clipping, and thereafter bi weekly in clipped plots and monthly in reference plots by clipping all shoots at the sediment surface within three randomly placed 10 x 10 cm quadrats (0.01 m 2 ). Biomass samples were collected in a similar manner from three quadrats randomly placed within 1 m of each of the four sediment trap trees in the naturally grazed area once in July. Blade length and width were measured in the lab f or 30 random blades from each sample. Blades were gently scraped to remove any epiphytes, rinsed in seawater, dried at 60Â° C
51 to a constant weight, and weighed for aboveground dry biomass. Belowground biomass samples were collected at the beginning (May) an d end (August) of the experiment from all plots and the naturally grazed area with a 7.7 cm diameter PVC corer. Samples in May were collected just outside of plots to avoid destructive sampling within the sampling area, and from the middle of plots at the end of the experiment in August. Samples were rinsed in seawater to remove all sediments, dried at 60Â° C to a constant weight, and weighed for belowground dry biomass. Seagrass characteristics from three reference plots located near sediment trap trees wer e used to represent the ungrazed meadow for comparison to the naturally grazed area. Surface sediment cores were collected at the beginning and end of the experiment to measure sediment dry bulk density (DBD), percent organic matter content (OM), and per cent carbon content (total (C), organic (C org ), and inorganic (C inorg )). Cores were collected using a 60 ml plastic syringe (2.6 cm diameter, 5.3 cm 2 area) with the end cut off inserted to a depth of 5 cm (26.6 cm 3 core volume). To account for natural spatial variability in sediment characteristics that may have been missed using small syringe cores, three cores were collected from the same area (withi n 25 cm of each other) and combined into a single sample (15.9 cm 2 core area, 79.7 cm 3 core volume). Three samples (of three cores each) were randomly collected within each experimental and reference plot (May and August) and from the four locations of sed iment trap trees within the naturally grazed area (August only). Surface sediment characteristics from three reference plots located near sediment trap trees were used to represent the ungrazed meadow for comparison to the naturally grazed area. Surface se diment samples were dried at 60Â° C to a constant weight and weighed for dry mass (DM). Dry
52 bulk density was calculated as DM divided by the sample volume. Samples were homogenized using a mortar and pestle prior to OM and carbon analyses. Organic matter co ntent was measured from a subset of the sample by drying at 105Â° C for 24 hours before combusting at 500Â° C for 4 hours and measuring the difference. Total carbon content of each sample was measured on an elemental analyzer ( Carlo Erba NA1500 CNHS elementa l analyzer ), and C inorg content was measured on a coulometer ( Coulometrics 5014 CO 2 coulometer ) coupled with an AutoMate automated carbonate preparation device ( AutoMateFX.com ) . Organic carbon was calculated as the difference between C inorg and total C. S ediment Erosion Sediment height was measured in all plots prior to clipping. Changes in sediment height were measured weekly thereafter in clipped and reference plots using two different methods to evaluate sediment erosion following the onset of experime ntal clipping. The first method, measuring depth of the unconsolidated surface layer, measured the depth of the loose, upper sediment layer that may have been vulnerable to erosion, and the second method measured changes in the elevation of the sediment su rface. The depth of the unconsolidated surface layer was measured at ten random locations in each plot by inserting a rigid ruler into the surface sediment until resistance was met (Moran and Bjorndal 2005) . To measure fine scale changes in sedimen t surface elevation, we used a sediment elevation table (Cahoon et al. 2002) slightly modified for use by SCUBA divers (Christianen et al. 2013) , referred to in this study as a RAKA bridge (named using author initials). Two permanent PVC poles were inserted into each of the ten plo ts at the start of the experiment to which the RAKA bridge could be attached. The RAKA consisted of a horizontal PVC pole with five vertical notches
53 into which a rigid ruler was placed and distance between the sediment surface and the horizontal RAKA bridg e was measured (Fig. 3 2a). This setup allowed measurement of the same five spots in each plot throughout the experiment to evaluate elevation changes. Particle Deposition a nd Resuspension Rates of short term particle deposition and resuspension were mea sured with sediment traps deployed in the middle of a large area of the meadow naturally grazed by turtles and an adjacent ungrazed area. Sediment traps were not deployed in the 2 x 2 m experimental plots, to avoid potential edge effects that could have af fected deposition and resuspension processes. Sediment trap construction was similar to those used in previous studies measuring short term deposition rates in seagrass meadows (Gacia et al. 1999; Gacia and Duarte 2001) . Individual traps consisted of 15 ml plastic centrifuge tubes (1.5 cm diameter, 11.6 cm height) with an aspect ratio ( height:diameter) of 7.7, above the recommended threshold of 5 to prevent particles from being resuspended out of sediment traps (Blomqvist and Kofoed 1981) . Traps Traps were equally spaced, ~7 cm apart, along each arm. The five arms were attached in a spiral so no traps were situated above other traps to a central pole inserted into the sediment at increasing heigh ts above the sediment surface (Fig. 3 2b). Sediment traps were at 11, 26, 41, 56, and 71 cm above the sediment surface. Sediment traps on the bottom arm (11 cm) were always located within the seagrass canopy in the ungrazed area, and traps at 26 cm were ab ove the canopy. All traps were above canopy height in the naturally grazed area.
54 Measuring total sediment flux at various heights above the sediment surface allows for the discrimination between primary flux (Fp, g DM m 2 d 1 ; downward flux of new particl es in an area) and resuspended flux (Fr, g DM m 2 d 1 ; downward flux of resuspended particles previously deposited in an area) (Pejrup et al. 1996; Gacia and Duarte 2001) . Particles caught in sediment traps represent the total downward flux (Ft, g DM m 2 d 1 ) of particles in the system and incorporate both primary and resuspended particles (Ft = Fp + Fr) (HÃ¥kanson et al. 1989; Gacia et al. 1999) . Resuspended flux (Fr) can be parsed apart from primary flux (Fp) by calculating the rate of Fp as a function of height above the sediment surface (Valeur 1994; Gacia and Duarte 2001) . This method of separating flux components assumes an initial condition of a uniformly distributed particle load throu ghout the water column, and that in the presence of resuspension, a significant gradient will occur approaching the sediment surface (Valeur 1994; Pejrup et al. 1996) . The rate of Fp is calculated from the linear regression of Ft in the up permost traps, which are assumed to not receive resuspended sediments and are thus equal to Fp (HÃ¥kanson et al. 1989; Pejrup et al. 1996) . Resuspended flux is the difference between total flux and the primary flux calculated from the linear regression at a given height (Fr = Ft Fp). This method, and its use in seagrass meadows, has been described in further detail previously (e.g. HÃ¥kanson et al. 1989; Pejrup et al. 1996; Gacia et al. 1999; Gacia and Duarte 2001) . Four sediment trap trees were deployed in each treatment (natu rally grazed, adjacent ungrazed) on three occasions (June, July, August) by SCUBA divers during summer 2016. Sediment traps were uncapped and filled with subsurface water, and the trees were then attached to the central pole in the sediment. Divers took ca re during
55 deployment to not disturb surface sediments that could be resuspended into traps. Sediment trap trees were left in the meadow for 3 4 days, following methods of previous studies (Gacia and Duar te 2001; Gacia et al. 2002, 2003) . Upon collection traps were capped, brought to the surface, and processed in the lab within 48 hours. Sediment trap samples were filtered onto pre combusted, pre weighed 47 mm glass microfiber filters (Whatman, GF/F), dried at 60Â° C to a constant weight, and weighed for dry mass. Organic matter content was measured from one of the three filters from each of the five heights from all sediment trap trees and deployments. Filters were dried at 105Â° C for 24 hours, weighed, and combusted at 500Â° C for four hours in a muffle furnace. Organic matter content was calculated as the difference between the combusted and dry weights. Carbon content (total C, C org , C inorg ) was measured from one of the three filters from each of the f ive heights from all sediment trap trees and deployments. Total carbon and inorganic carbon were each measured from half of the filter using the same analytical method as for surface sediments described above. Organic carbon content was calculated as the d ifference between C inorg and total C. Data Analysis In addition to the Fr calculated at discreet sediment trap heights, we calculated Fr at the meadow canopy height to estimate the amount of sediment resuspended out of the protective seagrass canopy. Th e relationships between Ft and height measured with sediment traps in this study fit a power law model, which we used to calculate Ft at heights not directly measured with sediment traps. We calculated Fr at the height of the ungrazed meadow canopy by subt racting the calculated Fp (from linear regression) from the calculated Ft at this height.
56 Multiple replicates of a variable (e.g. seagrass shoot density) collected from the same plot were averaged to obtain a single mean value for each plot. Likewise, the dry mass values of the three filtered sediment trap samples from a single arm (height) on a given sediment trap tree were averaged resulting in one value for each of the five heights on each tree for flux calculations. Linear mixed effects models were us ed to evaluate the effect of a treatment (experimental clipping or natural grazing) over time. In these models treatment and time were fixed effects, and blocks (clipped reference plot pairs, for seagrass and surface sediments) or individual sediment trap trees (for sediment fluxes) were treated as a random effect. We used mixed effects models with treatment as a fixed effect, and block as a random effect, to evaluate differences between clipped and reference plots at the end of the experiment (e.g. sedimen t elevation change). Directional changes in sediment height measurements were analyzed using linear regression, but differences between treatments for these measures were analyzed with linear mixed effects models. Sediment dry bulk density, organic matter, and carbon content from surface sediment samples and belowground seagrass biomass were analyzed using two way ANOVA with time and treatment as factors, since these were collected only at the beginning and end of the experiment. Previous studies using sediment traps to measure sediment fluxes have shown resuspension rates to be three to ten times higher over unvegetated areas compared to seagrasses (Gacia et al. 1999; Gacia and Duarte 2001) . Using a more conservative estimate (two times greater resuspension) as an anticipated effect of removing the aboveground seagrass c anopy (i.e. from grazing) and using the ungrazed seagrass
57 resuspension data (from this study) as an assumed initial condition, we estimated effect sizes of 1.2 to 2.3 for grazing in this study. Given these estimated effects, our sampling design allowed suf ficient statistical power (range 0.82 0.99 based on reference treatment conditions) to detect significant differences in sediment resuspension rates from grazing. Analyses were performed in R version 3.4.2 (R Core Team 2018) using the (Wickham et al. 2017) (Pinheiro et al. 2018) (de Mendiburu 2017) packages. Data are presented as mea ns Â± standard deviation (SD). Significance was evaluated at an alpha level of 0.05. Results Seagrass a nd Sediment Characteristics Thalassia testudinum shoot density was higher in clipped plots than reference plots during the 12 week clipping experiment ( F 1,40 = 11.64, p < 0.01). However, density was not different between clipped and reference plots at the end of the experiment ( F 1,4 = 0.02, p = 0.89; Table 3 1). Shoot density was also not different between the naturally grazed area and the adjacent ungraze d area ( F 1,5 = 0.01, p = 0.92). Seagrass canopy height was reduced as a result of clipping in plots. Mean height was 4.8 Â± 0.9 cm in clipped plots and 15.8 Â± 1.5 cm in reference plots during the course of the experiment. This resulted in significantly redu ced post clipping aboveground biomass in clipped plots (52.1 Â± 16.5 g DM m 2 ) compared to reference plots (259.4 Â± 44.6; F 1,34 = 182.78, p < 0.01; Table 3 1). Belowground seagrass biomass was not affected following twelve weeks of experimental clipping, no r was belowground biomass different among any areas (experimental plots, naturally grazed and ungrazed areas), including the area that had been grazed by green turtles for more than a year ( F 2 = 0.53, p = 0.60; Table 3 1).
58 Dry bulk density of the surface sediments (Table 3 2) was not affected by experimental clipping over the course of the study and did not differ among clipped plots, reference plots, the naturally grazed area, and ungrazed area ( F 3 = 0.50, p = 0.69). Organic carbon content of the sedimen t was low in this system, typically < 1%, and below the global median of 1.8% (Fourqurean et al. 2012) . Percent total C, %C org , and %OM of surface sedimen ts did not differ among any areas (experimental plots, naturally grazed and ungrazed areas) or over the course of the study (%C, F 3 = 0.85, p = 0.48; %C org , F 3 = 1.24, p = 0.32; %OM, F 3 = 2.40, p = 0.10; Table 3 2). Sediment Erosion Depth of the unconsolidated surface layer (loose upper layer of sediment) did not differ between clipped and reference plots during the 12 week clipping experiment ( F 1,94 = 0.99, p = 0.32). The depth of the unconsolidated surface layer increased slightly (0.9 cm) over time ( R 2 = 0.46, p < 0.01). However, this occurred simultaneously in both clipped and reference plots (Fig. 3 3a), indicating the increase was not an effect of clipping. Additionally, the depth of the unconsolidated surface layer of sediments in the natura lly grazed area (1.6 Â± 0.2 cm) was similar to clipped plots (1.9 Â± 0.4 cm), even though surface sediments had theoretically been vulnerable to erosion for longer in the grazed area. Clipping also did not lead to an increase in erosion of surface sediment s within plots. There was a slight decrease in elevation (mean 1.8 mm) over the course of the study ( R 2 = 0.47, p < 0.01), but this decrease was observed in both clipped and reference plots and was not an effect of clipping (Fig. 3 3b). There was a signifi cant difference in the change in sediment elevation between clipped and reference plots during the course of the experiment ( F 1,94 = 11.33, p < 0.01); however, the greater
59 erosion actually occurred in the reference plots. This difference in change in sedim ent elevation between clipped and reference plots was only 0.7 mm, and at the end of the experiment there was no longer a difference between treatments ( F 1,4 = 1.47, p = 0.29). Particle Deposition a nd Resuspension Rates of particle deposition (Fp, downwa rd flux of new particles) did not differ between the area naturally grazed long term by green turtles and the adjacent ungrazed area of seagrass during any deployment ( F 1,17 = 1.66, p = 0.22; Fig. 3 4). Fp was higher in July than in June or August for both grazed and ungrazed areas due to higher hydrodynamic conditions caused by Tropical Storm Earl, but did not differ between the two areas (Table 3 3). The linear rate of increase in Fp with depth (i.e. slope of the Fp linear regression) also did not differ between grazed and ungrazed areas at any time ( F 1,17 = 0.22, p = 0.65), indicating that starting concentrations or distributions of sediments within the water column above the meadows were not affected by the presence of grazing. Grazing did not affect the amount of sediments that were resuspended at any time, nor the percentage of the total downward particle flux that these resuspended sediments comprised. Even though traps in the ungrazed area were located within the supposedly protective seagrass cano py, Fr (flux of previously deposited particles, calculated from bottom traps, 11 cm) did not differ between grazed and ungrazed areas during any deployment ( F 1,17 = 0.08, p = 0.78). While mean Fr was high in July, it was not significantly different from ot her months due to high variation among sediment trap trees within treatments (Fig. 3 4). The fraction of the total downward particle flux comprised of resuspended sediments varied between 33.1 59.9% (mean 51%, Table 3 3) and was not affected by grazing ( F 1,17 = 0.05, p = 0.83).
60 Resuspended sediment flux reaching the canopy height in the ungrazed meadow was 11.6 g DM m 2 d 1 in June, 83.1 in July, and 16.3 in August. This represented 52%, 52%, and 51% of the total amount of resuspended sediments measured within the seagrass canopy (bottom traps) in June, July, and August, respectively. At the equivalent height of the ungrazed seagrass canopy, Fr in the grazed area was 22.5 g DM m 2 d 1 in June, 63.2 in July, and 11.4 in August, representing 40%, 55%, and 53% of total Fr in June, July, and August, respectively. Resuspended flux measured above the seagrass canopy (traps at 26 cm) also di d not differ between grazed and ungrazed areas during any deployment ( F 1,17 = 0.11, p = 0.75), and particles reaching this height on average represented 9% and 13% of total Fr in grazed and ungrazed areas, respectively. The presence of a full length canopy therefore did not affect the rate at which the amount of resuspended sediments in the water column decreased with height. Carbon a nd Organic Matter Content o f Sediment Fluxes Natural grazing by green turtles did not affect the total or organic carbon con tent of the downward flux of new particles (Fp; calculated from upper three trap heights) (%C, F 1,17 = 1.07, p = 0.32; %C org , F 1,17 = 0.18, p = 0.67). Total carbon content of particles settling out of the water column was relatively steady over time (mean 11.7%), and the majority was inorganic (Table 3 4). Grazing also did not affect either the total or organic carbon content of the total downward sediment flux (Ft; measured from bottom traps) which is comprised of both primary and resuspended particles (%C , F 1,17 = 0.17, p = 0.68; %C org , F 1,16 = 1.93, p = 0.18). The only aspect of particle flux affected by grazing was %OM. Organic matter content of the primary flux was slightly lower in the grazed area than ungrazed area in
61 July (Table 3 4), but did not di analysis; June, p = 0.99; July, p = 0.05; August, p = 1.0). Likewise, %OM of the total flux was lower in the grazed area, but only in July (June, p = 0.37; July, p = 0.02; August, p = 1.0). Discussion Sediment dynamics did not differ between an area naturally grazed by green turtles and an adjacent ungrazed area in a tropical Thalassia testudinum seagrass meadow in Little Cayman, Cayman Islands. The results of this study are important for understandin g how sediment resuspension and erosion processes are affected by green turtle grazing in shallow, coastal Caribbean seagrass meadows. Both of these processes play a role in the long term accumulation (or loss) of sediment and associated carbon in seagrass ecosystems and are important for understanding total seagrass carbon dynamics. We had predicted increased erosion of surface sediments following the onset of experimental clipping in plots, but clipped plots actually experienced less erosion than referenc e plots during the course of our clipping experiment (Fig. 3 3b). However, this difference in erosion was only 0.7 mm. While the mean 1.8 mm decrease observed in sediment elevation across all plots was statistically significant, the ecological significance is less clear, as this is only the thickness of ~2 sediment grains in coarse calcium carbonate based systems in the Caribbean (Burdige and Zimmerman 2002) . The lack of increased erosion following canopy removal in clipped plots demonstrates that as green turtles increase in abundance and graze more seagrass area, their grazing is not likely to directly result in erosion and loss of sediments from shallow, c oastal meadows.
62 As seagrass meadows are known to provide significant buffering capacity against resuspension (Ward et al. 1984) , presumably a result of their aboveground canopies, we had expected higher sediment resuspension in the area of green turtle grazing where the canopy had been largely removed. However, neither rates of resuspension nor particle deposition in these meadows were affected by grazing. Given that ungrazed seagrass beds offer modest, if any, increases in particle deposition rates compared to unvegetated areas (Ga cia et al. 1999; Gacia and Duarte 2001) , it is not surprising that depositional rates did not differ between grazed and ungrazed areas. Lack of a difference in resuspension between grazed and ungrazed areas during periods of both calm and higher hydrody namic conditions however, indicates that a tall seagrass canopy did not provide more buffering capacity than the minimal seagrass canopy in the grazed area during this study. The empirical data we have generated from our quantitative evaluation of particle fluxes in grazed and ungrazed meadows are relevant to modelling and predicting sediment flux under the conditions in our study. The meadows in this study were located at relatively shallow depths compared to those in other studies measuring effects of se agrass on particle resuspension. If resuspended sediments were distributed throughout the water column (i.e. reaching top traps), our resuspension measurements could have been affected. However, this did not occur based on the measured resuspension gradien ts, and the linearity in total downward flux among the top three trap heights (implying no supplementation to primary flux) for all deployments and treatments (Fig. 3 4). In addition, even if resuspended sediments had been distributed throughout the water column, our conclusion that grazing did not affect deposition/resuspension dynamics would not be
63 affected, because the relationship between total flux and height above the sediment did not differ between grazed and ungrazed areas at any time. Patterns of sediment carbon flux were similar to those of total sediment flux (Fig. 3 5). Total carbon content (%C) of the downward sediment flux exhibited little variation with height above the sediment, and total C flux dynamics were therefore largely controlled by sediment fluxes in this system. Organic carbon content (%C org ) of the total flux was more variable than %C with height but did not display any consistent patterns. That neither primary nor total sediment flux carbon content differed between grazed and ungr azed areas indicates that grazing did not affect resuspended sediment carbon flux not sur prising that grazing did not affect the carbon content of resuspended particles. The seagrass meadows in this study were bounded by a fringing coral reef and sand beach, and likely did not receive high inputs of allochthonous particulate organic carbon. La ck of organic rich inputs, and the dominance of inorganic carbon of the sedimentary pool, may help explain the high similarity between carbon and sediment flux dynamics in this system. Seagrass meadows in other areas (e.g. adjacent to mangroves) with highe r organic content of the primary flux material (Chen et al. 2017) may exhibit different relationships between C org flux and height above the sediment than those observe d here. Not all meadows are the same, and sediment and carbon flux dynamics may operate differently in seagrass systems with different characteristics such as stronger hydrodynamic conditions or finer sediments than those in the shallow meadows of
64 Little Cayman. Total sediment fluxes from our study fall within the range of those reported for Mediterranean Posidonia oceanica meadows. However, resuspension was always higher in the Mediterranean (70 85% of total flux) (Dauby et al. 1995; Gacia et al. 1999; Gacia and Duarte 2001) , even though meadows in Little Cayman were considerably shallower. In a Tanzanian seagrass meadow ( Thalassia he mprichii ), erosion of the surface sediments occurred following a clipping experiment in which 100% of aboveground biomass was removed to mimic urchin grazing (Dahl et al. 2016) . However, trampling occurred in experimental plots during clipping in that study, which may have been responsible for some of the erosion (Dahl et al. 2016) . Water depth (10 cm at low tide), tidal fluctuation, seagrass species, and sediment characteristics in Tanzania all differed from those in the present study, highlighting the importance of studying these processes across systems. Future investigations of sediment processes in grazed meadows in different geographic regions, with different seagrass species assemblages, as well as in deeper waters with high hydrodynamics would be particularl y informative. Most previous studies have focused on ungrazed seagrasses compared to unvegetated areas (Ward et al. 1984; Lawson et al. 2012) . Meadows grazed by green turtles differ from unvegetated areas in several important ways, and it is important to understand how sediment processes operate in these systems as more are returned to a naturally grazed state. While canopy height may be sev erely reduced in grazed areas (e.g. 88% in the present study), it is not gone. Protruding seagrass shoots alter the speed and direction of water flowing over the sediment surface (Fonseca et al. 1982; Hendriks et al. 2008) , and grazed shoots may sti ll at least partially fulfill this role.
65 Clipped seagrass blades decrease shear velocity at the canopy water interface (Gacia et al. 1999) , but the effects of clipping or grazing on hydrodynamics at the sediment surface have not been studied to our kno wledge. Unlike bare areas, most grazed meadows also still have an intact, functioning belowground rhizosphere to stabilize the sediments in addition to a short, aboveground canopy. High densities of green turtles are capable of consuming all aboveground s eagrass biomass within a meadow (Williams 1988a) . In Bermuda, it has been suggested that high turtle abundance has led to overgrazing of the seagrass (Fourqurean et al. 2010) , and in Indonesia, overgrazing by a turtle population led to a collapse of the ecosystem (Christianen et al. 2014) . Green turtles within a marine protected area in Derawan, Indonesia, reached hyper abundance, causing the turtles to begin digging for and consuming belowground seagrass ( Halodule uninervis ) tissues (Christianen et al. 2014) . Removal of the stabilizing rhizomes decreased sedim ent stability and ultimately led to sediment erosion in this system. However, in an earlier experiment in these same meadows in Derawan, prior to the report of digging by turtles (Christianen et al. 2013) , exclusion cages were used to allow seagrass to regrow within areas grazed by turtles. Areas of ungrazed seagrass inside cages did not accumulate any more sediment than grazed areas, indicating that stabilization of sediments was similar in grazed and ungrazed areas (when rhizomes were left intact) (Christianen et al. 2013) . The meadows in Derawan differed in both seagrass species and hydrodynamic conditions from those in our study, lendin g additional support to our conclusion that under normal grazing grazing of aboveground seagrass tissues seagrass meadows can still prevent sediment erosion. The grazing of belowground tissues observed at
66 Derawan is not the usual green turtle grazing behav ior and has not been reported elsewhere to our knowledge. Additionally, Thalassia testudinum seagrass forms a much more robust belowground rhizome mat than the H. uninervis seagrasses consumed in Derawan , making the belowground grazing strategy less likely for green turtles in the Caribbean. The consumption of belowground seagrass biomass is however a common grazing strategy among dugongs (Preen 1995) , and sediment processes where dugongs graze may operate differently than in those where turtles graze normally, such as the current study. Removal of the aboveground seagrass canopy by g reen turtle grazing may have indirect effects within meadows that can affect carbon stored in the sediments, such as altering consumer behavior within grazed areas. Bioturbators can vertically transport organic matter through seagrass sediments, potentiall y bringing buried carbon back to the surface where it may be subject to degradation (Kristensen et al. 2012) . Burrowing organisms transport oxygen deeper into the sediment, potentially stimulating aerobic respiration and remineralization of carbon in deeper layers (Thomson et al. 2018) . Remo val of the canopy may also allow increased predation by consumers that dig for prey (e.g. fishes and rays), thus increasing disturbance of the surface sediments (Yahel et al. 2008) . While small in spatial and temporal scale, and therefore not captured in particle flux dynamics, faunal mediated processes such as these may affect long term sediment a nd carbon dynamics in grazed seagrass meadows differently than ungrazed meadows. Future studies would benefit from including effects of other bioturbators and consumers in addition to grazing when evaluating long term effects of green turtles on seagrass m eadow sediment dynamics.
67 Green turtle grazing did not directly lead to a loss of surface sediments (and associated carbon) from the seagrass meadow in the present study. Other forces (e.g. climatological, anthropogenic) may also affect the long term sedime nt dynamics in meadows in which green turtles are grazing, however. Two tropical storms passed near our study site during this experiment, one of which partially overlapped with a deployment of sediment traps (as seen by higher flux rates in the July measu rement; Fig. 3 4). Rates of sediment resuspension in the area grazed by green turtles were not affected by the higher hydrodynamic conditions created by this storm relative to the ungrazed area. Storms more powerful than those experienced during this study may interact with grazing in ways not captured here, however, and could result in short term events of sediment loss in grazed areas compared to ungrazed areas of seagrass. With climate change increasing the frequency and severity of weather events, and t he expected return of more seagrass area to a natural grazed state with higher green turtle abundances, potential interactions between grazing and environmental processes affecting long term sediment accumulation would be expected to increase. Green turtle grazing (natural and experimental) did not affect the capacity of a shallow Caribbean seagrass meadow to buffer sediments against resuspension and erosion in this study. Similarly, simulated grazing (clipping) did not stimulate remineralization of c arbon in this meadow (Johnson et al. 2017) . These results suggest that sediment carbon stores in shallow, coastal seagrass meadows are not necessarily vulnerable to loss under a normal green turtle grazing regime (consumption of aboveground biomass) in the Caribbe an region. Additional studies in grazed meadows under different conditions (e.g. deeper, open systems) will be particularly useful for
68 furthering our understanding of effects of grazing. Other factors may also affect long term sediment dynamics within mead ows, potentially interacting with grazing to affect carbon storage, and these should be explored further in future studies. Increasing green turtle populations will lead to more seagrass area being returned to a natural grazed state, and it is important to understand what effects, or lack thereof, this may have on these ecosystems and the services they provide so that conservation practices and protections can be applied most effectively.
69 Table 3 1 . Seagrass meadow parameters at the beginning and end of the experiment. n Canopy height Thalassia density AB biomass BG biomass cm shoots m 2 g DM m 2 kg DM m 2 May Grazed area 4 812.0 Â± 91.6 4.1 Â± 0.8 Ungrazed area 3 832.0 Â± 134.6 4.3 Â± 0.8 Clipped plots 5 18.6 Â± 4.4 855.5 Â± 206.4 455.2 Â± 101.0 4.4 Â± 1.3 Reference plots 5 15.4 Â± 5.6 882.1 Â± 187.5 281.6 Â± 82.3 4.4 Â± 0.9 August Grazed area 4 1.9 Â± 0.4 750.7 Â± 108.2 8.6 Â± 4.3 3.0 Â± 0.9 Ungrazed area 3 15.1 Â± 1.9 760.9 Â± 136.2 193.8 Â± 21.5 4.0 Â± 1.3 Clipped plots 5 4.4 Â± 0.8 770.1 Â± 224.2 35.4 Â± 6.6 3.3 Â± 1.0 Reference plots 5 16.1 Â± 2.5 781.9 Â± 100.5 242.3 Â± 36.4 3.4 Â± 1.2 AG: aboveground; BG: belowground. Seagrass canopy height and aboveground biomass were not measured in naturally grazed and ungrazed areas in May. Starting Thalassia density in grazed and ungrazed areas was measured in June. Final aboveground seagrass biomass in grazed and ungrazed areas was collected in July.
70 Table 3 2 . Surface sediment parameters at the beginning and end of the experiment. n DBD C C org OM g DM cm 3 % % % May Clipped plots 5 1.0 Â± 0.1 11.8 Â± 0.1 1.0 Â± 0.4 3.6 Â± 0.6 Reference plots 5 1.0 Â± 0.0 11.9 Â± 0.1 0.9 Â± 0.2 3.8 Â± 0.3 August Grazed area 4 1.1 Â± 0.1 11.8 Â± 0.1 0.7 Â± 0.2 3.3 Â± 0.6 Ungrazed area 3 1.0 Â± 0.0 11.9 Â± 0.1 0.9 Â± 0.1 3.9 Â± 0.2 Clipped plots 5 1.0 Â± 0.1 12.0 Â± 0.1 1.0 Â± 0.2 3.7 Â± 0.3 Reference plots 5 1.0 Â± 0.0 11.9 Â± 0.1 0.9 Â± 0.1 3.9 Â± 0.3 DBD: dry bulk density; C: total carbon; C org : organic carbon; OM: organic matter Surface sediment cores from naturally grazed and ungrazed areas were collected in August only.
71 Table 3 3 . Monthly rates of sediment fluxes and percent resuspension. n Ft Fp Fr Resuspension g DM m 2 d 1 g DM m 2 d 1 g DM m 2 d 1 % June Grazed 4 80.7 Â± 56.9 24.3 Â± 5.2 56.4 Â± 59.2 59.9 Â± 19.2 Ungrazed 4 42.2 Â± 11.7 19.8 Â± 0.6 22.4 Â± 11.8 50.5 Â± 12.2 July Grazed 4 178.0 Â± 83.7 64.0 Â± 7.2 114.0 Â± 86.3 59.1 Â± 15.0 Ungrazed 4 240.4 Â± 127.2 81.1 Â± 4.8 159.3 Â± 129.3 58.8 Â± 19.2 August Grazed 4 58.0 Â± 17.9 36.3 Â± 4.1 21.6 Â± 18.9 33.1 Â± 20.6 Ungrazed 4 66.7 Â± 12.1 34.4 Â± 2.8 32.3 Â± 14.3 46.9 Â± 11.6 Ft: total flux; Fp: primary flux; Fr: resuspended flux Fluxes are measured from bottom sediment traps in naturally grazed and ungrazed areas. Percent resuspension is the proportion of Ft comprised of Fr.
72 Table 3 4 . Carbon and organic matter content of sediment fluxes. n Total sediment flux content Primary sediment flux content %C %C org %OM %C %C org %OM June Grazed 4 12.6 Â± 0.6 1.9 Â± 1.1 2.6 Â± 0.4 11.3 Â± 1.0 2.8 Â± 0.9 1.7 Â± 0.2 Ungrazed 4 12.6 Â± 0.6 2.5 Â± 1.1 3.6 Â± 0.7 10.2 Â± 0.6 2.1 Â± 1.3 1.8 Â± 0.1 July Grazed 4 13.1 Â± 0.3 2.1 Â± 0.1 4.5 Â± 0.6 12.9 Â± 0.3 2.7 Â± 0.7 3.2 Â± 0.4 Ungrazed 4 12.3 Â± 2.6 2.4 Â± 1.7 6.3 Â± 1.0 13.1 Â± 0.2 3.3 Â± 0.3 3.8 Â± 0.3 August Grazed 4 12.6 Â± 0.4 2.4 Â± 1.7 3.1 Â± 0.3 11.4 Â± 0.6 3.1 Â± 1.0 2.1 Â± 0.3 Ungrazed 4 12.8 Â± 0.7 3.6 Â± 1.2 3.2 Â± 0.5 11.5 Â± 0.2 2.7 Â± 1.2 2.0 Â± 0.1 C: total carbon; C org : organic carbon; OM: organic matter Total sediment flux content was measured from bottom sediment traps. Primary sediment flux contents are means of the traps from the top three heights.
73 Figure 3 1 . A border between an area grazed by green turtles (right) and an ungrazed area (left) in a Thalassia testudinum seagrass meadow in Little Cayman. The grazed area is characterized by short, cropped blades, creating a lower meadow canopy height compared to the ungrazed area Photo: R. Johnson.
74 Figure 3 2 . Sediment resuspension and erosion measurement methods. ( a ) Using the RAKA bridge (a modified sediment elevation table) to measure changes in sediment elevation in a reference plot. Photo: A. Gulick. ( b ) A sediment trap tree deployed in an area of the Thalassia testudinum meadow naturally grazed by green turtles . Photo: R. Johnson.
75 Figure 3 3 . Depth of the unconsolidated surface layer (in cm) ( a ) and cumulative change in sediment elevation (in mm) ( b ) over time in clipped plots (open points ) and reference plots ( closed points ). Vertical dashed lines show whe n experimental clipping began, and clipping continued at two week intervals in clipped plots. ( a ) is a measure of the thickness of the loose surface sediments, and ( b ) is a measure of how much sediments have eroded. Values are means Â± SD. Horizontal dashed line in ( b ) is the baseline sediment height prior to clipping. Clipped and reference plots in ( a ) were not significantly different during the experiment. Clipped and reference plots differed during the experiment in ( b ) but were not different at the end.
76 Figure 3 4 . Total downward sediment flux (Ft) profiles measured with sediment traps at various heights in ungrazed ( a c ) and naturally grazed ( d f ) areas of the Thalassia testudinum meadow. Boxes show the interquartile range, the bold line is the medi an, and whiskers extend to the min and max values of Ft. Sloping dotted lines are linear regressions of Fp with height (calculated from top three heights), and the model equations are given in each panel, where h is height in cm. Horizontal dashed lines ar e height of the seagrass canopy at the time of deployment, and hashed lines are Â± SD of seagrass canopy height. Top panels ( a , d ) are June, middle panels ( b , e ) are July, and bottom panels ( c , f ) are August deployments. Sediment fluxes did not differ betwe en grazed and ungrazed areas during any deployment ( p > 0.05).
77 Figure 3 5 . Downward carbon flux profiles measured from sediment traps . Total ( a f ) and organic ( g l ) carbon flux in ungrazed ( a c , g i ) and naturally grazed ( d f , j l ) areas of the Thalassia testudinum meadow. Boxes show the interquartile range, the bold line is the median, and whiskers extend to the min and max values. Horizontal dashed lines are height of the seagrass canopy at the time of deployment, and hashed line s are Â±SD of seagrass canopy height. Values of C org from 71 cm in August were removed due to errors in the inorganic carbon analysis. Fluxes of total and organic carbon did not differ between grazed and ungrazed areas during any deployment ( p > 0.05).
78 CH APTER 4 SEAGRASS ECOSYSTEM METABOLIC CARBON CAPTURE IN RESPONSE TO GREEN TURTLE GRAZING ACROSS CARIBBEAN MEADOWS Introduction Seagrasses form some of the most productive ecosystems on the planet (Duarte and Chiscano 1999) . High rates of metabolic carbon capture, and subsequent biomass production, is one of the main processes by which seagrasses contribute to carbon (Duarte et al. 2010; Kennedy et al. 2010) . Among seagrass mea dows globally, about half of the total carbon stored in a meadow on average may be derived from seagrass biomass (Kennedy et al. 2010) . Meadows also ex port large amounts of biomass annually, contributing to carbon storage in peripheral habitats (Duarte and CebriÃ¡n 1996; Duarte and Krause Jensen 2017) . Coupled wit h the ability to store carbon for centuries to millennia through the creation of a hypoxic sedimentary environment (Mateo et al. 1997; Terrados et al. 1999; Serrano et al. 2012) , highly productive seagrass meadows form efficient natural carbon sinks (Duarte et al. 2013 a) , and their protection has been suggested as a climate change mitigation strategy (Murdiyarso et al. 2015; Macreadie et al. 2017) . There is an estimated 150,000 km 2 of seagrass habitat acr oss the Caribbean Sea, Gulf of Mexico, and The Bahamas (Jackson 1997; Green and Short 2003; Wabnitz et al. 2008) . These meadows provide important foraging habitat for numerous species (Ogden 1976; Hemminga and Duarte 2000; Scott et al. 2018) , including green turtles ( Chelonia mydas ), which were historically abundant in this region prior to overexploitation by humans (Jackson 1997 ; Jackson et al. 2001) . Green turtle abundance is currently increasing in the Caribbean (and elsewhere) (Chaloupka et al. 2008; Mazaris et al. 2017) , which will lead to more seagrass areas being returned to a
79 natural graz ed state (Fig. 4 1 a). Green turtles forage by creating grazing patches within seagrass meadows in which they crop the blades to short heights above the sediment surface, and continually re graze these same areas to consume new tissue growth (Bjorndal 1980; Ogden 1980) . This grazing strategy leads to a reduction of the photosynthetic biomass in a meadow. Given the desire to conserve seagrasses for their ability to sequester and store carbon (Macreadie e t al. 2017) , and the expectation that more seagrass will return to a natural grazed state in the future, it is necessary to understand how grazing affects metabolic carbon capture which affects biomass production and storage across seagrass meadows. Mu ch of the research on seagrass metabolic carbon dynamics to date has focused on differences among species (Murray and Wetzel 1987; Lindeboom and Sande e 1989; Pollard and Moriarty 1991) or between seagrasses and unvegetated sediments (BarrÃ³n et al. 2006; Stutes et al. 2007; Rheuban et al. 2014) , but little attention has been given to grazed meadows. In a Caribbean Thalassia testudinum meadow naturally grazed by green turtles, rat es of metabolic carbon capture were found to be lower compared to ungrazed seagrass, but net ecosystem production remained positive (Johnson et al. 2017) . Long term grazing (greater than a year) did not lead to a proportional increase in benthic respiration and ca rbon remineralization. These results are from a single Caribbean location (Little Cayman, Cayman Islands), which has some of the highest recorded metabolic rates among seagrass meadows (Johnson et al. 2017) . It is not known if grazing will have similar effects acr oss seagrass meadows in other locations or under different conditions. A better understanding of the variability among meadows in response to grazing is needed to understand how increasing green
80 turtle abundance and increased grazing will affect seagrass c arbon capture in the Greater Caribbean region. In this study, we investigate if green turtle grazing has a similar overall effect on metabolic carbon capture across meadows, and if the strength of the response varies among meadows varying in biotic and abi otic characteristics. We measured seagrass ecosystem metabolic rates in four locations around the Greater Caribbean (Petuch 2013; Robertson and Cramer 2014) and Gulf of Mexico in addition to those previously sampled in Little Cayman in wh ich green turtles had established foraging areas. The five locations encompass a wide geographic area and span a range of environmental and seagrass meadow characteristics representative of green turtle foraging areas in this region. The invasive seagrass Halophila stipulacea (Fig. 4 1 b) was present in three sites across two locations, and we also compared metabolic dynamics between meadows of the native T. testudinum and this invasive species. Methods Study Sites We sampled seagrass meadows at seven site s from five locations across the Greater Caribbean and Gulf of Mexico where green turtles had established foraging areas (Fig. 4 2 ). One of these locations, Little Cayman, Cayman Islands, was sampled three times during June and July 2016 as part of a previ ous study ( Johnson et al. 2017) . The remaining four locations were sampled in 2018 St. Croix, U.S. Virgin Islands (February); west coast of Florida (Gulf of Mexico side), USA (May); Bonaire, Caribbean Netherlands (July); Eleuthera, The Bahamas (August). We sampled two different sites at two of these locations Bonaire and Eleuthera and one site at all other locations.
81 Coordinates and general environmental characteristics (daylight hours, water depth, salinity, temperature, and irradiance) for all sites are given in Table 4 1 . The meadows sampled in this study varied greatly in size, from relatively small (<150 m 2 ) for the grazed Thalassia testudinum meadow at the northwest Lac Bay site, Bonaire, to very large (several square kilometers of unbroken seagrass habitat) for the ungrazed T. testudinum meadow at the North Rack site, Florida (west coast). Meadows at other sites encompassed a range of sizes b etween these two, such as ~350 m 2 each for the grazed and ungrazed T. testudinum meadows in Little Cayman, and ~600 m 2 for the grazed and >50,000 m 2 for the ungrazed T. testudinum meadows at the Arvida Bay site in Eleuthera. All sampled grazed meadows were nearly uniformly grazed. The proportion of grazed T. testudinum blades ranged from 95 100% in all grazed meadows. Grazing by green turtles was not observed in ungrazed meadows, except for in St. Croix, where ~3% of T. testudinum blades showed evidence o f grazing (cropped blade tips). No evidence of grazing was observed in any Halophila stipulacea meadows. Some of the sampled seagrass meadows, such as the T. testudinum meadows in Bonaire, were largely monospecific, while others were dominated by one spec ies ( T. testudinum or H. stipulacea ) with interspersed Syringodium filiforme or Halodule wrightii seagrasses and various macroalgae ( Appendix Table A 1 ). The substrate in all meadows was primarily calcium carbonate sand. Additional meadow characteristics a re provided in Results. Sampling Seagrass Meadow Characteristics Seagrass meadow and environmental characteristics were surveyed at each site at the time metabolic incubations were conducted. Environmental temperature and
82 irradiance were measured during m etabolic incubations at seagrass canopy height (5 minute intervals) at each site with a HOBO Pendant logger (Onset Computer Corporation, Bourne, MA). A water sample was collected at canopy height and salinity was measured with a handheld AgTec Salinity Ref ractometer that was calibrated with freshwater each day before sampling (Agriculture Solutions, Strong, ME) (no sample was collected at the Florida site). Meadow depth was measured at each location with (accuracy Â±10 cm). Seagrass species composition, shoot density, blade morphometry (length, width, and surface area), and aboveground biomass were determined in both the grazed and ungrazed areas of each sampled meadow within one meter of where incubation chambers were placed. Shoot densities were measured using 25 x 25 cm quadrats (0.0625 m 2 area) in T. testudinum meadows. Due to the high densities of H. stipulacea , 10 x 10 cm quadrats (0.01 m 2 area) were used to measure shoot densities in meadows dominate d by this species. Aboveground biomass samples were collected using 10 x 10 cm quadrats (all meadows) by clipping all blades at the sediment surface with scissors. Six replicate quadrats (for both shoot density and biomass) were collected from all meadows at each site, except for the St. Croix (n = 5) and Florida (n = 3) sites (Table 4 2 ). Blade length and width were measured for the dominant species ( T. testudinum or H. stipulacea ) from 30 randomly selected seagrass blades from these biomass samples for ea ch meadow. Blade surface area was calculated as two times the product of blade length and width (for each species). Following measurement, all blades were gently scraped clean of sediments and epiphytes (minor to negligible epiphyte loads at all
83 sites), ri nsed in freshwater, and dried to a constant weight at 60Â° C before weighing for dry mass. Ecosystem Metabolism Measurements Metabolic carbon dynamics of the seagrass ecosystems were measured using benthic incubation chambers, similar to previous studies (BarrÃ³n et al. 2004; Calleja et al. 2006; OlivÃ© et al. 2015; Johnson et al. 2017) . Chambers were constructed by inserting a PVC cylinder (16 cm diameter, 0.02 m 2 area) ~7.5 cm into the sediment, and attaching a flexible, gas tight polyethylene bag with sampling port to the top (Hansen et al. 2000) . The use of flexible, gas tight bags in chamber construction allows the propagation of wave turbulence t environmental conditions. Incubation chamber volume was measured in the lab to be 5.5 6 L. On sampling days, chambers were set up in the meadows between 10:30 11:30, and incubations were run for 2.5 3 hours. This length of incubation was chosen because saturation effects can occur within chambers during longer incubations and have been shown to lead to underestimates of metabolic rates (OlivÃ© et al. 2015) . Metabolic dynamics were measured three times at two week intervals in Little Cayman as part of a previous study and shown to be relatively stable among sampling events (Johnson et al. 2017) . Incubations were therefore conducted once at each additional site in th e present study. We set up three light (clear) and three dark (opaque) chambers in each meadow to measure metabolic rates of the system (n = 3 for all meadows). Light chambers were used to measure rates of net ecosystem production (NEP) and dark chambers were used to measure rates of ecosystem respiration (R E ). Metabolic rates were estimated from changes in dissolved oxygen (DO) concentration inside the chambers between the
84 beginning and end of the incubation period. Three water samples were collected from the chamber, via the sampling port, in 60 ml plastic syringes at the beginning and end of the incubation. Upon collection, syringes were capped with a silicon cap and brought to the surface where DO concentration was measured directly in the syringe with an optical DO probe (Vadeboncoeur 2011) (YSI ProODO, Yellow Springs, Ohio) that was calibrated at the beginning of each sampling day with water saturated air. We measured water column metabolism using BOD bottles at the same t ime as benthic chamber incubations to distinguish metabolic dynamics of the seagrass/benthic ecosystem from those of phytoplankton in the water column. Three clear 300 ml glass BOD bottles and three dark 300 ml glass BOD bottles were used to measure water column production and respiration, respectively. Bottles were filled with water at seagrass canopy height, anchored to the bottom, and incubated under in situ conditions in the seagrass meadow. Water column samples were collected at the beginning of incuba tions in 60 ml syringes at canopy height, and DO concentration was measured in the same manner as samples from incubation chambers. Following the incubation period, bottles were collected and returned to the surface, one sample was collected from each bott le with a syringe, and DO concentration was measured. Hourly metabolic rates were calculated from the difference in DO concentration within light (NEP) and dark (R E ) incubation chambers between the beginning and end of the incubation. Water column metabol ic rates measured from BOD bottles were subtracted from rates measured in chambers. Hourly gross primary production (GPP) was then calculated as the sum of hourly NEP and R E , and daily metabolic rates were calculated by multiplying by the length of the pho toperiod (GPP) or by 24 hours (R E ).
85 Daily NEP is the difference between daily rates of GPP and R E . To account for lower metabolic rates at dawn and dusk (OlivÃ© et al. 2015) , we used a corrected photoperiod to calculate daily rates by multiplying the number of daylight hours at each site by 0.75 (Johnson et al. 2017) . There were a few instances when an increase in DO concentr ation was measured within a dark chamber during an incubation, suggesting an (0.1 mg O 2 L 1 ), and we assumed respiration to be zero during these instances. Measured DO c oncentrations were converted from mg O 2 to mmol O 2 , and then to carbon units (mmol C) assuming photosynthetic and respiratory quotients of one (BarrÃ³n & Duarte, 2009) . Data Analyses Net ecosystem production (NEP) was the metabolic variable of primary interest, as it accounts for both carbon uptake and loss and can be used as a n indicator of whether a system is currently a metabolic carbon sink or source. Therefore, NEP is the metabolic variable for which statistical results are presented. Our aim was to evaluate if green turtle grazing has a consistent effect on metabolic ra tes across seagrass ecosystems that is, to evaluate if at any given site, at any given time, rates of NEP are lower in grazed meadows than in ungrazed meadows. We evaluated differences in rates of NEP between grazed and adjacent ungrazed T. testudinum mead ows from each site with paired t tests. T tests were used because the comparison of interest was the difference between grazed and ungrazed meadows (effect of grazing) within a given site, not differences in NEP among sites. For sites where the invasive se agrass H. stipulacea was present, we used one way ANOVAs to test for significant differences in rates of NEP between H. stipulacea meadows and
86 nearby grazed and ungrazed T. testudinum meadows (St. Croix site and Lac Cai Beach site, Bonaire). For the northw est site in Lac Bay, Bonaire, where only a grazed T. testudinum meadow and invasive H. stipulacea meadow were compared, an unpaired t performed to identify significant comparison s. We tested for differences in seagrass meadow parameters between grazed and ungrazed T. testudinum meadows and H. stipulacea meadows at sites using paired t tests and ANOVAs in the same manner as for NEP. We used linear regression to test for significant relationships between NEP and explanatory variables (e.g. seagrass biomass, environmental temperature). Mean values from each meadow for NEP and explanatory variables were used for linear regressions tests, so there was a single data point for each grazed T. testudinum , ungrazed T. testudinum , or H. stipulacea meadow at each sampling site. All analyses were performed in R version 3.4.3 (R C ore Team 2018) . Data were (Wickham et al. 2017) and post hoc analyses were (de Mendiburu 2017) . All statistical analyses were evaluated at a significance level of 0.05 . Results Seagrass Meadow Characteristics We sampled seven Thalassia testudinum dominated seagrass meadows that were grazed by green turtles (hereafter grazed meadows), six T. testudinum dominated meadows that were not grazed (hereafter ungrazed meadows), and three meadows that were dominated by the invasive seagrass Halophila stipul acea (hereafter H. stipulacea meadows). These meadows were spread across a wide area of the Greater Caribbean and Gulf of Mexico regions and varied in their biotic and abiotic characteristics. Data
87 from the Little Cayman site have been reported previously (Johnson et al. 2017) , and are presented again here for comparison to the other sites. Total seagrass shoot densities varied widely among meadows and locations (Table 4 2 ), though shoot densities were always highest in H. stipulacea meadows. The lowest shoot dens ity observed (237.3 shoots m 2 ) was in a grazed meadow (NW site in Lac Bay, Bonaire), but there was no clear trend for higher or lower seagrass densities between grazed and ungrazed meadows across the surveyed sites. Thalassia testudinum density in Little Cayman was greater in the grazed meadow than ungrazed meadow (t test, p = 0.02), while T. testudinum density at the Lac Cai Beach site in Bonaire was greater in the ungrazed meadow (t test, p = 0.02). Thalassia testudinum density did not differ between gra zed and ungrazed meadows at the remaining sites (t tests, p > 0.05). Shoot density of Halophila stipulacea was not significantly different between any of the three H. stipulacea meadows measured (ANOVA, F 2 = 0.09, p = 0.92). Green turtles, by cropping se agrass blades near the sediment surface and reducing canopy height, significantly reduce the aboveground seagrass biomass of meadows within their grazing areas. Aboveground biomass ranged from 4.8 Â± 2.8 to 41.8 Â± 13.9 g DM m 2 among grazed meadows (Table 4 2 ), and biomass was always lower in grazed meadows than adjacent meadows (ungrazed or H. stipulacea ) ( p < 0.05 for all comparisons). The only exception was at the Lac Cai Beach site in Bonaire, at which mean aboveground biomass in the grazed meadow was no t significantly different from the H. stipulacea p = 0.88). Ungrazed meadows always had greater aboveground biomass (range 51.6 Â± 11.0 to 202.2 Â± 34.8 g DM m 2 )
88 than adjacent meadows ( p < 0.05 for all comparisons), except for in St. Croix, where biomass in the ungrazed meadow was not significantly differe nt from the H. stipulacea p = 0.53). Evidence of green turtle grazing was only observed in T. testudinum dominated meadows in this study, not H. stipulacea meadows, and grazing resulted in a >79% decrease in biomass in grazed comp ared to ungrazed meadows at all sites, except St. Croix (47%; Table 4 3 ). Biomass among the three sampled H. stipulacea meadows ranged from 17.1 Â± 5.9 to 63.0 Â± 24.0 g DM m 2 . The H. stipulacea meadow in St. Croix had significantly greater aboveground biom ass than either of the H. stipulacea meadows sampled in Bonaire (ANOVA, F 2 = 12.02, p < 0.01). Ecosystem Metabolic Rates Rates of net ecosystem production (NEP) differed among meadow types (grazed, ungrazed, H. stipulacea ) across the studied locations ( A ppendix Table A 2 ). Across all sites, NEP ranged from 4.2 Â± 7.0 mmol C m 2 d 1 to 51.7 Â± 11.0 among grazed meadows, and from 52.3 Â± 5.7 mmol C m 2 d 1 to 225.3 Â± 19.1 among ungrazed meadows. The large standard deviations on some of the ungrazed meadow NEP estimates from Little Cayman were a result of one incubation chamber producing rates much different from the other two (though not enough to be an outlier). Across all sites, NEP was consistently lower in grazed meadows than ungrazed meadows (Fig. 4 3 ). Ne t ecosystem production was significantly lower in grazed meadows than ungrazed meadows at most of the individual sites surveyed ( p < 0.05). Although NEP appeared lower in the grazed meadows, rates were not significantly different from those measured in the adjacent ungrazed meadows at the Lac Cai Beach site in Bonaire (t test, p = 0.08) or the Half Sound site in Eleuthera (t test, p = 0.09).
89 Though NEP was significantly lower in grazed than ungrazed meadows at most sites, NEP remained positive in all graz ed meadows, even with considerable reduction of photosynthetic seagrass biomass by grazing (>79% biomass reduction at all sites except St. Croix; Table 4 3 ). NEP ranged from 56 96% lower in grazed meadows than ungrazed meadows (Table 4 3 ). A similar rang e in the difference between grazed and ungrazed meadows was measured for gross primary production (54 92% lower in grazed) and ecosystem respiration (40 96% lower in grazed). Ecosystem metabolic rates of H. stipulacea meadows were between those of gr azed and ungrazed T. testudinum dominated meadows (Fig. 4 4 ), but the difference in NEP between H. stipulacea meadows and adjacent grazed or ungrazed meadows was not significant at any site ( p > 0.05 for all locations). Halophila stipulacea meadow NEP rang ed from 61.2 Â± 22.6 mmol C m 2 d 1 to 98.9 Â± 23.3 ( Appendix Table A 2 ), and did not differ significantly among the three sites at which it was present (ANOVA, F 2 = 1.92, p = 0.23), even though the establishment of H. stipulacea seagrass meadows in St. Croix has been relatively recent (Gulick, unpublished data) compared to those in Bonaire (Willette et al. 2014) . Drivers o f Metabolic Rates Variation in rates of seagrass ecosystem metabolism across grazed and ungrazed meadows appear ed to be driven by some characteristics of the seagrass meadows, but not by environmental factors. Meadow NEP was strongly, positively related to aboveground seagrass biomass (linear regression, R 2 = 0.82, p < 0.01; Fig. 4 5 ). Net ecosystem production was also positively correlated with seagrass canopy height (R 2 = 0.84, p < 0.01) and blade surface area (R 2 = 0.92, p < 0.01). Aboveground biomass is likely the more useful predictor of meadow NEP however, as variation in
90 canopy height and blade surface area a re accounted for within measures of biomass. Net ecosystem production was not related to seagrass shoot density (linear regression, R 2 < 0.01, p = 0.96) across meadows (Fig. 4 5 ), nor were there significant relationships between NEP and environmental tempe rature (R 2 = 0.01, p = 0.70) or irradiance (R 2 < 0.01, p = 1.0) across sites ( Appendix Fig. A 1 ). Discussion Through a comparative study of seagrass meadows both grazed and ungrazed across a wide area encompassing sites in the Greater Caribbean and Gulf o f Mexico, we found that green turtle grazing has a consistent effect on the metabolic carbon capture rates of Thalassia testudinum dominated seagrass meadows in this region. Rates of metabolic carbon uptake were always lower in grazed meadows up to 96% in Little Cayman but remained positive across sites. Regardless of the amount of aboveground seagrass biomass removed by grazing, or other differences in biotic and abiotic factors between sites, primary production remained high enough in grazed meadows to ex ceed daily rates of ecosystem respiration. These findings lend additional support to the hypothesis that though rates of future metabolic carbon capture may be lower in areas grazed by turtles, grazing does not stimulate ecosystem respiration and result in a large metabolic loss of carbon currently stored in these seagrass habitats (Johnson et al. 2017) . Though there was a general consistent response in metabolic rates to grazing across meadows, the strength of this response varied among individual sites. Some sit es, such as Little Cayman, had considerably lower rates of NEP in grazed meadows than ungrazed meadows (mean 91.7%), whereas the response to grazing (difference in NEP between grazed and adjacent ungrazed meadows) was less pronounced at other
91 sites (Table 4 3 ). Variability in the strength of the metabolic response to grazing resulted in the lack of a significant difference between grazed and ungrazed meadows at two sites Lac Cai Beach in Bonaire, and Half Sound in Eleuthera (Fig. 4 3 a). Grazing reduces the amount of aboveground biomass (i.e. amount of photosynthetic material) in a meadow. Given the strong relationship between NEP and seagrass biomass (Fig. 4 5 a), we expected sites for which the difference in NEP was not significant to be those with smaller d ifferences in seagrass biomass between the grazed and ungrazed meadows. This was not observed, however. The Lac Cai Beach and Half Sound sites had greater differences between grazed and ungrazed meadows in both NEP and aboveground seagrass biomass than som e other sites (e.g. St. Croix) where rates of NEP were significantly different between grazed and ungrazed meadows (Table 4 3 ). It is possible that local environmental or seasonal factors not accounted for here contributed to higher variability in metaboli c rates at some sites. In addition, metabolic rates, similar to belowground carbon storage in meadows (Oreska et al. 2017a) , may exhibit within meadow variation (patchiness), leading to higher variability in estimates and helping to explain the lack of a significant difference in NEP at these two sites. High within meadow variability in metabolic rates, if present, may also help explain the large variability measured in the ungrazed meadow in Little Cayman. Environmental temperature and irradiance have been shown to be strong drivers of seagrass meadow productivity (PÃ©rez and Romero 1992; Lee and Dunton 1997; Calleja et al. 2006; Lee et al. 2007) . In our study, however, NEP was not correlated with either environmental temperature or irra diance across meadows (grazed, ungrazed, or all meadows combined; Appendix Fig. A 1 ). Previous studies examining controls on
92 seagrass metabolism have focused on ungrazed seagrasses (e.g. Apostolaki et al., 2010; Gacia et al., 2005) . Meadow metabolic productivity correlates strongly with abov eground seagrass biomass though (Johnson et al., 2017; Fig. 4 5 a) , which can be strongly affected by green turtle grazing (Bjorndal 1980; Williams 1988b; CebriÃ¡n and Duarte 1998; Fourqurean et al. 2010; Christianen et al. 2012) . Aboveground seagrass biomass explained 82% of the variation measured in NEP across grazed and ungrazed T. testudinum meadows in this study. We therefore investigated if the lack of a relationship between NEP and temperature and irradiance was perhaps due to a larger effect of differences in biomass caused by grazing. We scaled metabolic rates to the abovegroun d biomass in each meadow (unit metabolism per unit biomass) and compared them to temperature and irradiance. Accounting for differences in aboveground biomass did not reveal an underlying relationship between NEP and these environmental parameters across m eadows (grazed, ungrazed, or all meadows combined; Appendix Fig. A 2 ). The lack of a relationship between NEP and these environmental parameters may be due in part to the relatively limited range of temperature and irradiance in seagrass meadows grazed by green turtles in the Greater Caribbean. All meadows were located in relatively shallow, warm tropical and subtropical environments receiving high levels of incident sunlight. Metabolic Dynamics o f Halophila stipulacea Compared t o Native Seagrass In additi on to the new metabolic carbon capture estimates we contribute from grazed T. testudinum meadows, we also present here the first estimates of metabolic rates from meadows dominated by Halophila stipulacea seagrass. Having been first reported in the Caribbe an in 2002 (Ruiz and Ballantine 2004) , H. stipulacea has spread rapidly among islands and become invasive in this region (Willette et al. 2014;
93 Christianen et al. 2018) . Halophila stipulacea may also spr ead more quickly within areas of seagrass grazed by green turtles than in ungrazed areas (Christianen et al. 2018) . In Lac Bay, Bonaire, H. stipulacea is invading areas of T. testudinum grazed by green turtles, and the turtles have begun to expand their grazing patches into areas of previously ungrazed T. testudinum (Sm ulders et al. 2017; Christianen et al. 2018) . If this relationship between green turtles, native seagrasses, and invasive seagrass exists in other areas, it may have implications for total ecosystem carbon dynamics and carbon storage as well as other ec osystem functions within these meadows. Our results demonstrate that rates of metabolic carbon capture in H. stipulacea meadows are comparable to the native T. testudinum meadows grazed and ungrazed which they are replacing (Fig. 4 4 ). However, it is not known whether this translates to comparable carbon sequestration or storage in meadows dominated by invasive H. stipulacea . Unlike T. testudinum (van Tussenbroek et al. 2006) , seagrass species within the genus Halophila do not form a deep belowground rhizome mat (Fonseca 1989) . Robust rhizome mats form ed by some seagrass species contribute to higher belowground carbon storage (Christianen et al. 2013) , and the lack o f a similar rhizome mat may result in lower sediment carbon storage in H. stipulacea meadows. Halophila tissues also decompose quickly in sediment (Josselyn et al. 1986) , and this may affect seagrass derived carbon input to the sediments in invaded meadows compared to native species with higher refractory organic matter content (Trevathan Tackett et al. 2017) . Given the comparable metabolic rates between the invasive and native seagrasses, some benefits of seagrass presence may be retained in H. stipulacea meadows, such as local mitigation of ocean acidification (through metabolic
94 buffering of pH) (Unsworth et al. 2012; Hendriks et al. 2014; Camp et al. 2016) . Further study is needed to confirm this however. Future studies on carbon dynamics in H. stipulacea meadows will be needed to form a better understanding of how the continued invasion by this species will affect total ecosystem carbon dynamics in seagrass meadows in the Caribbean. Greater Global Assessment o f Grazed Meadows Needed Ungraze d T. testudinum dominated seagrass meadows from this study exhibited rates of NEP (median 104.0 mmol C m 2 d 1 ) near the upper end of those reported in the literature for seagrass ecosystems, and grazed meadows in this study exhibited rates of NEP (median 38.1) near the median value of NEP for all reported ungrazed seagrass meadows (median 27.1) (Fig. 4 6 ). Measurements of seagrass metabolic carbon capture are relatively geographically limited however, with the majority of measurements mostly from T. testud inum dominated meadows coming from the Greater Caribbean and Gulf of Mexico regions. The Caribbean and Gulf of Mexico are ecologically important grazing regions, but only a single estimate of seagrass metabolism from a grazed meadow existed prior to this s tudy (Johns on et al. 2017) . All meadows in the present study exhibited a consistent response in metabolic carbon capture to green turtle grazing; however, our study was limited to the effects of grazing in tropical and subtropical T. testudinum dominated meadows. It is possible that meadows in other regions of the world, perhaps dominated by different seagrass species and under different environmental conditions, will not exhibit the same response to grazing in terms of metabolic carbon dynamics. We are aware of only one other study i n which this was investigated. In a tropical Thalassia hemprichii meadow (Tanzania), net ecosystem production was reduced as a result of experimental clipping to simulate
95 grazing (Dahl et al. 2016) ; however, only hourly rates of daytime NEP were measured. Without corresponding rates of respiration, we cannot evaluate if NEP remained positive in this T. hemprichii meadow over a diel cycle, and therefore if the response to experimental clipping in Tanzania was similar to the response to grazing measured in our study. Further studies on effects of grazing on sea grass meadow metabolic carbon dynamics would be particularly beneficial in: 1) additional areas where grazer abundance is increasing, and 2) meadows dominated by species other than Thalassia testudinum . Conclusion Given the importance of seagrass meadows as blue carbon ecosystems (Fourqurean et al. 2012) and their role in carbon sequestration and potential climate change mitigation, it is necessary to unde rstand how meadow carbon dynamics are affected by grazing. Our study adds critical information to our understanding of the effects of green turtle grazing on seagrass metabolic carbon dynamics, and how these processes may be affected with increasing green turtle abundance and grazing. We demonstrate that the response in metabolic carbon capture to green turtle grazing is consistent across seagrass ecosystems in the Greater Caribbean and Gulf of Mexico regions an area that supports vast expanses of seagrass (Jackson 1997; Green and Short 2003; Wabnitz et al. 2008) . Even meadows with low seagrass density and biomass (e.g. NW site, Lac Bay, Bonaire; Nort h Rack site, Florida), which could be expected to be net heterotrophic (Duarte et al. 2010) , maintained positive rates of NEP and metabolic carbon capture. This is an ecologically important grazing region for green turtle populations, and given increasing green turtle abundance in the Caribbean (Cha loupka et al. 2008) , it is expected that more seagrass area will return to a natural
96 grazed state. Increased grazing will translate to lower rates of metabolic carbon capture, but not a metabolic release of carbon from seagrass meadows. Total ecosystem carbon dynamics are controlled by numerous processes in seagrass meadows (Mateo et al. 2006) . To fully understand the effects of green turtle grazing on total carbon dynamics and blue carbon storage in seagrass ecosystems, we need to better understand how the carbon content of seagrass tissues (above and belowground) and sediment carbon stocks are also affected by grazing. Future studies on these topics will be beneficial and complementary to our results on metabolic carbon dynamics, which show that an increase in green turtle abundance and grazing pressure is not expected to stimulate respiration and lead to a metabolic release of carbon from these important ecosystems.
97 Table 4 1 . Coordinates of each sampling site and environmental parameters measured in the seagrass meadows on days of metabolic incubations. Site Latitude Longitude Daylight Depth Salinity Temperature Irradiance Decimal degrees hours m Â°C Lux Bonaire Lac Cai Beach 12.104417 68.223183 12.75 1.0 35 30.0 32451.8 NW Lac Bay 12.108546 68.231562 12.75 1.4 36 31.1 43221.7 St. Croix Buck Island 17.784612 64.624566 11.5 4.5 35 27.9 21190.4 Little Cayman Grape Tree Bay 19.696518 80.059652 13 1.0 36 31.9 55184.0 19.696518 80.059652 13 1.0 37 31.7 46573.2 19.696518 80.059652 13 1.0 38 33.6 60071.5 Eleuthera Arvida Bay 24.722297 76.190644 12.5 2.4 39 32.2 26764.8 Half Sound 24.936641 76.15342 12.5 0.6 36 34.4 71721.5 Florida North Rack 28.56476 82.779335 13.75 2.0 29.2 40980.1 Temperature and irradiance data are mean values measured during incubations.
98 Table 4 2 . Seagrass characteristics of grazed and ungrazed Thalassia testudinum meadows and Halophila stipulacea meadows measured at each sampling site. Site Meadow n Seagrass density Canopy height Blade surface area AG biomass shoots m 2 cm cm 2 g DM m 2 Bonaire Lac Cai Beach Grazed 6 741.3 Â± 126.3 4.7 Â± 2.2 6.3 Â± 3.2 27.1 Â± 7.6 Ungrazed 6 954.7 Â± 93.7 9.6 Â± 3.9 21.4 Â± 11.6 130.9 Â± 54.4 H. stipulacea 6 3850.0 Â± 806.8 2.1 Â± 0.6 1.9 Â± 0.9 36.4 Â± 13.3 NW Lac Bay Grazed 6 237.3 Â± 44.6 2.8 Â± 1.4 2.8 Â± 1.4 4.8 Â± 2.8 H. stipulacea 6 3483.3 Â± 549.2 2.0 Â± 0.4 1.8 Â± 0.6 17.1 Â± 5.9 St. Croix Buck Island Grazed 5 1964.8 Â± 226.1 4.0 Â± 2.3 4.4 Â± 2.7 27.4 Â± 10.2 Ungrazed 5 2406.4 Â± 250.1 8.1 Â± 3.1 9.7 Â± 4.5 51.6 Â± 11 H. stipulacea 5 3733.6 Â± 1249.4 2.8 Â± 0.5 3.2 Â± 0.9 63.0 Â± 24.0 Little Cayman Grape Tree Bay Grazed 6 957.3 Â± 148.5 2.0 Â± 0.9 2.3 Â± 1.2 12.3 Â± 3.7 Ungrazed 6 1018.7 Â± 210.2 14.9 Â± 8.5 30.7 Â± 20.4 183.2 Â± 94.8 Eleuthera Arvida Bay Grazed 6 1648.0 Â± 392.7 3.5 Â± 1.9 4.2 Â± 2.2 41.8 Â± 13.9 Ungrazed 6 2053.3 Â± 329.9 15.2 Â± 4.6 25.8 Â± 8.8 202.2 Â± 34.8 Half Sound Grazed 6 1674.7 Â± 316.7 1.8 Â± 0.7 1.4 Â± 0.6 8.8 Â± 2.0 Ungrazed 6 1784.0 Â± 284.7 9.4 Â± 2.0 10.8 Â± 3.1 96.5 Â± 16.2 Florida North Rack Grazed 3 1322.7 Â± 444.1 2.5 Â± 1.1 1.6 Â± 0.8 9.3 Â± 2.4 Ungrazed 3 1386.7 Â± 161.9 8.9 Â± 2.7 7.8 Â± 2.9 70.7 Â± 23.3 n is number of replicates at each site. Values are means Â± SD. Canopy height is measured from mean blade length. Blade surface area is per blade of seagrass. AG: aboveground; DM: dry mass
99 Table 4 3 . Percent difference in metabolic rates and aboveground (AG) seagrass biomass between grazed and adjacent ungrazed Thalassia testudinum meadows. Site GPP R E NEP AG Biomass % difference Bonaire Lac Cai Beach 81.8 90.9 65.6 79.3 St. Croix Buck Island 60.4 78.3 56.0 46.9 Little Cayman Grape Tree Bay 92.2 95.5 89.1 93.3 89.9 79.6 96.0 93.3 88.5 87.0 90.0 93.3 Eleuthera Arvida Bay 54.3 39.5 62.9 79.3 Half Sound 80.1 88.9 67.9 90.8 Florida North Rack 61.5 41.3 91.9 86.8 GPP: gross primary production; R E : ecosystem respiration; NEP: net ecosystem production
100 Figure 4 1 . A Thalassia testudinum seagrass meadow grazed by green turtles (panel a , left side) and an adjacent ungrazed meadow (panel a , right side) in Eleuthera, The Bahamas ( a ). The invasive seagrass Halophila stipulacea in St. Croix, U.S. Virgin Islands ( b ). Photos: R. Johnson.
101 Figure 4 2 . Map of seagrass meadow sampling locations. Two sites were sampled in both Bonaire and Eleuthera, and one site was sampled at each of the other three locations.
102 Figure 4 3 . Metabolic r ates (mean Â± SD) in grazed (open points) and adjacent ungrazed (solid points) Thalassia testudinum seagrass meadows. N et ecosystem production ( a ), gross primary production ( b ), and ecosystem respiration ( c ) . Rates of NEP were significantly lower in grazed meadows than ungrazed meadows at all sites except at Lac Cai Beach, Bonaire, and Half Sound, Eleuth era ( Results ). Sites ordered south to north. BON: Bonaire (LCB: Lac Cai Beach site); STX: St. Croix; LC: Little Cayman; EL: Eleuthera (AB: Arvida Bay site; HS: Half Sound site); FL: Florida. Dashed line in ( a ) denotes metabolic balance (NEP = 0). Values ab ove line represent net metabolic carbon capture.
103 Figure 4 4 . Rates (mean Â± SD) of net ecosystem production in meadows dominated by the invasive seagrass Halophila stipulacea (triangles) compared to nearby grazed (open circles) and ungrazed (solid circles) Thalassia testudinum meadows. BON: Bonaire (LCB: Lac Cai Beach site; NW: northwest Lac Bay site); STX: St. Croix.
104 Figure 4 5 . Relationship between net ecosystem produc tion and aboveground seagrass biomass ( a ) and total seagrass shoot density ( b ) from grazed (open points) and ungrazed (solid points) Thalassia testudinum meadows. Solid line in ( a ) is the significant linear regression (R 2 = 0.82, p < 0.01) between NEP and biomass. NEP was not related to seagrass shoot density across meadows.
105 Figure 4 6 . Rates (mean Â± SD) of net ecosystem production from meadows of various seagrass species collected from the literature. Open circles, closed bla ck circles, and open triangles are grazed Thalassia testudinum (n = 7), ungrazed T. testudinum (n = 6), and Halophila stipulacea (n = 3) meadows, respectively, from the present study. All measured grazed meadows maintained positive rates of NEP. Colored ci rcles are previously published estimates of seagrass meadow NEP. Published estimates come from various locations around the world, and not all estimates were measured using the same method. Figure modified from Johnson et al. (Johnson et al. 2017) to include addit ional estimates. Criteria for literature collection and data inclusion are described in Johnson et al. (2017) . Letters above points denote sites from the present study B: Bonaire; C: Little Cayman; E: Eleuthera; F: Florida; S: St. Croix.
106 CHAPTER 5 SIMULATED GREEN TURTLE GRAZING AFFEC TS BENTHIC INFAUNA ABUNDANCE AND COMMUNITY COMPOSITION BUT NOT DIVERSITY IN A THALASSIA TESTUDINUM SEAGRASS MEADOW Introduction Seagrass ecosystems host abundant and diverse invertebrate infaunal communities (animals inhabiting the benthic environment) (Orth et al. 1984; Hemminga and Duarte 2000) . Different infaunal species contribute to many important roles in the functioning of seagrass meadows. Polychaetes act as pollinators for Thalassia testudinum seagrass in the Caribbean, suggesting that infauna play a role in the health and longevity of se agrass meadows (van Tussenbro ek et al. 2016) . Detritivorous species of infauna consume organic matter at the sediment surface and are important for recycling nutrients and organic matter within meadows (Klumpp et al. 1989; Edgar and Shaw 1995; Hemminga and Duarte 2000) . The infaunal community can also serve as a large source of prey for consumers, including small fishes, within meadows (Virnstein 1977; Hemminga and Duarte 2000) , thereby transferring energy from the bacterial and detrital pathways up the meadow food web to higher order consumers (Edgar and Shaw 1995) . Seagrass meadows host significantly higher densities of infauna compared to unvegetated sediments (Orth et al. 1984) , with higher infauna densities found in meadows with greater seagrass biomass (Stoner 1980; Stoner and Lewis 1985; Skilleter et al. 2007) . Structural complexity of the seagrass canopy is also important, with canopies of higher complexity hosting greater infaunal diversity and abundance (Gartner et al. 2013) . The role that infauna play in the cycling of organic matter within seagrass mea dows is likely to make the infaunal community susceptible to changes in organic
107 matter input to the benthic environment. By removing seagrass and algal biomass, grazers decrease the amount of primary production within a meadow available for incorporation i nto the benthic detrital compartment. While many small grazers, such as fishes, may not remove large amounts of biomass (CebriÃ¡n and Duarte 1998; Hemminga and Duarte 2000) , megagrazers, such as dugongs ( Du gong dugon ) and green turtles ( Chelonia mydas ), consume large amounts of seagrass (Thayer et al. 1984; Williams 1988a; Sco tt et al. 2018) and are likely to have a greater effect on the amount of seagrass and algal material entering the detrital pathway. Dugongs create and belowground seagrass plant material (Preen 1995) . The large physical disturbance to the benthic environment caused by this feeding strategy decreases the abundance of i nfauna within these trails relative to the surrounding seagrass habitat (Skilleter et al. 2007) . Green turtles exhibit a different foraging strategy from that of dugongs, however. Under normal grazing conditions, green turtles only consume the aboveground portions of seagrass, cropping blades to a short height above the sediment surface (Bjorndal 1980) . Green ay remain in a grazed state for months or years (Bjorndal 1980; Fourqurean et al. 2010; HernÃ¡ndez and van Tussenbroek 2014) . By cropping only aboveground seagrass blades, green turtles alter the amount of seagrass biomass in a meadow, but do not cause the same direc t physical disturbance to the benthic environment as in the feeding trails of dugongs. Green turtle abundance is increasing in many areas (Mazaris et al. 2017) , including the Caribbean (Chaloupka et al. 2008) . Greater numbers of green turtles will
108 lead to higher grazing pressure and more seagrass area being returned to a natural grazed state. Effects of green turtle grazing on the infaunal communities of seagrass meadows are not well understood, however. Changes to the seagrass infaunal community are likely to have effects not only on the amount o f infauna available as a source of prey for consumers, but also on detrital matter consumption and the recycling of nutrients and organic matter within meadows. It is therefore critical to understand what effects grazing has on infaunal communities in orde r to better understand how seagrass meadow functioning may be affected by increasing green turtle abundance and grazing. Given that meadow faunal abundance is positively related to both seagrass biomass and complexity of the seagrass canopy (Stoner 1980; Gartner et al. 2013) , both of which are decreased by grazing, we hypothesized that green turtle grazing would lead to a decrease in infauna abundance. We further hypothesized that grazing would decrease infaunal diversity and lead to differences in the composition of the infaunal community within grazed areas, simil ar to changes observed following dugong grazing (Skilleter et al. 2007) . To investigate our hypotheses, we analyzed infauna collections made during a long term clipping experiment (16 months) in a Thalassia testudinum seagrass meadow in The Bahamas to simulate green turtle grazing (Moran and Bjorndal 2005, 2007) . We characterized the infaunal community and aspects of the seagrass meadow in both experimentally clipped and unclipped reference plots prior to clipping initiat ion and throughout the 16 month simulated grazing experiment to evaluate changes in abundance and community composition.
109 Methods Site Description a nd Experimental Design Marine Research Center on Lee Stocking Island, Exuma, The Bahamas (23.772963, 76.106910) from July 1999 to November 2000. The experiment was conducted in a large, monospecific Tha lassia testudinum seagrass meadow 400 m to the southwest of the island. Water depth was ~3 m at low tide with a tidal range of ~1.5 m. Mean shoot density was moderate within the meadow (589 Â± 16.5 (SE) shoots m 2 ), and mean canopy height of the meadow was 12.4 Â± 0.5 (SE) cm at the beginning of the study. The substrate was composed of sandy carbonate sediments within the study area. Thirty 3 x 3 m plots were established in the T. testudinum seagrass meadow to the southwest of Lee Stocking Island in a blocke d design. Three blocks of ten plots each were established roughly 50 m apart, with each block consisting of five plot pairs. Each pair of plots consisted of one experimentally clipped plot to simulate green turtle grazing, and one plot that was left unclip ped to serve as an ungrazed reference. All plots within blocks were spaced at least 4 m apart. Simulated grazing was initiated in clipped plots by clipping all seagrass blades at the blade sheath junction (~2 cm above the sediment surface) with stainless s teel scissors to mimic natural green turtle grazing (Bjorndal 1980; Moran and Bjorndal 2005) . All blades within experimentally clipped plots were re clipped every time mean blade length in the plot reached 5 cm, consistent with natural green turtle grazing behavio r in the Caribbean region (Bjorndal 1980; Ogden 1980) . Intervals between clipping events varied between 12 and 37 days, as blade growth varies with temperature (Moran and Bjorndal 2005) , and clipped plots were maintained for the entire 16 m onth duration of the study. Seagrass blades were
110 removed from plots at the time of clipping to simulate the removal of biomass and nutrients when green turtles graze an area. Rhizomes were severed around the perimeters of clipped plots at the beginning of the experiment and every 6 8 weeks thereafter with a flat bladed shovel to prevent nutrient translocation into the experimental plots. A 0.5 m wide buffer zone was established around the inner border of all plots, and all samples were collected from the remaining inner 2 x 2 m area, to avoid possible edge effects created by the surrounding unclipped meadow. Infaunal Sample Collection a nd Analysis The invertebrate infaunal community was sampled from all plots, clipped and reference, prior to initiation o f the clipping experiment (July 1999; 0 months), and at 2, 6, 11, and 16 months following the initiation of experimental clipping. Infauna communities were sampled from sediment cores collected within each plot with a 7.62 cm I.D. PVC corer inserted to a d epth of 25 cm (1140 cm 3 total sample volume). One core was collected near the middle of each plot at each sampling time, taking care to not resample the same locations as previous coring events. Cores were washed with seawater over a series of two metal mesh sieves (2.0 and 0.5 mm) in the lab, and seagrass root and rhizome material was removed. Only material and organisms >0.5 mm were retained on the sieves, with smaller particles and organisms being washed through. The material and organisms retained on each sieve were transferred to jars and preserved with a mixture of 5% buffered formalin and seawater. Samples were left in this buffered formalin mixture for at least two weeks, during which time a few drops of Rose Bengal mixed with ethanol were added to each jar as a staining agent. After this period of preservation and staining, samples were rinsed of the buffered formalin mixture and transferred to 70% ethanol in glass
111 scintillation vials for storage. The >2.0 mm and 0.5 2.0 mm size fractions were ke pt separate during this process. Infauna data were analyzed separately for each of these size fractions, in addition to analyses for all data pooled together. Infaunal organisms were counted and identified under a dissecting microscope. Most organisms we re identified to class or order. Aschelminth phyla were split into two groups during identification: the nematodes (phylum: Nematoda) and all other communities were enumerated for t he same eight clipped plots and eight reference plots at each sampling time point during the experiment. Seagrass a nd Sediment Sample Collection a nd Analyses Seagrass meadow characteristics Thalassia testudinum shoot density, blade length and width, abov eground seagrass biomass, and belowground seagrass biomass were measured at regular intervals throughout the study in all experimentally clipped and reference plots. Seagrass structural parameters were measured bi weekly in all plots. Shoot density was mea sured from three randomly placed 25 x 25 cm quadrats (0.0625 m 2 ), and blade length and width were measured from 30 randomly selected blades from each plot. Aboveground seagrass biomass was measured in all plots, clipped and reference, prior to the initiati on of clipping (July 1999; 0 months). Thereafter, aboveground biomass was measured in clipped plots at each clipping event (when blade length reached 5 cm), and in reference plots at 2, 6, 11, and 16 months following initiation of the experiment. In clippe d plots, all blades from the inner 4 m 2 of each plot were collected for biomass, and in reference plots all blades from three randomly placed 25 x 25 cm quadrats were collected. Collected blades were rinsed in seawater in the lab to remove sediments and dr ied to a constant weight at 60Â° C for dry
112 mass. Belowground seagrass biomass (roots and rhizomes) was collected prior to the initiation of clipping (0 months) and at 2, 6, 11, and 16 months following clipping from the same cores as infauna samples. Roots a nd rhizomes were separated from the surrounding sediment in the lab (at the time infauna samples were collected on sieves), rinsed in seawater, and dried to a constant weight at 60Â° C for dry mass. Characteristics of the surface sediments (where most inf auna reside) in the meadow were measured in all plots throughout the experiment. The depth of the detrital layer (layer of loose particles and detritus at the sediment surface) was measured bi weekly by inserting a rigid ruler into the sediment until resis tance was met within each of the three randomly placed quadrats used for measuring seagrass shoot density. Organic matter content and particle size distribution in the surface sediments of the meadow were measured from three shallow sediment cores (5.08 cm I.D.; 15 cm depth; 304 cm 3 volume) collected in each plot prior to clipping initiation (0 months), and at 2, 6, 11, and 16 months. Seagrass material and large invertebrates that were visible were removed in the lab. Sediment samples were then dried to a c onstant weight at 60Â° C, ground with a mortar and pestle to pass through a 1 mm mesh sieve, and re dried at 105Â° C for at least 16 hours for dry mass measurement. Organic matter content was measured by combusting dried sediment samples at 500Â° C for 3 hour s in a muffle furnace. Sediment particle size was measured using the hydrometer technique (Buoyoucus 1936; Gee and Bauder 1986) . Particle size classes were assigned as follows: sand, >0.05 mm; sil t, 0.002 0.05 mm; clay, <0.002 mm. Data Analyses For variables collected more frequently than infauna samples from plots (e.g. seagrass density, collected bi weekly), only data corresponding to the times of infauna
113 collection were used for analyses (i.e . sampling times at 0, 2, 6, 11, and 16 months). Mean values were calculated for variables for which multiple measurements were collected at each time point (e.g. seagrass density, three quadrats from each plot), and these mean values (i.e. single value pe r plot per sampling time) were used for analyses. Effects of clipping on infauna abundance over time were analyzed using linear mixed effects models, where treatment and time (sampling event) were treated as fixed effects, and experimental plot blocks wer e treated as a random effect. Effects of clipping on meadow and environmental factors over time (e.g. aboveground seagrass biomass, depth of detrital layer) were analyzed using linear mixed effects models in the same manner. Differences in variables (infau na abundance, seagrass and sediment parameters) between clipped and reference plots at the end of the 16 month clipping experiment were analyzed with a linear mixed effects model with treatment as a fixed effect and experimental plot block as a random effe ct. Relationships between infauna abundance and meadow variables (e.g. aboveground seagrass biomass) were analyzed using linear regression, and infauna abundance data were square root transformed to meet normality assumptions. s used to measure the diversity of the infaunal community. This index accounts for both group richness and abundance and is a good measure of diversity in situations when one or a few groups are considerably more abundant than others (e.g. nematodes in thi s study). A linear mixed effects model was used to analyze effects of clipping over time on the diversity of the infauna community, where treatment and time were treated as fixed effects, and block was treated as a random effect. Differences in measures of infaunal diversity between treatments at the
114 end of the experiment were analyzed with a linear mixed effects model with treatment infaunal group richness and evennes s were also evaluated across treatments and sampling events. Changes to the composition of the infauna community over time as a result of experimental clipping were analyzed using a PERMANOVA (Anderson 2001) with treatment and time as factors. Differen ces in the community composition between treatments (clipped and reference plots) at the beginning (0 months) and end of the experiment (16 months) were analyzed using a PERMANOVA with treatment as the only factor. PERMANOVA calculations were based on a Br ay Curtis distance matrix for community composition. All analyses were performed in R version 3.5.1 (R Core Team 2018) using the effects models (Pinheiro et al. 2018) for PERMANOVA tests (Oksanen et al. 2018) . Significance for all tests was evaluated at an alpha value of 0.05. Results Most of the infaunal organisms identified in this study were from the 0.5 2.0 mm size fraction, and results for infauna abundance from this size fraction were t he same as results from all data when both size fractions were pooled. There were not enough data for multiple groups in the >2.0 mm size fraction to evaluate infauna abundance, and abundance was low for remaining groups. Because including or omitting data from the >2.0 mm size fraction from analyses did not affect results, only results from the pooled data are presented here.
115 Effects o f Simulated Grazing o n t he Infaunal Community Effects of experimental clipping on the seagrass meadow infaunal community occurred relatively quickly within six months following the onset of simulated grazing. Between two and six months post clipping, total infauna abundance decreased by 59% within clipped plots (Fig. 5 1) and became significantly different over time from that in unclipped reference plots (linear mixed effects model (LME); F 1,41 = 36.7, p < 0.01). The decrease in total infauna abundance was driven by a decrease in abundance observed across all individual infaunal groups within clipped plots between t wo and six months post clipping (Fig. 5 2). The decline in abundance between two and six months was greater in some groups (e.g. copepods) than others (e.g. oligochaetes), and the differences between clipped and reference plots were not significant over time for all groups (Table 5 1). The abundance of crab claws was also counted (a potential sign of predation), and there was a significant increase in crab claws between two and six months within clipped plots (Table 5 1, Fig. 5 2). Nematodes were the mos t abundant individual infaunal group and comprised on average ~30 40% of total infauna abundance within the meadow. As a result, total infauna abundance was strongly, positively related to nematode abundance (linear regression; R 2 = 0.69, p < 0.01), and the decrease in total infauna abundance was likely largely driven by the decrease in nematode abundance. than reference plots (0.74) (LME; F 1,13 = 4.6, p = 0.05) prior to clipping initiation; during the experiment (LME; F 1,55 = 1.4, p = 0.25) (Fig. 5 infaunal group richness and evenness did not differ between treatments prior to clipping
116 (p = 0.39 and p = 0.47, respectively). Following the onset of clipping, infaunal group richness decreased in clipped plots, while group evenness increased, resulting in the time between clipped and reference plots. Though the diversity of the infaunal community was not affected by clipping, the relative abundance of individual groups comprising the community varied during the experiment (Fig. 5 4). As a result, the composi tion of the infaunal community became significantly different between clipped and reference plots during the experiment (PERMANOVA; full model, Table 5 2). Community composition did not differ between treatments prior to clipping (0 months, Table 5 2). Ef fects o f Long Term Simulated Grazing o n t he Infaunal Community Following 16 months of experimental clipping, total infauna abundance remained significantly lower in clipped plots compared to reference plots (LME; F 1,13 = 9.6, p = 0.01; Table 5 3). However , differences in abundance were no longer present at the end of 16 months for every individual infaunal groups that had experienced decreases within clipped plots during the experiment. By the end of the clipping experiment, only nematodes (p < 0.01) and p eracarids (p = 0.01) still had significantly lower abundances within clipped plots compared to reference plots (in addition to total infauna abundance) (Table 5 3). The abundances of the remaining groups that had been lower in clipped plots became similar to those in reference plots, whether due to an increase in abundance in clipped plots (e.g. polychaetes), or a decrease in abundance within reference plots (e.g. gastropods) (Fig. 5 2). e plots at the end of 16 months (LME; F 1,13 = 0.4, p = 0.56). Though infaunal group richness and
117 evenness exhibited temporal variation within clipped plots, by the end of the 16 month experiment both richness and evenness had become similar between clipped and reference plots again (richness; F 1,13 = 0.3, p = 0.59; evenness; F 1,13 = 0.3, p = 0.60). All infaunal groups were present at all sampling times, though abundances were temporally variable. This variation led to shifting relative dominance among indiv idual groups, which affected the community composition. The significant differences in infaunal community composition observed between clipped and reference plots during the experiment (above section) were still present at the end of the experiment (PERMAN OVA; 16 months, Table 5 2). Effects o f Simulated Grazing o n Meadow Characteristics Results from this experiment on the effects of simulated grazing (clipping) on the growth and morphometry of seagrass have been described previously (Moran and Bjorndal 2005) . Some results are presented again here to aid in the interpretation of th e results of experimental grazing on the invertebrate infauna community. Experimental clipping significantly altered the aboveground seagrass canopy compared to reference plots that remained unclipped (Table 5 4). Canopy height (measured from blade length ) was significantly lower in clipped plots than reference plots throughout the experiment (LME; F 1,55 = 537.7, p < 0.01). Seagrass blade width in clipped plots also became significantly narrower than within reference plots by the end of the experiment (16 months; LME; F 1,13 = 35.9, p < 0.01). Aboveground seagrass biomass was of course also lower in clipped plots than reference plots during the experiment (LME; F 1,55 = 322.8, p < 0.01), as the clipping regime mimicked the natural grazing strategy of green tu rtles cropping all blades to short heights above the sediment surface and removing the majority of biomass. Seagrass shoot density did not
118 differ between clipped and reference plots following 16 months of clipping (LME; F 1,13 = 0.9, p = 0.35). Experimenta l clipping of the seagrass canopy also affected the sedimentary environment (Table 5 5). The depth of the detrital layer loose surface layer of unconsolidated sediment and organic matter was reduced in clipped plots compared to reference plots over the cou rse of the experiment (LME; F 1,55 = 5.0, p = 0.03). Though the detrital layer became smaller in clipped plots, the organic matter content of the surface sediments was not affected by clipping (LME; F 1,55 = 2.7, p = 0.10). The composition of the sediment (percent sand, silt, and clay) vari ed significantly over the course of the experiment (significant time component) in both treatments but did not differ between clipped and reference treatments. Relationships Between Infauna Abundance a nd Meadow Characteristics We investigated potential r elationships between abundance of the infaunal community and characteristics of the seagrass meadow with linear regression. Total infauna abundance was significantly, positively related to aboveground seagrass biomass (p < 0.01; Fig. 5 5a), meadow canopy h eight (p < 0.01), and blade surface area (p < 0.01). Each of these meadow characteristics explained similar amounts of variation in total abundance (R 2 = 0.15, R 2 = 0.18, R 2 = 0.17, respectively), as all three are directly related to each other. Total infa una abundance was not significantly related to belowground seagrass biomass (R 2 = 0.01, p = 0.16). However, belowground biomass did not differ between clipped and reference plots during the experiment (LME; F 1,55 = 0.17, p = 0.68), as did measures of the a boveground canopy. With the exception of organic matter content, infauna abundance was not related to any measured sedimentary characteristics (i.e. detrital layer depth, percent sand, percent silt, percent
119 clay) in the meadow. Total infauna abundance was positively related to the organic matter content of the surface sediments across all plots (R 2 = 0.04, p = 0.04); however, the relationship was weak and differences in organic matter explained little of the variation in abundance (Fig. 5 5b). Discussion T hrough the creation of grazing patches in which seagrass blades are cropped to a uniform, short height, green turtles create areas within meadows that are structurally different from the surrounding areas of ungrazed seagrass. The creation of these areas i n which the aboveground seagrass canopy has been removed affects physical processes at the sediment surface, such as wave attenuation (Fonseca and Cahalan 1992) , and chemical processes, such as metabolic carbon flux (Johnson et al. 2017) . Our results from the present study show that canopy removal also affects the invertebrate infaunal communities inhabiting Thalassia testudinum seagrass meadows. Infaunal communities pl ay important roles in seagrass meadows, and it is important to understand how they may be affected by grazing to better understand the role of green turtles within seagrass ecosystems. Simulated grazing resulted in a decrease in the total abundance of inf auna within six months of clipping initiation, after which total abundance remained reduced throughout the clipping experiment. However, temporal dynamics in abundance varied among individual infaunal groups within the community. More than half of the infa unal groups present in the meadow (6 of 11) decreased in abundance within a relatively short period of time (within six months) following a large structural change to the ecosystem (canopy removal via clipping) (Fig. 5 2; Table 5 1). However, following sus tained long term clipping (16 months), abundances of four of the six groups that
120 initially decreased returned to levels comparable to abundances in unclipped reference plots. This suggests that the decrease in abundance for some infaunal groups may have be en a response to the perturbation to the ecosystem from which the population was able to recover as it adapted to the altered ecosystem state. Other groups (e.g. nematodes), following a decrease between two and six months post clipping, remained at lower abundance levels for the duration of simulated grazing. Reasons for these different responses among infaunal groups are not known. Some groups may have been more affected by environmental factors, whereas others may have been more affected by predation pr essure. An increase in a food source, such as benthic microalgae, following canopy removal may have facilitated increases in abundance for some infauna in the long term. Changes in environmental factors or physical processes may also have differentially af fected infaunal groups. Investigating the mechanisms behind temporal dynamics in infauna abundance within grazed areas will be a beneficial future research direction for understanding the ecological effects of green turtle grazing. Though total infauna ab undance decreased, and individual group dynamics were temporally variable, the diversity of the infaunal community was not affected by long number of groups present (rich ness) and relative abundance of each group. As abundances changed over time within groups, some became more dominant while others became less dominant. Clipping reduced the dominance (relative abundance) of some groups (e.g. nematodes and peracarids) compa red to reference plots after 16 months, allowing other groups, such as non nematode aschelminthes and polychaetes,
121 to increase their dominance within the community (16 months, Fig. 5 4). While shifting relative dominance among groups did not affect infauna l diversity, it did affect the composition of the infaunal community within clipped plots compared to reference plots during the experiment (Fig. 5 4; Table 5 2). Infaunal communities differ among seagrass meadows (Ansari et al. 1991) , and some infauna have even been found to be associated with specific species of seagrass (Liao et al. 2016) . It is possible that other meadows, differing in their seagrass species and infaunal communities, may respond differently to green turtle grazing from that in the present study. Grazing is likely to affect the c omposition of the infaunal community regardless of the community composition prior to grazing, however. Unless abundance decreases proportionally across all groups inhabiting a meadow (so infaunal group evenness is unaffected), changes to the composition o f the community will occur. Green turtle grazing may lead to additional indirect effects within seagrass meadows through changes to the infaunal community. As abundance and composition of the community change, other ecosystem functions and services may als o be affected, such as prey availability for juvenile fishes using the meadow as foraging habitat. These effects of grazing may not affect entire seagrass meadows, however. Green turtles do not often graze entire seagrass meadows (though there are exceptio ns; e.g. Fourqurean et al. 2010; Christianen et al. 2014) . Green turtles in the Caribbean create discrete foraging patches within a meadow (Bjorndal 1980) , resulting in a mosaic of grazed and ungrazed areas of seagr ass. The effects of green turtle grazing on the infaunal community and indirect effects stemming therefrom may scale to the amount of grazing within a meadow. However, the distance to which edge
122 effects may propagate beyond a grazed area may differ among i nfaunal groups (Bell et al. 2001; Tanner 2005) , and it is possible certain groups may be affected across an area greater than that which is grazed by turtles. With increasing green turtle abundance and grazing in areas such as the Caribbean (Chaloupka et al. 2008) , more seagrass area will return to a natural grazed state, and the effects on ecosystem functions mediated by the infaunal communities within seagrass meadows will become wider spread. Seagrass infauna play important roles in seagrass ecosystem functioning, such as the consumption of detrital matter and energy transfer up the food web (Edgar and Shaw 1995) . Seagrass meadow canopies baffle waves and reduce current speed, creating a calmer microhabitat near the sediment surface (Fonseca et al. 1982; Gacia et al. 1999; Reidenbach and Thomas 2018) . This helps meadows trap organic matter, such as senescent seagrass blades or matter brought in from external habitats (Gacia et al. 2002; Kennedy et al. 2010) , which may be incorporated into the detrital layer. Many infaunal species are detritivores, and play an essential role in the recycling of this organic matter (Hemminga and Duarte 2000) . A decrease in the detritivorous infauna within a meadow could lead to a slowing down of organic matter and nutrient recycling, possibly affecting the nutrien ts available for seagrass growth. Foraging habitat for juveniles of many fish species is an important ecosystem service provided by seagrass meadows, and changes to the infaunal community through grazing may affect these consumer populations. Grazing open s up the sediment surface, potentially creating a habitat in which foraging for infauna becomes easier. The decrease in abundance observed across many infaunal groups may have been a result of increased predation following removal of the seagrass canopy. S urface -
123 dwelling fauna, such as small crabs, may be especially vulnerable within grazed areas, and we measured the abundance of crab claws within plots as a potential sign of predation to investigate this. The significant increase in crab claws within clipp ed plots during the study, preceded by a sharp decline in decapod abundance (Fig. 5 2), could be a sign that predation increased following removal of the protective seagrass canopy. An increase in predation of infauna by fishes within experimentally clippe d plots was not directly observed during this study. However, normal fish behavior may have been altered by the presence of divers within the meadow. Additionally, all activities were conducted during daytime, so any potential nighttime or crepuscular pred ation by fishes or other consumers would not have been observed. Further research is needed to better understand the cause of declines in infauna abundance within grazed areas. A decline in the structural complexity of the seagrass meadow canopy may also explain the decrease in infauna abundance following experimental clipping. Simply removing the epiphytes from artificial seagrass blades deployed in an Australia Amphibolis griffithii meadow led to a significant decrease in abundance of invertebrate fauna (Gartner et al. 2013) . When green turtles graze seagrass, they crop blades near the sediment su rface. This decreases the meadow canopy height, but also removes epiphytes, which typically grow on the older portions of seagrass blades (Hemminga and Duarte 2000) . In monospecifi c seagrass meadows, such as the Thalassia testudinum meadow in the present study, grazing may result in a much more structurally simple canopy compared to areas left ungrazed. The decrease observed across many infaunal groups following clipping may therefo re not have been a result of increased predation, but rather a result of decreased canopy complexity and reduced protection
124 (e.g. baffling) against physical processes (Gartner et al. 2013) . Future investigations conducted in meadows dominated by species other than T. testudinum , in which the seagrass canopy complexity and epiphyte community differ f rom those in the present study would be particularly beneficial for furthering our understanding on the effects of grazing on the infaunal communities in seagrass meadows. Green turtles affect the infaunal communities inhabiting seagrass ecosystems through removal of the aboveground canopy. These changes are likely to cause further changes to ecosystem processes which are mediated by aspects of the infaunal community. Infaunal communities are comprised of many species (Edgar et al. 1994; Skilleter et al. 2007; Liao et al. 2016) , and they play important roles in seagrass ecosystems, from affecting the rate at which organic matter and nutrients are passed through the detrital cycle, t o acting as a source of prey for numerous consumers, such as juvenile fishes. Infauna therefore affect the rate at which energy and nutrients may be transferred through seagrass food webs. As green turtle populations increase, more seagrass will be consume d, and greater areas of seagrass meadows will be returned to a naturally grazed state. Our results demonstrate that grazing affects infaunal communities and can lead to a decrease in abundance. Given the roles infauna play in seagrass ecosystems, effects o n these communities should be taken into account when evaluating the effects of green turtles in seagrass habitats.
125 Table 5 1. Results of linear mixed effects models evaluating effects of experimental clipping on individual infaunal groups (and crab cla ws) during the 16 month experiment. Significant results are in bold and denoted with an (*). In all cases of a significant effect of treatment, abundance was lower in clipped plots than reference plots (except Crab claws, for which abundance was greater in clipped plots). Group df F p Nematodes Time 2, 41 0.14 0.87 Treatment 1, 41 17.40 <0.01 * time:treatment 2, 41 0.32 0.73 Non nematode Aschelminthes Time 2, 41 4.06 0.02 * Treatment 1, 41 1.56 0.22 time:treatment 2, 41 1.08 0.35 Polychaetes Time 2, 41 3.47 0.04 * Treatment 1, 41 13.30 <0.01 * time:treatment 2, 41 1.14 0.33 Oligochaetes Time 2, 41 3.33 0.05 * Treatment 1, 41 1.77 0.19 time:treatment 2, 41 1.46 0.24 Gastropods time 2, 41 2.14 0.13 treatment 1, 41 6.72 0.01 * time:treatment 2, 41 0.40 0.67 Bivalves time 2, 41 3.63 0.04 * treatment 1, 41 0.78 0.38 time:treatment 2, 41 3.83 0.03 * Peracarids time 2, 41 3.54 0.04 * treatment 1, 41 19.26 <0.01 * time:treatment 2, 41 2.75 0.08
126 Table 5 1. Continued. Group df F p Copepods time 2, 41 3.30 0.05 * treatment 1, 41 9.80 <0.01 * time:treatment 2, 41 0.27 0.76 Ostracods time 2, 41 0.36 0.70 treatment 1, 41 3.20 0.08 time:treatment 2, 41 0.12 0.89 Isopods time 2, 41 0.39 0.68 treatment 1, 41 4.76 0.03 * time:treatment 2, 41 0.68 0.51 Decapods time 2, 41 1.08 0.35 treatment 1, 41 3.64 0.06 time:treatment 2, 41 0.48 0.62 Crab claws time 2, 41 6.07 <0.01 * treatment 1, 41 6.89 0.01 * time:treatment 2, 41 1.97 0.15
127 Table 5 2. Results of the permutational multivariate analysis of variance (PERMANOVA) evaluating infaunal community composition using Bray Curtis distances. Differences in community composition between clipped and reference plots were evaluated over the course of the clipping experiment (full model), as well as between treatments (i.e. no time component) at zero months (pre clipping) and 16 months (end of the clipping experiment). Infaunal communities differed in their composition both throughout and at the end of the experiment but did not differ prior to experimental clipping. Significant results are in bold and denoted with an (*). df MS F p Full Model treatment 1 0.95 7.67 <0.01 * time 2 0.20 1.58 0.10 treatment:time 2 0.16 1.32 0.20 Residuals 42 0.12 Total 47 Zero Months treatment 1 0.13 0.94 0.47 Residuals 14 0.14 Total 15 Sixteen months treatment 1 0.43 4.97 0.01 * Residuals 14 0.09 Total 15
128 Table 5 3. Results of linear mixed effects models evaluating differences in abundance between experimentally clipped and unclipped reference plots after 16 months of clipping for all infaunal groups (plus crab claws) and total infauna abundance. Significant results are in bold and denoted with an (*). Nematode, peracarid, and total infauna abundance were all significantly lower in clipped plots than reference plots at the end of th e experiment. Group df F p Individual infaunal groups Nematodes 1, 13 16.16 <0.01 * Non nematode Aschelminths 1, 13 0.51 0.49 Polychaetes 1, 13 1.96 0.19 Oligochaetes 1, 13 0.00 1.00 Gastropods 1, 13 0.68 0.42 Bivalves 1, 13 1.62 0.23 Peracarids 1, 13 9.39 0.01 * Copepods 1, 13 2.15 0.17 Ostracods 1, 13 3.36 0.09 Isopods 1, 13 1.30 0.27 Decapods 1, 13 1.34 0.27 Crab claws 1, 13 3.06 0.10 Total infauna abundance 1, 13 9.55 0.01 *
129 Table 5 4. Seagrass meadow characteristics in experimentally clipped and unclipped reference plots during the experiment. Data are mean Â± SE of the eight plots sampled for each treatment. Treatment Blade length Blade width Shoot density AG biomass BG biomass cm cm shoots m 2 g DM m 2 g DM m 2 Pre clipping 0 months Clipped 12.6 Â± 0.4 0.7 Â± 0.0 599.3 Â± 16.5 80.7 Â± 5.8 682.5 Â± 79.5 Reference 12.2 Â± 1.0 0.7 Â± 0.0 579.3 Â± 29.6 79.6 Â± 2.8 590.7 Â± 94.1 Post clipping 2 months Clipped 8.0 Â± 0.3 0.6 Â± 0.0 697.3 Â± 42.3 30.8 Â± 2.2 748.8 Â± 108.6 Reference 12.5 Â± 0.7 0.7 Â± 0.0 772.0 Â± 38.7 63.2 Â± 5.2 715.1 Â± 59.0 6 months Clipped 2.9 Â± 0.3 0.6 Â± 0.0 668.0 Â± 41.1 15.1 Â± 0.7 797.1 Â± 156.5 Reference 11.9 Â± 0.6 0.7 Â± 0.0 716.0 Â± 45.3 59.3 Â± 4.6 837.4 Â± 110.4 11 months Clipped 4.0 Â± 0.5 0.6 Â± 0.0 732.7 Â± 46.1 25.9 Â± 2.1 572.0 Â± 97.4 Reference 11.7 Â± 0.4 0.7 Â± 0.0 779.3 Â± 50.7 60.4 Â± 4.1 756.8 Â± 164.7 16 months Clipped 4.5 Â± 0.2 0.6 Â± 0.0 698.0 Â± 43.8 14.0 Â± 0.6 718.4 Â± 178.5 Reference 13.9 Â± 0.6 0.7 Â± 0.0 761.3 Â± 48.2 67.6 Â± 4.2 657.8 Â± 111.6 AG: aboveground; BG: belowground; DM: dry mass
130 Table 5 5. Sediment characteristics in experimentally clipped and unclipped reference plots in the seagrass meadow during the experiment. Data are mean Â± SE of the eight plots sampled for each treatment. Treatment Detrital depth OM Sand Silt Clay cm % % % % Pre clipping 0 months Clipped 1.9 Â± 0.2 4.6 Â± 0.2 85.7 Â± 1.3 11.4 Â± 1.2 2.9 Â± 0.2 Reference 1.9 Â± 0.1 4.5 Â± 0.1 84.1 Â± 1.5 13.0 Â± 1.2 2.8 Â± 0.3 Post clipping 2 months Clipped 3.5 Â± 0.5 4.7 Â± 0.1 84.5 Â± 2.4 12.0 Â± 2.1 3.4 Â± 0.4 Reference 4.0 Â± 0.6 4.7 Â± 0.1 87.0 Â± 1.4 10.1 Â± 1.3 3.0 Â± 0.3 6 months Clipped 2.7 Â± 0.2 4.7 Â± 0.1 74.1 Â± 0.8 22.9 Â± 0.9 3.0 Â± 0.2 Reference 3.0 Â± 0.3 4.8 Â± 0.1 73.6 Â± 0.9 23.5 Â± 0.9 2.9 Â± 0.3 11 months Clipped 2.4 Â± 0.2 4.3 Â± 0.1 73.3 Â± 1.3 23.5 Â± 1.4 3.2 Â± 0.3 Reference 2.9 Â± 0.2 4.7 Â± 0.1 71.7 Â± 2.1 24.3 Â± 1.9 4.0 Â± 0.4 16 months Clipped 2.7 Â± 0.2 4.7 Â± 0.1 76.0 Â± 2.7 19.6 Â± 2.1 4.4 Â± 0.7 Reference 3.5 Â± 0.3 4.7 Â± 0.1 76.7 Â± 3.3 18.5 Â± 2.7 4.7 Â± 0.6 OM: organic matter
131 Figure 5 1. Total infauna abundance over the course of the 16 month clipping experiment in clipped (open points) and reference (closed points) plots. Data are mean Â± SE. Abundance was significantly different between clipped and reference plots at 16 months (LME; F 1,13 = 9.6, p = 0.01). Sampling at zero months was prior to clipping initiation.
132 Figu re 5 2. Abundance of individual infaunal groups (and crab claws) over the course of the 16 month clipping experiment in clipped (open points) and reference (closed points) plots. Data are mean Â± SE. Abundance was significantly different between clipped and reference plots at 16 months for nematodes and peracarids (Table 5 3). Sampling at zero months was prior to clipping initiation.
133 Figure 5 points) and reference (closed points) plots over the course of the 16 month clipping experiment. Data are mean Â± SE. Diversity was not significantly affected by clipping over the course of the experiment. Sampling at zero months was prior to clipping initiation.
134 Figure 5 4. Infau nal community composition. Percentage of the total infauna abundance that infaunal groups comprised at each sampling time during the experiment in clipped (left) and reference (right) plots. Infaunal community composition became significantly different bet ween clipped and reference plots during the experiment (PERMANOVA; Table 5 2). NNA: non nematode aschelminthes. at very low abundances (e.g. tube worms). Sampling at zero months was prior to clipping initiation.
135 Figure 5 5. Relationships between total infauna abundance and aboveground seagrass biomass ( a ) and the organic matter content of the surface sediments ( b ). Open points are clipped plots and closed points are referenc e plots. While linear regression was significant for each relationship (solid lines; a : R 2 = 0.15, p < 0.01; b : R 2 = 0.04, p = 0.04), neither variable explained much of the variation in abundance. Abundance data were square root transformed prior to analys is. High aboveground biomass values for open data points in ( a ) are from clipped plots prior to clipping (0 months).
136 CHAPTER 6 FINAL THOUGHTS AND FUTURE DIRECTIONS What Have We Learned? Green turtles play important roles in the tropical and subtropical seagrass ecosystems in which they forage. Their distinct grazing strategy, in which turtles uniformly crop the seagrass blades within an area to a short height above the sediment surface, and subsequently re graze the same area, structurally alters seagra ss meadows in a way unlike that of other seagrass grazers. Green turtles may graze like this because it stimulates new tissue growth within these grazing patches that is higher in nutrient content than the surrounding ungrazed seagrass blades (Bjorndal 1980; Mor an and Bjorndal 2007) , but this grazing strategy also has effects that cascade through the seagrass ecosystem. Grazing directly affects the structure and productivity of the seagrass (Moran and Bjorndal 2005) , and through these structural change s grazing can indirectly affect the seagrass species composition of the meadow (Lal et al. 2010; HernÃ¡ndez and van Tussenbroek 2014) as well as the diversity of the seagrass and macr oalgae community (Hearne et al. 2019) . Reductions in the structural complexity of the abo veground canopy can also alter the species and abundance of fauna inhabiting or using grazed areas. These effects of grazing focus on the flora and fauna of meadows, however, and have left open opportunities to further explore the effects of grazing on phy sical and chemical properties and processes within seagrass ecosystems. Studies of physical and chemical processes within meadows to date have generally focused on the benefits provided by, or effects of, the presence of seagrass compared to unvegetated areas (e.g. Gacia and Duarte 2001; B arrÃ³n et al. 2006) . The
137 studies comprising this dissertation are the first to investigate and experimentally test the effects of grazing on carbon dynamics within seagrass meadows naturally grazed by green turtles, sometimes with surprising results. Se agrass meadows form some of the most productive ecosystems in the world, in part due to their high rates of metabolism (Duarte et al. 2010) . In Chapter 2 we investigated the effects of green turtle grazing on the metabolic carbon dynamics of a seagrass ecosystem in Little Cayman, Cayman Islands. We measured metabolic rates of the ecosystem over time following experi mental clipping to simulate grazing, and corroborated results from this experiment with metabolic rates measured in a nearby seagrass meadow that was actively grazed by green turtles. I had hypothesized that metabolic rates would be lower in areas grazed b y turtles (clipped or grazed) compared to ungrazed areas of seagrass, as a result of less photosynthetic biomass present in grazed areas, which we did observe. However, I had further hypothesized that following removal of the seagrass canopy, surface sedim ents would receive greater oxygen input from the water column and stimulate benthic microbial respiration, leading to an increase in benthic remineralization and a metabolic loss of stored carbon. Our results did not support this, however. Production and r espiration responded proportionally to grazing, and the meadow remained net autotrophic, demonstrating that grazing did not lead to a metabolic release of carbon in Little Cayman. Though metabolism contributes to high rates of carbon capture, the majority of carbon within meadows is stored belowground in the sediments (Fourqurean et al. 2012) . In Chapter 3 we investigated how grazing affects the sediment carbon dynamics in seagrass meadows. From the same clipping experiment we conducted for Chapter 2,
138 we measured the erosion of the surface sediments in a meadow following the initial removal of the seagrass canopy. We also measured rates of sediment particle deposition and resuspension in a meadow that had been naturally grazed by turtles for more than a year. The methods for measuring these processes were adapted from previous studies (e.g. Gacia et al. 1999; Hu et al. 2015) which were important for demonstrating the sediment capture and retention capabilities of seagrass meadows, but which had not been applied to grazed areas. Of all my chapters, the results of Chapter 3 were perhaps most surprising. I had hypothesized that erosion of surface sediment would increase following clipping, and that rates of sediment resuspension would be higher in grazed than ungrazed meadows. Specifically, when the aboveground seagrass canopy was removed, sur face sediments would no longer be protected and become more vulnerable to loss. What we found was that sediment dynamics in grazed areas (clipped or natural) were no different from those in ungrazed seagrass areas, demonstrating that grazed seagrass meadow s can be just as important for sediment retention (and all of its associated carbon) as ungrazed meadows. Once we had experimentally shown that green turtle grazing does not necessarily lead to a loss of carbon reserves from a seagrass meadow, this led u s to ask if this pattern also held in other meadows. My Chapter 4 was inspired by our results from Chapter 2, and we investigated metabolic carbon dynamics in meadows grazed by green turtles at five locations across the Greater Caribbean and Gulf of Mexico . As hypothesized, metabolic carbon capture rates were lower in grazed meadows than ungrazed meadows, but remained positive at all sites measured. Grazing had a consistent effect on the metabolic carbon dynamics across meadows in this region. The
139 strength of this response to grazing was not the same across all sites, however, and investigating drivers of these differences among meadows presents an exciting avenue for future research. While more attention has been give to the effects of grazing on species composition within meadows than physical and chemical processes, there are still many aspects of meadow communities yet to be investigated in meadows grazing by green turtles. In chapter 5, we present the first study to experimentally investigate the effe cts of green turtle grazing on the invertebrate infaunal community of a seagrass ecosystem. Skilleter et al. (2007) measured how infauna communities were affected within dugong grazing trails, but the green tur tle grazing strategy is distinct from that of other seagrass grazers, and may therefore have different effects on these infaunal communities. In our experiment, we had hypothesized that simulated green turtle grazing would lead to a reduction in infaunal a bundance within clipped areas and lead to a shift in the composition of the community. Our results offered support for both hypotheses. Total infaunal abundance decreased in plots within two months of clipping initiation, and at the end of the experiment, infaunal community composition differed between clipped and reference plots. Taken together, the results of these chapters further our understanding of the effects that green turtle grazing has on ecological dynamics within seagrass meadows. Through two independent experiments investigating separate aspects of ecosystem carbon dynamics metabolic and physical sediment processes we have reached a similar conclusion: green turtle grazing does not stimulate a direct release of carbon stored in Caribbean seagr ass ecosystems. Reduced rates of metabolic carbon capture
140 within grazing patches may result in a future reduction in carbon sequestration rates, but carbon already stored within meadows will not suddenly be lost following grazing. The effects that grazing has on the infaunal communities within meadows may also lead to further indirect effects on ecosystem processes, such as organic matter recycling and energy transfer. This dissertation research demonstrates that the current paradigm for viewing seagrass m eadows as either vegetated or unvegetated is in need of an update. At the very least, grazed meadows need to be added to the picture, as we have shown here that ecological processes within grazed meadows do not operate entirely like those in either ungraze d or unvegetated areas, but fall somewhere in between. This is likely still an oversimplification of meadows however, which should instead be viewed along a continuum from ungrazed seagrass to unvegetated sediment, with various degrees of grazing in betwee n. Implications f or Green Turtle Grazing Carbon sequestration and the storage capacities of seagrass meadows have gained considerable attention in the literature in recent decades (Duarte et al. 2005; McLeod et al. 2011; Fourqurean et al. 2012) . More recently, the importance of seagrass meadows, and even their possible use/protection to help mitigate global climate change (MarbÃ et al. 2015; Macreadie et al. 2017) , has begun to make its way to the attention of the wider public and mainstream media. This focus on carbon sequestration and storage has led to the suggestion of helping recover predator populations to control grazer population sizes so as to prioritize the protection of carbon stocks as a conservation strategy (Atwood et al. 201 5) .
141 While we have shown that rates of metabolic carbon capture are lower in areas grazed by turtles than areas left ungrazed, we have also shown independently for metabolic carbon dynamics and sediment carbon dynamics that grazing did not stimulate a release o f carbon stored in the meadow prior to grazing. These results of ours suggest that increasing grazer population sizes and protecting carbon stocks do not need to be mutually exclusive conservation strategies for seagrass ecosystems. Protection of belowgrou nd carbon stores can remain a viable conservation strategy for seagrasses without having to reduce or exclude herbivores from the system. Throughout the preceding chapters of this dissertation, we have shown there are numerous effects of grazing beyond dir ect effects to the seagrass, and these indirect and cascading effects of grazing should be taken into account when planning the conservation of seagrass habitats. Looking Forward The importance of green turtle grazing, and its ecological role within seag rass meadows, is well known (e.g. Valentine and Duffy 2006; Heithaus 2013) . However, there are still many aspects of seagrass meadow ecology that have received little attention in regard to the effects of grazing. The preceding chapters fill in so me of the gaps in our knowledge on how ecosystem processes are affected within grazed meadows, but there are still many questions to be answered. Does grazing in meadows in other regions of the world have similar effects to those in the Greater Caribbean? Do seagrass meadows in deeper waters respond to grazing in similar ways to shallow, coastal meadows? Does grazing within meadows dominated by small, fast growing seagrass species (e.g. Halophila spp.) have similar effects to those in the Thalassia testudin um dominated meadows of the Caribbean? This last question is particularly
142 important given the current invasion of Caribbean meadows by the seagrass Halophila stipulacea (Willette et al. 2014) . If it continues to outcompete the native species and reduce available T. testudinum grazing habitat (Christianen et al. 2018) , green turtles may be forced to choose between feeding on this invader, or moving to a new habitats. We know considerably more about the effects of green turtle grazing on seagrass meadow carbon dynamics than we did before. Further studies of green turtle grazing are needed, however, to gain a more complete understanding of the effects of this circumglobal species in coastal habitats. Grazing studies from regions of the world beyond the Caribbean, and from meadows dominated by seagrasses other than T. testudinum including both local (e.g . single location) experimental studies and large scale, regional comparisons will be particularly beneficial for our understanding of the roles of green turtles in seagrass meadows. There are plenty more questions to ask, and I am excited for all the sci ence yet to come. Through all that I have learned during this dissertation , I am left feeling optimistic about the future of green turtles and seagrass ecosystems as we prepare for more seagrass to return to a natural grazed state.
143 APPENDIX CHAPTER 4 SUPPLEMENTAL INFORMATION Table A 1. Shoot densities for Thalassia testudinum , Halophila stipulacea , Syringodium filiforme , and Halodule wrightii seagrass species, total seagrass shoot density, and total macroalgae density from all meadows at all sites. Values are mean Â± SD. Site Meadow Thalassia density Halophila density Syringodium density Halodule density Total seagrass Total macroalgae shoots m 2 shoots m 2 shoots m 2 shoots m 2 shoots m 2 thalli m 2 Bonaire Lac Cai Beach Grazed 741.3 Â± 126.3 741.3 Â± 126.3 Ungrazed 954.7 Â± 93.7 954.7 Â± 93.7 40.0 Â± 67.7 H. stipulacea 400.0 Â± 189.7 3450.0 Â± 731.4 3850.0 Â± 806.8 NW Lac Bay Grazed 237.3 Â± 44.6 237.3 Â± 44.6 5.3 Â± 13.1 H. stipulacea 150.0 Â± 151.7 3333.3 Â± 637.7 3483.3 Â± 549.2 33.3 Â± 81.6 St. Croix Buck Island Grazed 886.4 Â± 124.2 1020.8 Â± 164.5 57.6 Â± 66.5 1964.8 Â± 226.1 3.2 Â± 7.2 Ungrazed 851.2 Â± 328.6 1555.2 Â± 474.6 2406.4 Â± 250.1 6.4 Â± 8.8 H. stipulacea 3540.8 Â± 1117.9 60.0 Â± 134.2 132.8 Â± 262.6 3733.6 Â± 1249.4 Little Cayman Grape Tree Bay Grazed 906.7 Â± 130.2 50.7 Â± 87.3 957.3 Â± 148.5 32.0 Â± 36.5 Ungrazed 736.0 Â± 73.0 282.7 Â± 167.4 1018.7 Â± 210.2 37.3 Â± 24.1 Eleuthera Arvida Bay Grazed 1437.3 Â± 255.1 210.7 Â± 197.4 1648.0 Â± 392.7 13.3 Â± 25.6 Ungrazed 1544.0 Â± 164.7 509.3 Â± 258.9 2053.3 Â± 329.9 26.7 Â± 29.8 Half Sound Grazed 1429.3 Â± 226.0 245.3 Â± 127.5 1674.7 Â± 316.7 16.0 Â± 20.2 Ungrazed 1482.7 Â± 152.7 293.3 Â± 187.9 8.0 Â± 19.6 1784.0 Â± 284.7 5.3 Â± 8.3 Florida North Rack Grazed 1184.0 Â± 489.8 64.0 Â± 42.3 37.3 Â± 33.3 1322.7 Â± 444.1 53.3 Â± 24.4 Ungrazed 1146.7 Â± 205.7 170.7 Â± 24.4 69.3 Â± 40.3 1386.7 Â± 161.9 85.3 Â± 51.4
144 Table A 2. Metabolic rates (GPP, R E , NEP) from grazed and ungrazed Thalassia testudinum meadows and Halophila stipulacea meadows from each sampling site. Values are mean Â± SD of the three metabolic incubation chambers deployed in each meadow. Site Sampling date Meadow GPP R E NEP mmol C m 2 d 1 Bonaire Lac Cai Beach July 2018 Grazed 56.0 Â± 7.5 17.9 Â± 3.8 38.1 Â± 11.2 Ungrazed 307.3 Â± 51.5 196.7 Â± 46.0 110.6 Â± 45.5 H. stipulacea 232.3 Â± 21.1 133.3 Â± 44.3 98.9 Â± 23.3 NW Lac Bay July 2018 Grazed 51.7 Â± 11.0 0.0 Â± 0.0 51.7 Â± 11.0 H. stipulacea 95.0 Â± 38.8 33.8 Â± 18.4 61.2 Â± 22.6 St. Croix Buck Island February 2018 Grazed 48.3 Â± 16.7 5.3 Â± 3.3 43.1 Â± 15.1 Ungrazed 122.1 Â± 14.2 24.3 Â± 16.6 97.9 Â± 30.7 H. stipulacea 84.3 Â± 21.1 13.1 Â± 12.3 71.1 Â± 27.2 Little Cayman Grape Tree Bay June 2016 Grazed 29.5 Â± 6.7 7.9 Â± 3.9 21.6 Â± 8.3 Ungrazed 376.5 Â± 107.0 177.9 Â± 34.2 198.6 Â± 89.0 July 2016 Grazed 36.1 Â± 20.2 27.0 Â± 42.5 9.1 Â± 23.4 Ungrazed 358.0 Â± 22.2 132.7 Â± 10.3 225.3 Â± 19.1 July 2016 Grazed 40.0 Â± 14.9 22.2 Â± 20.9 17.8 Â± 14.9 Ungrazed 347.6 Â± 50.0 170.6 Â± 28.7 177.0 Â± 68.3 Eleuthera Arvida Bay August 2018 Grazed 91.7 Â± 11.9 44.5 Â± 23.4 47.1 Â± 12.7 Ungrazed 200.8 Â± 13.6 73.6 Â± 23.1 127.2 Â± 31.4 Half Sound August 2018 Grazed 30.4 Â± 13.3 9.8 Â± 6.8 20.6 Â± 17.6 Ungrazed 152.6 Â± 11.2 88.3 Â± 41.9 64.3 Â± 30.8 Florida North Rack May 2018 Grazed 50.5 Â± 9.6 46.3 Â± 4.5 4.2 Â± 7.0 Ungrazed 131.2 Â± 37.8 78.9 Â± 37.1 52.3 Â± 5.7
145 Figure A 1. Relationship between net ecosystem production and environmental temperature ( a ) and irradiance ( b ) from grazed (open points) and adjacent ungrazed (closed points) Thalassia testudinum seagrass meadows. NEP was not significantly related to temperature (R 2 = 0.01, p = 0.70) or irradiance (R 2 < 0.01, p = 1.0).
146 Figure A 2. Relationship between rates of net ecosystem production after accounting for differences in aboveground seagrass biomass (unit NEP per unit biomass) and environmental temperature ( a ) and irradiance ( b ) from grazed (open points) and adjacent ungrazed (close d points) Thalassia testudinum seagrass meadows. Mass specific NEP was not significantly related to either temperature (R 2 < 0.01, p = 0.91) or irradiance (R 2 = 0.01, p = 0.72).
147 LIST OF REFERENCES Agawin, N. S. R., and C. M. Duarte. 2002. Evidence of direct particle trapping by a tropical seagrass meadow. Estuaries 25 : 1205 1209. doi:10.1007/BF02692217 Alcoverro, T., M. Manzanera, and J. Romero. 2001. Ann ual metabolic carbon balance of the seagrass Posidonia oceanica : The importance of carbohydrate reserves. Mar. Ecol. Prog. Ser. 211 : 105 116. doi:10.3354/meps211105 Anderson, M. J. 2001. A new method for non parametric multivariate analysis of variance. Au stral Ecol. 32 46. Ansari, Z. A., C. U. Rivonker, P. Ramani, and A. H. Parulekar. 1991. Seagrass habitat complexity and macroinvertebrate abundance in Lakshadweep coral reef lagoons, Arabian Sea. Coral Reefs 10 : 127 131. doi:10.1007/BF00572170 Anton, A., J . Cebrian, C. M. Duarte, K. L. H. Jr, and J. Goff. 2009. Low impact of Hurricane Katrina on seagrass community structure and functioning in the northern Gulf of Mexico. Bull. Mar. Sci. 85 : 45 59. Apostolaki, E. T., M. Holmer, N. MarbÃ , and I. Karakassis. 2 010. Metabolic imbalance in coastal vegetated ( Posidonia oceanica ) and unvegetated benthic ecosystems. Ecosystems 13 : 459 471. doi:10.1007/s10021 010 9330 9 Atwood, T. B., R. M. Connolly, E. G. Ritchie, C. E. Lovelock, M. R. Heithaus, G. C. Hays, J. W. Fou rqurean, and P. I. Macreadie. 2015. Predators help protect carbon stocks in blue carbon ecosystems. Nat. Clim. Chang. 5 : 1038 1045. doi:10.1038/NCLIMATE2763 BarrÃ³n, C., and C. M. Duarte. 2009. Dissolved organic matter release in a Posidonia oceanica meadow . Mar. Ecol. Prog. Ser. 374 : 75 84. doi:10.3354/meps07715 BarrÃ³n, C., C. M. Duarte, M. Frankignoulle, and A. V. Borges. 2006. Organic carbon metabolism and carbonate dynamics in a Mediterranean seagrass ( Posidonia oceanica ) meadow. Estuaries and Coasts 29 : 417 426. doi:10.1007/BF02784990 BarrÃ³n, C., N. MarbÃ , J. Terrados, H. Kennedy, and C. M. Duarte. 2004. Community metabolism and carbon budgets along a gradient of seagrass ( Cymodocea nodosa ) colonization. Limnol. Oceanogr. 49 : 1642 1651. doi:10.4319/lo.20 04.49.5.1642 Bell, S. S., R. A. Brooks, B. D. Robbins, M. S. Fonseca, and M. O. Hall. 2001. Faunal response to fragmentation in seagrass habitats: Implications for seagrass conservation. Biol. Conserv. 100 : 115 123. doi:10.1016/S0006 3207(00)00212 3 Bjornd al, K. A. 1980. Nutrition and grazing behavior of the green turtle Chelonia mydas . Mar. Biol. 56 : 147 154. doi:10.1007/BF00397131
148 Bjorndal, K. A. 1997. Foraging ecology and nutrition in sea turtles, p. 199 232. In P.L. Lutz and J.A. Musick [eds.], The Biol ogy of Sea Turtles. CRC Press. Blomqvist, S., and C. Kofoed. 1981. Sediment trapping A subaquatic in situ experiment. Limnol. Oceanogr. 26 : 585 590. doi:10.4319/lo.1981.26.3.0585 Bowles, J. W., and S. S. Bell. 2004. Simulated herbivory and the dynamics of disease in Thalassia testudinum . Mar. Ecol. Prog. Ser. 283 : 127 132. doi:10.3354/meps283127 Buoyoucus, G. J. 1936. Directions for making mechanical analyses of soils by the hydrom eter method. Soil Sci. 42 : 225 229. Burdige, D. J., and R. C. Zimmerman. 2002. Impact of sea grass density on carbonate dissolution in Bahamian sediments. Limnol. Oceanogr. 47 : 1751 1763. doi:10.4319/lo.2002.47.6.1751 Cahoon, D. R., J. C. Lynch, P. Hensel, R. Boumans, B. C. Perez, B. Segura, and J. W. Day. 2002. High precision measurements of sediment elevation: I. Recent improvements to the sediment erosion table. J. Sediment. Res. 72 : 730 733. doi:10.1306/020702720730 Calleja, M. L., C. BarrÃ³n, J. A. Hale , T. K. Frazer, and C. M. Duarte. 2006. Light regulation of benthic sulfate reduction rates mediated by seagrass ( Thalassia testudinum ) metabolism. Estuaries and Coasts 29 : 1255 1264. Camp, E. F., D. J. Suggett, G. Gendron, J. Jompa, C. Manfrino, and D. J. Smith. 2016. Mangrove and seagrass beds provide different biogeochemical services for corals threatened by climate change. Front. Mar. Sci. 3 : 1 16. doi:10.3389/fmars.2016.00052 Cebri Ã¡n, J., and C. M. Duarte. 1998. Patterns in leaf herbivory on seagrasses. Aquat. Bot. 60 : 67 82. doi:10.1016/S0304 3770(97)00070 3 Chaloupka, M., K. A. Bjorndal, G. H. Balazs, and others. 2008. Encouraging outlook for recovery of a once severely exploited marine megaherbivore. Glob. Ecol. Biogeogr. 17 : 297 304. doi:10.1111/j.1466 8238.2007.00367.x Champenois, W., and A. V. Borges. 2012. Seasonal and interannual variations of community metabolism rates of a Posidonia oceanica seagrass meadow. Limnol. Oceanog r. 57 : 347 361. doi:10.4319/lo.2012.57.1.0347 Chen, G., M. H. Azkab, G. L. Chmura, and others. 2017. Mangroves as a major source of soil carbon storage in adjacent seagrass meadows. Sci. Rep. 7 : 42406. doi:10.1038/srep42406 Christianen, M. J. A., J. van Be lzen, P. M. J. Herman, M. M. van Katwijk, L. P. M.
149 Lamers, P. J. M. van Leent, and T. J. Bouma. 2013. Low canopy seagrass beds still provide important coastal protection services. PLoS One 8 : e62413. doi:10.1371/journal.pone.0062413 Christianen, M. J. A., L. L. Govers, T. J. Bouma, W. Kiswara, J. G. M. Roelofs, L. P. M. Lamers, and M. M. van Katwijk. 2012. Marine megaherbivore grazing may increase seagrass tolerance to high nutrient loads. J. Ecol. 100 : 546 560. doi:10.1111/j.1365 2745.2011.01900.x Christia nen, M. J. A., P. M. J. Herman, T. J. Bouma, and others. 2014. Habitat collapse due to overgrazing threatens turtle conservation in marine protected areas. Proc. R. Soc. B Biol. Sci. 281 : 20132890. Christianen, M. J. A., F. O. H. Smulders, M. S. Engel, and others. 2018. Megaherbivores may impact expansion of invasive seagrass in the Caribbean. J. Ecol. 00 : 1 13. doi:10.1111/1365 2745.13021 Dahl, M., D. Deyanova, L. D. Lyimo, J. NÃ¤slund, G. S. Samuelsson, M. S. P. P. Mtolera, M. BjÃ¶rk, and M. GullstrÃ¶m. 2016 . Effects of shading and simulated grazing on carbon sequestration in a tropical seagrass meadow. J. Ecol. 104 : 654 664. doi:10.1111/1365 2745.12564 Dauby, P., A. J. Bale, N. Bloomer, and others. 1995. Particle fluxes over a Mediterranean seagrass bed: a o ne year case study. Mar. Ecol. Prog. Ser. 126 : 233 246. Duarte, C. M. 2002. The future of seagrass meadows. Environ. Conserv. 29 : 192 206. doi:10.1017/S0376892902000127 Duarte, C. M., and J. CebriÃ¡n. 1996. The fate of marine autotrophic production. Limnol. Oceanogr. 41 : 1758 1766. doi:10.4319/lo.19188.8.131.528 Duarte, C. M., and C. L. Chiscano. 1999. Seagrass biomass and production: A reassessment. Aquat. Bot. 65 : 159 174. doi:10.1016/S0304 3770(99)00038 8 Duarte, C. M., H. Kennedy, N. MarbÃ , and I. Hendriks. 2013a. Assessing the capacity of seagrass meadows for carbon burial: Current limitations and future strategies. Ocean Coast. Manag. 83 : 32 38. doi:10.1016/j.ocecoaman.2011.09.001 Duarte, C. M., and D. Krause Jensen. 2017. Export from seagrass meadows contributes to marine carbon sequestration. Front. Mar. Sc i. 4 : 1 7. doi:10.3389/fmars.2017.00013 Duarte, C. M., I. J. Losada, I. E. Hendriks, I. Mazarrasa, and N. MarbÃ . 2013b. The role of coastal plant communities for climate change mitigation and adaptation. Nat. Clim. Chang. 3 : 961 968. doi:10.1038/nclimate19 70
150 Duarte, C. M., N. MarbÃ , E. Gacia, J. W. Fourqurean, J. Beggins, C. BarrÃ³n, and E. T. Apostolaki. 2010. Seagrass community metabolism: Assessing the carbon sink capacity of seagrass meadows. Global Biogeochem. Cycles 24 : 1 8. doi:10.1029/2010GB003793 Du arte, C. M., J. J. Middelburg, and N. Caraco. 2005. Major role of marine vegetation on the oceanic carbon cycle. Biogeosciences 2 : 1 8. doi:10.5194/bgd 1 659 2004 Edgar, G. J., and C. Shaw. 1995. The production and trophic ecology of shallow water fish ass emblages in southern Australia II . Diets of fishes and trophic relationships between fishes and benthos at Western Port , Victoria. J. Exp. Mar. Bio. Ecol. 194 : 83 106. Edgar, G. J., C. Shaw, G. F. Watsona, and L. S. Hammond. 1994. Comparisons of species richness, size structure and production of benthos in vegetated and unvegetated habitats in Western Port, Victoria. J. Exp. Mar. Bio. Ecol. 176 : 201 226. doi:10.1016/0022 0981(94)90185 6 Fonseca, M. S. 1989. Sediment stabilization by Halophila decipiens in comparison to other seagrasses. Estuar. Coast. Shelf Sci. 29 : 501 507. doi:10.1016/0272 7714(89)90083 8 Fonseca, M. S., and J. A. Cahalan. 1992. A preliminary evaluation of wave attenuation by four species of seagrass. Estuar. Coast. Shelf Sci. 35 : 565 57 6. doi:10.1016/S0272 7714(05)80039 3 Fonseca, M. S., J. S. Fisher, J. C. Zieman, and G. W. Thayer. 1982. Influence of the seagrass, Zostera marina L., on current flow. Estuar. Coast. Shelf Sci. 15 : 351 364. doi:10.1016/0272 7714(82)90046 4 Fourqurean, J. W ., C. M. Duarte, H. Kennedy, and others. 2012. Seagrass ecosystems as a globally significant carbon stock. Nat. Geosci. 5 : 505 509. doi:10.1038/ngeo1477 Fourqurean, J. W., S. Manuel, K. A. Coates, W. J. Kenworthy, and S. R. Smith. 2010. Effects of excludin g sea turtle herbivores from a seagrass bed: Overgrazing may have led to loss of seagrass meadows in Bermuda. Mar. Ecol. Prog. Ser. 419 : 223 232. doi:10.3354/meps08853 Gacia, E., and C. M. Duarte. 2001. Sediment retention by a Mediterranean Posidonia ocean ica meadow: The balance between deposition and resuspension. Estuar. Coast. Shelf Sci. 52 : 505 514. doi:10.1006/ecss.2000.0753 Gacia, E., C. M. Duarte, N. MarbÃ , J. Terrados, H. Kennedy, M. D. Fortes, and N. H. Tri. 2003. Sediment deposition and production in SE Asia seagrass meadows. Estuar. Coast. Shelf Sci. 56 : 909 919. doi:10.1016/S0272 7714(02)00286 X
151 Gacia, E., C. M. Duarte, and J. J. Middelburg. 2002. Carbon and nutrient deposition in a Mediterranean seagrass ( Posidonia oceanica ) meadow. Limnol. Oceanogr. 47 : 23 32. doi:10.4319/lo.2002.47.1.0023 Gacia, E., T. C. Granata, and C. M. Duarte. 1999. An approach to measurement of particle flux and sediment retention within seagrass ( Posidonia oceanica ) meadows. Aquat. Bot. 65 : 255 268. doi:10.1016/S0304 3770(99)00044 3 Gacia, E., H. Kennedy, C. M. Duarte, J. Terrados, N. MarbÃ , S. Papadimitriou, and M. Fortes. 2005. Light dependenc e of the metabolic balance of a highly productive Philippine seagrass community. J. Exp. Mar. Bio. Ecol. 316 : 55 67. doi:10.1016/j.jembe.2004.10.008 Gartner, A., F. Tuya, P. S. Lavery, and K. McMahon. 2013. Habitat preferences of macroinvertebrate fauna among seagrasses with varying structural forms. J. Exp. Mar. Bio. Ecol. 439 : 143 151. doi:10.1016/j.jembe.2012.11.009 Gee, G. W., and J. W. Bauder. 198 6. Particle size analysis., p. 383 411. In A. Klute [ed.], Method of Soil Analysis, Part 1. American Society of Agronomy. Green, E. P., and F. T. Short. 2003. World Atlas of Seagrasses, University of California Press. Greiner, J. T., G. M. Wilkinson, K. J. McGlathery, and K. A. Emery. 2016. Sources of sediment carbon sequestered in restored seagrass meadows. Mar. Ecol. Prog. Ser. 551 : 95 105. doi:10.3354/meps11722 HÃ¥kanson, L., S. Floderus, and M. Wallin. 1989. Sediment trap assemblages a methodological des cription. Sediment/Water Interact. 176/177 : 481 490. doi:10.1007/978 94 009 2376 8_46 Hansen, J. C. R., and M. A. Reidenbach. 2013. Seasonal growth and senescence of a Zostera marina seagrass meadow alters wave dominated flow and sediment suspension within a coastal bay. Estuaries and Coasts 36 : 1099 1114. doi:10.1007/s12237 013 9620 5 Hansen, J. W., B. Thamdrup, and B. B. JÃ¸rgensen. 2000. Anoxic incubation of sediment in gas tight plastic bags: A method for biogeochemical process studies. Mar. Ecol. Prog. Ser. 208 : 273 282. doi:10.3354/meps208273 Hearne, E. L., R. A. Johnson, A. G. Gulick, A. Candelmo, A. B. Bolten, and K. A. Bjorndal. 2019. Effects of green turtle grazing on seagrass and macroalgae diversity vary spatially among seagrass meadows. Aquat. Bo t. 152 : 10 15. doi:10.1016/j.aquabot.2018.09.005 Heithaus, M. R. 2013. Predators, Prey, and the Ecological Roles of Sea Turtles. Biol. Sea Turtles, Vol. III 249 285. doi:doi:10.1201/b13895 11
152 Heithaus, M. R., T. Alcoverro, R. Arthur, and others. 2014. Seag rasses in the age of sea turtle conservation and shark overfishing. Front. Mar. Sci. 1 : 1 6. doi:10.3389/fmars.2014.00028 Hemminga, M. A., and C. M. Duarte. 2000. Seagrass Ecology, Cambridge University Press. Hendriks, I. E., Y. S. Olsen, L. Ramajo, L. Bas so, A. Steckbauer, T. S. Moore, J. Howard, and C. M. Duarte. 2014. Photosynthetic activity buffers ocean acidification in seagrass meadows. Biogeosciences 11 : 333 346. doi:10.5194/bg 11 333 2014 Hendriks, I. E., T. Sintes, T. J. Bouma, and C. M. Duarte. 20 08. Experimental assessment and modeling evaluation of the effects of the seagrass Posidonia oceanica on flow and particle trapping. Mar. Ecol. Prog. Ser. 356 : 163 173. doi:10.3354/meps07316 HernÃ¡ndez, A. L. M., and B. I. van Tussenbroek. 2014. Patch dynam ics and species shifts in seagrass communities under moderate and high grazing pressure by green sea turtles. Mar. Ecol. Prog. Ser. 517 : 143 157. doi:10.3354/meps11068 Holzer, K. K., and K. J. McGlathery. 2016. Cultivation grazing response in seagrass may depend on phosphorus availability. Mar. Biol. 163 : 88. doi:10.1007/s00227 016 2855 5 Hu, Z., W. Lenting, D. van der Wal, and T. J. Bouma. 2015. Continuous monitoring bed level dynamics on an intertidal flat: Introducing novel, stand alone high resolution S ED sensors. Geomorphology 245 : 223 230. doi:10.1016/j.geomorph.2015.05.027 Jackson, J. B. C. 1997. Reefs since Columbus. Coral Reefs 16 : S23 S32. doi:10.1007/s003380050238 Jackson, J. B. C. 2001. What was natural in the coastal oceans? Proc. Natl. Acad. Sc i. 98 : 5411 5418. doi:10.1073/pnas.091092898 Jackson, J. B. C., M. X. Kirby, W. H. Berger, and others. 2001. Historical overfishing and the recent collapse of coastal ecosystems. Science 293 : 629 637. doi:10.1126/science.1059199 Johnson, R. A., A. G. Gulic k, A. B. Bolten, and K. A. Bjorndal. 2017. Blue carbon stores in tropical seagrass meadows maintained under green turtle grazing. Sci. Rep. 7 : 13545. doi:10.1038/s41598 017 13142 4 Josselyn, M., M. Fonseca, T. Niesen, and R. Larson. 1986. Biomass, producti on and decomposition of a deep water seagrass, Halophila decipiens ostenf. Aquat. Bot. 25 : 47 61. doi:10.1016/0304 3770(86)90039 2 Kennedy, H., J. Beggins, C. M. Duarte, J. W. Fourqurean, M. Holmer, N. MarbÃ¡, and J.
153 J. Middelburg. 2010. Seagrass sediments as a global carbon sink: Isotopic constraints. Global Biogeochem. Cycles 24 : 1 8. doi:10.1029/2010GB003848 Kennedy, H., E. Gacia, D . P. Kennedy, S. Papadimitriou, and C. M. Duarte. 2004. Organic carbon sources to SE Asian coastal sediments. Estuar. Coast. Shelf Sci. 60 : 59 68. doi:10.1016/j.ecss.2003.11.019 Klumpp, D. W., R. K. Howard, and D. A. Pollard. 1989. Trophodynamics and nutritional ecology of seagrass communities, p. 394 457. In A.W.D. Larkum, A.J. McComb, and S.A. Sheperd [eds.], Biology of Seagrasses. Elsevier. Koch, E. W. 1999. Sediment resuspensi on in a shallow Thalassia testudinum banks ex Konig bed. Aquat. Bot. 65 : 269 280. doi:10.1016/S0304 3770(99)00045 5 Koch, M. S., and C. J. Madden. 2001. Patterns of primary production and nutrient availability in a Bahamas lagoon with fringing mangroves. M ar. Ecol. Prog. Ser. 219 : 109 119. doi:10.3354/meps219109 Kristensen, E., G. Penha Lopes, M. Delefosse, T. Valdemarsen, C. O. Quintana, and G. T. Banta. 2012. What is bioturbation? The need for a precise definition for fauna in aquatic sciences. Mar. Ecol. Prog. Ser. 446 : 285 302. doi:10.3354/meps09506 Kuiper Linley, M., C. R. Johnson, and J. M. Lanyon. 2007. Effects of simulated green turtle regrazing on seagrass abundance, growth and nutritional status in Moreton Bay, south east Queensland, Australia. Mar . Freshw. Res. 58 : 492 503. doi:10.1071/MF06241 Lal, A., R. Arthur, N. MarbÃ , A. W. T. Lill, and T. Alcoverro. 2010. Implications of conserving an ecosystem modifier: Increasing green turtle ( Chelonia mydas ) densities substantially alters seagrass meadows. Biol. Conserv. 143 : 2730 2738. doi:10.1016/j.biocon.2010.07.020 Lamb, J. B., J. A. J. M. van de Water, D. G. Bourne, and others. 2017. Seagrass ecosystems reduce exposure to bacterial pathogens of humans, fishes, and invertebrates. Science 355 : 731 733. d oi:10.1126/science.aal1956 Lawson, S. E., K. J. McGlathery, and P. L. Wiberg. 2012. Enhancement of sediment suspension and nutrient flux by benthic macrophytes at low biomass. Mar. Ecol. Prog. Ser. 448 : 259 270. doi:10.3354/meps09579 Lee, K. S., and K. H. Dunton. 1997. Effects of in situ light reduction on the maintenance, growth and partitioning of carbon resources in Thalassia testudinum Banks ex Konig. J. Exp. Mar. Bio. Ecol. 210 : 53 73. doi:10.1016/S0022 0981(96)02720 7 Lee, K. S., and K. H. Dunton. 199 9. Inorganic nitrogen acquisition in the seagrass Thalassia testudinum: Development of a whole plant nitrogen budget. Limnol. Oceanogr. 44 : 1204 1215. doi:10.4319/lo.19184.108.40.2064
154 Lee, K. S., S. R. Park, and Y. K. Kim. 2007. Effects of irradiance, tempera ture, and nutrients on growth dynamics of seagrasses: A review. J. Exp. Mar. Bio. Ecol. 350 : 144 175. doi:10.1016/j.jembe.2007.06.016 Liao, J. X., H. M. Yeh, and H. K. Mok. 2016. Do the abundance, diversity, and community structure of sediment meiofauna di ffer among seagrass species? J. Mar. Biol. Assoc. United Kingdom 96 : 725 735. doi:10.1017/S0025315415000879 Lindeboom, H. J., and A. J. J. Sandee. 1989. Production and consumption of tropical seagrass fields in Eastern Indonesia measured with bell jars and microelectrodes. Netherlands J. Sea Res. 23 : 181 190. doi:10.1016/0077 7579(89)90012 4 Macreadie, P. I., D. A. Nielsen, J. J. Kelleway, and others. 2017. Can we manage coastal ecosystems to sequester more blue carbon? Front. Ecol. Environ. doi:10.1002/fee .1484 MarbÃ , N., A. Arias Ortiz, P. MasquÃ©, G. A. Kendrick, I. Mazarrasa, G. R. Bastyan, J. Garcia Orellana, and C. M. Duarte. 2015. Impact of seagrass loss and subsequent revegetation on carbon sequestration and stocks. J. Ecol. 103 : 296 302. doi:10.1111/ 1365 2745.12370 MarbÃ , N., and C. M. Duarte. 1998. Rhizome elongation and seagrass clonal growth. Mar. Ecol. Prog. Ser. 174 : 269 280. doi:10.3354/meps174269 MarbÃ , N., M. Holmer, E. Gacia, and C. BarrÃ³n. 2006. Seagrass beds and coastal biogeochemistry, p. 135 157. In A.W.D. Larkum, R.J. Orth, and C.M. Duarte [eds.], Seagrasses: Biology, Ecology and Conservation. Springer. Mateo, M. A., J. CebriÃ¡n, K. Dunton, and T. Mutchler. 2006. Carbon flux in seagrass ecosystems, p. 159 192. In A.W.D. Larkum, R.J. Orth, and C.M. Duarte [eds.], Seagrasses: Biology, Ecology and Conservation. Springer. Mateo, M. A., J. Romero, M. PÃ©rez, M. M. Littler, and D. S. Littler. 1997. Dynamics of millenary organic deposits resulting from the growth of the Mediterranean seagrass Posidonia oceanica . Estuar. Coast. Shelf Sci. 44 : 103 110. doi:10.1006/ecss.1996.0116 Mazaris, A. D., G. Schofield, C. Gkazi nou, V. Almpanidou, and G. C. Hays. 2017. Global sea turtle conservation successes. Sci. Adv. 3 : e1600730. doi:10.1126/sciadv.1600730 McClenachan, L., J. B. C. Jackson, and M. J. H. Newman. 2006. Conservation implications of historic sea turtle nesting bea ch loss. Front. Ecol. Environ. 4 : 290 296. doi:10.1890/1540 9295(2006)4[290:CIOHST]2.0.CO;2 McLeod, E., G. L. Chmura, S. Bouillon, and others. 2011. A blueprint for blue carbon: Toward an improved understanding of the role of vegetated coastal habitats in
155 sequestering CO2. Front. Ecol. Environ. 9 : 552 560. doi:10.1890/110004 de Mendiburu, F. 2017. agricolae: Statistical Procedures for Agricultural Research. R package version 1.2 6. Moran, K. L., and K. A. Bjorndal. 2005. Simulated green turtle grazing affec ts structure and productivity of seagrass pastures. Mar. Ecol. Prog. Ser. 305 : 235 247. doi:10.3354/meps305235 Moran, K. L., and K. A. Bjorndal. 2007. Simulated green turtle grazing affects nutrient composition of the seagrass Thalassia testudinum . Mar. Bi ol. 150 : 1083 1092. doi:10.1007/s00227 006 0427 9 Mortimer, J. A. 1981. The feeding ecology of the west Caribbean green turtle ( Chelonia mydas ) in Nicaragua. Ecology 13 : 49 58. doi:10.2307/2387870 Murdiyarso, D., J. Purbopuspito, J. B. Kauffman, and others . 2015. The potential of Indonesian mangrove forests for global climate change mitigation. Nat. Clim. Chang. 5 : 1089 1092. doi:10.1038/nclimate2734 Murray, L., and R. Wetzel. 1987. Oxygen production and consumption associated with the major autotrophic components in two temperate seagrass communities. Mar. Ecol. Prog. Ser. 38 : 231 239. doi:10.3354/meps038231 Nagel, J. L., W. M. Kemp, J. C. Cornwell, M . S. Owens, D. Hinkle, and C. J. Madden. 2009. Seasonal and regional variations in net ecosystem production in Thalassia testudinum communities throughout Florida Bay. Contrib. Mar. Sci. 38 : 91 108. Ogden, J. C. 1976. Some aspects of herbivore plant relati onships on Caribbean reefs and seagrass beds. Aquat. Bot. 2 : 103 116. doi:10.1016/0304 3770(76)90013 9 Ogden, J. C. 1980. Faunal relationships in Caribbean seagrass beds, p. 173 198. In R.C. Phillips and C.P. McRoy [eds.], Handbook of seagrass biology: An ecosystem perspective. Garland STPM Press. Ogden, J. C., L. Robinson, K. Whitlock, H. Daganhardt, and R. Cebula. 1983. Diel foraging patterns in juvenile green turtles ( Chelonia mydas L.) in St. Croix United States virgin islands. J. Exp. Mar. Bio. Ecol. 6 6 : 199 205. doi:10.1016/0022 0981(83)90160 0 Oksanen, J., F. G. Blanchet, M. Friendly, and others. 2018. vegan: Community Ecology Package. OlivÃ©, I., J. Silva, M. M. Costa, and R. Santos. 2015. Estimating seagrass community metabolism using benthic chamber s: The effect of incubation time. Estuaries and Coasts 39 : 138 144. doi:10.1007/s12237 015 9973 z
156 Oreska, M. P. J., K. J. Mcglathery, and J. H. Porter. 2017a. Seagrass blue carbon spatial patterns at the meadow scale. PLoS One 1 18. doi:https://doi.org/10. 1371/journal.pone.0176630 Oreska, M. P. J., G. M. Wilkinson, K. J. Mcglathery, M. Bost, and B. A. Mckee. 2017b. Non seagrass carbon contributions to seagrass sediment blue carbon. Limnol. Oceanogr. doi:10.1002/lno.10718 Orth, R. J., T. J. B. Carruthers, W. C. Dennison, and others. 2006. A global crisis for seagrass ecosystems. Bioscience 56 : 987 996. Orth, R. J., K. L. Heck, and J. van Montfrans. 1984. Faunal communities in seagrass predator prey relationships. Estuaries 7 : 339 350. Pejrup, M., J. Valeur, and A. Jensen. 199 6. Vertical fluxes of particulate matter in Aarhus Bight, Denmark. Cont. Shelf Res. 16 : 1047 1064. doi:10.1016/0278 4343(95)00044 5 egradation of vegetated coastal ecosystems. PLoS One 7 : e43542. doi:10.1371/journal.pone.0043542 PÃ©rez, M., and J. Romero. 1992. Photosynthetic response to light and temperature of the seagrass Cymodocea nodosa and the prediction of its seasonality. Aquat. Bot. 43 : 51 62. doi:10.1016/0304 3770(92)90013 9 Petuch, E. 2013. Biogeography and Biodiversity of Western Atlantic Mollusks, 1st Editio. CRC Press. Pinheiro, J., D. Bates, S. DebRoy, D. Sarkar, and R Core Team. 2018. _nlme: Linear and Nonlinear Mixed Eff ects Models_. Plus, M., J. Deslous Paoli, I. Auby, and F. Dagault. 2001. Factors influencing primary production of seagrass beds ( Zostera noltii Hornem.) in the Thau lagoon (French Mediterranean coast). J. Exp. Mar. Bio. Ecol. 259 : 63 84. Pollard, P. C., a nd D. J. W. Moriarty. 1991. Organic carbon decomposition, primary and bacterial productivity, and sulphate reduction, in tropical seagrass beds of the Gulf of Carpentaria, Australia. Mar. Ecol. Prog. Ser. 69 : 149 159. doi:10.3354/meps069149 Potouroglou, M. , J. C. Bull, K. W. Krauss, and others. 2017. Measuring the role of seagrasses in regulating sediment surface elevation. Sci. Rep. 1 11. doi:10.1038/s41598 017 12354 y Preen, A. 1995. Impacts of dugong foraging on seagrass habitats: Observational and
157 exper imental evidence for cultivation grazing. Mar. Ecol. Prog. Ser. 124 : 201 213. doi:10.3354/meps124201 R Core Team. 2018. R: A language and environment for statistical computing. R Foundation for Statistical Computing. Reidenbach, M. A., and E. L. Thomas. 20 18. Influence of the seagrass, Zostera marina , on wave attenuation and bed shear stress within a shallow coastal bay. 5 : 1 16. doi:10.3389/fmars.2018.00397 Reyes, E., and M. Merino. 1991. Diel dissolved oxygen dynamics and eutrophication in a shallow, well mixed tropical lagoon (Cancun, Mexico). Estuaries 14 : 372 381. doi:10.1007/BF02689416 Rheuban, J. E., P. Berg, and K. J. McGlathery. 2014. Ecosystem metabolism along a colonization gradient of eelgrass ( Zostera marina ) measured by eddy correlation. Limnol . Oceanogr. 59 : 1376 1387. doi:10.4319/lo.2014.59.4.1376 Robertson, D. R., and K. L. Cramer. 2014. Defining and dividing the Greater Caribbean: Insights from the biogeography of shorefishes. PLoS One 9 . doi:10.1371/journal.pone.0102918 Ruiz, H., and D. L. Ballantine. 2004. Occurrence of the seagrass Halophila stipulacea in the tropical west Atlantic. Bull. Mar. Sci. 75 : 131 135. Scott, A. L., P. H. York, C. Duncan, and others. 2018. The role of herbivory in structuring tropical seagrass ecosystem service de livery. Front. Plant Sci. 9 : 1 10. doi:10.3389/fpls.2018.00127 Serrano, O., M. A. Mateo, P. Renom, and R. JuliÃ . 2012. Characterization of soils beneath a Posidonia oceanica meadow. Geoderma 185 186 : 26 36. doi:10.1016/j.geoderma.2012.03.020 Short, F. T., and S. Wyllie Echeverria. 1996. Natural and human induced disturbance of seagrasses. Environ. Conserv. 23 : 17. doi:10.1017/S0376892900038212 Skilleter, G. A., C. Wegscheidl, and J. M. Lanyon. 2007. Effects of grazing by a marine mega herbivore on benthic assemblages in a subtropical seagrass bed. Mar. Ecol. Prog. Ser. 351 : 287 300. doi:10.3354/meps07174 Smulders, F. O. H., J. A. Vonk, M. S. Eng el, and M. J. A. Christianen. 2017. Expansion and fragment settlement of the non native seagrass Halophila stipulacea in a Caribbean bay. Mar. Biol. Res. doi:10.1080/17451000.2017.1333620 Stoner, A. W. 1980. The role of seagrass biomass in the organization of benthic macrofaunal assemblages. Bull. Mar. Sci. 30 : 537 551.
158 Stoner, A. W., and F. G. Lewis. 1985. The influence of quantitative and qualitative aspects of habitat complexity in tropical sea grass meadows. J. Exp. Mar. Bio. Ecol. 94 : 19 40. doi:10.101 6/0022 0981(85)90048 6 Stutes, J., J. Cebrian, A. L. Stutes, A. Hunter, and A. A. Corcoran. 2007. Benthic metabolism across a gradient of anthropogenic impact in three shallow coastal lagoons in NW Florida. Mar. Ecol. Prog. Ser. 348 : 55 70. doi:10.3354/mep s07036 Tanner, J. E. 2005. Edge effects on fauna in fragmented seagrass meadows. Austral Ecol. 30 : 210 218. doi:10.1111/j.1442 9993.2005.01438.x Terrados, J., and C. M. Duarte. 2000. Experimental evidence of reduced particle resuspension within a seagrass ( Posidonia oceanica L.) meadow. J. Exp. Mar. Bio. Ecol. 243 : 45 53. doi:Doi 10.1016/S0022 0981(99)00110 0 Terrados, J., C. M. Duarte, L. Kamp Nielsen, and others. 1999. Are seagrass growth and survival constrained by the reducing conditions of the sediment ? Aquat. Bot. 65 : 175 197. doi:10.1016/S0304 3770(99)00039 X Thayer, G. W., K. A. Bjorndal, J. C. Ogden, S. L. Williams, and J. C. Zieman. 1984. Role of large herbivores in seagrass communities. Estuaries 7 : 351 376. Thayer, G. W., D. W. Engel, and K. A. B jorndal. 1982. Evidence for short circuiting of the detritus cycle of seagrass beds by the green turtle, Chelonia mydas L. J. Exp. Mar. Bio. Ecol. 62 : 173 183. Thomson, A. C. G., S. M. Trevathan Tackett, D. T. Maher, P. J. Ralph, and P. I. Macreadie. 2018. Bioturbator stimulated loss of seagrass sediment carbon stocks. Limnol. Oceanogr. 1 15. doi:10.1002/lno.11044 Trevathan Tackett, S. M., P. I. Macreadie, J. Sanderman, J. Baldock, J. M. Howes, and P. J. Ralph. 2017. A global assessment of the chemical reca lcitrance of seagrass tissues: Implications for long term carbon sequestration. Front. Plant Sci. 8 : 1 18. doi:10.3389/fpls.2017.00925 van Tussenbroek, B. I., N. Villamil, J. MÃ¡rquez GuzmÃ¡n, R. Wong, L. V. Monroy VelÃ¡zquez, and V. Solis Weiss. 2016. Experi mental evidence of pollination in marine flowers by invertebrate fauna. Nat. Commun. 7 : 12980. doi:10.1038/ncomms12980 van Tussenbroek, B. I., J. A. Vonk, J. Stapel, P. L. A. Erftemeijer, J. J. Middelburg, and J. C. Zieman. 2006. The biology of Thalassia : Paradigms and recent advances in research, p. 409 439. In A.W.D. Larkum, R.J. Orth, and C.M. Duarte [eds.], Seagrasses: Biology, Ecology and Conservation. Springer. Unsworth, R. K. F., C. J. Collier, G. M. Henderson, and L. J. McKenzie. 2012. Tropical seag rass meadows modify seawater carbon chemistry: Implications for coral reefs
159 impacted by ocean acidification. Environ. Res. Lett. 7 : 9. doi:10.1088/1748 9326/7/2/024026 Vadeboncoeur, Y. 2011. Fish diversity, nutrient recycling and periphyton productivity in the littoral zone of Lake Tanganyika. YSI Environ. Valentine, J. F., and J. E. Duffy. 2006. The central role of grazing in seagrass ecology, p. 463 501. In A.W.D. Larkum, R.J. Orth, and C.M. Duarte [eds.], Seagrasses: Biology, Ecology and Conservation. Sp ringer. Valeur, J. R. 1994. Resuspension mechanisms and measuring methods., p. 185 203. In S. Floderus, A. Heiskanen, M. Olesen, and P. Wassmann [eds.], Sediment trap studies in the nordic countries. Marine Biological Lab. Virnstein, R. W. 1977. The import ance of predation by crabs and fishes on benthic infauna in Chesapeake Bay. Ecology 58 : 1199 1217. Wabnitz, C. C., S. Andrefouet, D. Torres Pulliza, F. E. Muller Karger, and P. A. Kramer. 2008. Regional scale seagrass habitat mapping in the Wider Caribbean region using Landsat sensors: Applications to conservation and ecology. Remote Sens. Environ. 112 : 3455 3467. doi:10.1016/j.rse.2008.01.020 Ward, L. G., W. M. Kemp, and W. R. Boynton. 1984. The influence of waves and seagrass communities on suspended particulates in an estuarine embayment. Mar. Geol. 59 : 85 103. doi:10.1016/0025 3227(84)90089 6 Waycott , M., C. M. Duarte, T. J. B. Carruthers, and others. 2009. Accelerating loss of seagrasses across the globe threatens coastal ecosystems. Proc. Natl. Acad. Sci. 106 : 12377 12381. doi:10.1073/pnas.0905620106 Wickham, H. 2007. Reshaping data with the reshape package. J. Stat. Softw. 21 : 1 20. Wickham, H., R. Francois, L. Henry, and K. MÃ¼ller. 2017. dplyr: A grammar of data manipulation. R package version 0.7.2. Willette, D. A., J. Chalifour, A. O. D. Debrot, and others. 2014. Continued expansion of the trans Atlantic invasive marine angiosperm Halophila stipulacea in the Eastern Caribbean. Aquat. Bot. 112 : 98 102. doi:10.1016/j.aquabot.2013.10.001 Williams, S. L. 1988a. Thalassia testudinum productivity and grazing by green turtles in a highly disturbed seagrass bed. Mar. Biol. 98 : 447 455. doi:10.1007/BF00391121 Williams, S. L. 1988b. Thalassia testudinum productivity and grazing by green turtles in a highly disturbed seagrass bed. Mar. Bio l. 98 : 447 455. Yahel, G., R. Yahel, T. Katz, B. Lazar, B. Herut, and V. Tunnicliffe. 2008. Fish activity: A major mechanism for sediment resuspension and organic matter remineralization in
160 coastal marine sediments. Mar. Ecol. Prog. Ser. 372 : 195 209. doi: 10.3354/meps07688 Yu, O. T., and G. L. Chmura. 2010. Soil carbon may be maintained under grazing in a St Lawrence Estuary tidal marsh. Environ. Conserv. 36 : 312 320. doi:10.1017/S0376892910000184 Ziegler, S., and R. Benner. 1998. Ecosystem metabolism in a subtropical, seagrass dominated lagoon. Mar. Ecol. Prog. Ser. 173 : 1 12. doi:10.3354/meps173001 Ziegler, S., and R. Benner. 1999. Dissolved organic carbon cycling in a subtropical seagrass dominated lagoon. Mar. Ecol. Prog. Ser. 180 : 149 160. doi:10.3354/m eps180149 Zieman, J. C., R. L. Iverson, and J. C. Ogden. 1984. Herbivory effects on Thalassia testudinum leaf growth and nitrogen content. Mar. Ecol. Prog. Ser. 15 : 151 158. doi:10.3354/meps015151
161 BIOGRAPHICAL SKETCH Robert A. Johnson was born in Madi son, Wisconsin and raised in McFarland, Wisconsin. He earned a Bachelor of Science degree in z oology and a Bachelor of Science degree in b iological a spects of c onservation from the University of Wisconsin Madison in 2010. During this time, he worked as an undergraduate research assistant in streams. Robert worked as a lab manager un der the supervision of Dr. Michael Pace at the University of Virginia from 2010 2013 investigating energy flow and resilience in lakes. In 2013, Robert began the pursuit of his Doctor of Philosophy under the direction of Dr. Karen Bjorndal in the Archie Carr Center for Sea Turtle Research at the University of Florida where he studied the ecological effects of green turtle grazing in Caribbean seagrass meadows. Robert compl eted his doctoral degree in 2019 with a degree in zoology.