FUNCTIONAL DIVERSITY OF SOUTHEASTERN UNITED STATES FISH COMMUNITIES By JOSHUA M. EPSTEIN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2016
2016 Joshua M. Epstein
To my wife, Samm
4 ACKNOWLEDGMENTS I acknowledge my advisor, Ben Baiser, and my committee members, Bill Pine, Christina Romagaosa, and Catherine Ph illips for their support and assistance during this project. Additionally, I thank Mark Scott and the South Carolina Department of Natural Resources for allowi ng me to analyze their fish sampling data, and Rua Mordechai, Amy Keister, Noel B urkhead, Stephen Walsh, and Amy Benson for their assistance in locating and/ or providing necessary data I also thank the South Atlantic Landscape Conservation Cooperative (SALCC) and the U.S. Fish and Wildlife Service for funding this project. Last but n ot least, I thank my family and friends for their continued support of my research and academic endeavors.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF F IGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ..................... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 11 Southeastern Unit ed States Fish Communities ................................ ...................... 11 Functional Diversity ................................ ................................ ................................ 12 Functional Diversity and Species Richness ................................ ............................ 13 Functional Diversity and the Environment ................................ ............................... 14 Our Research Study ................................ ................................ ............................... 16 2 METHODS ................................ ................................ ................................ .............. 18 Study Site ................................ ................................ ................................ ............... 18 Fish Sampling Data ................................ ................................ ................................ 18 Environmental Variables ................................ ................................ ......................... 19 Fish Traits Data ................................ ................................ ................................ ...... 21 Functional Diversity Metric ................................ ................................ ...................... 21 Random Forest ................................ ................................ ................................ ....... 23 3 RESULTS ................................ ................................ ................................ ............... 32 Functional Diversity Patterns ................................ ................................ .................. 32 Random Forest Regression Mode ls ................................ ................................ ....... 33 4 DISCUSSION ................................ ................................ ................................ ......... 43 Drivers of Functional Diversity ................................ ................................ ................ 43 Trophic SES FDis ................................ ................................ ............................. 43 Reproductive SES FDis ................................ ................................ .................... 46 Habitat SES FDis ................................ ................................ ............................. 48 Site C omparisons ................................ ................................ ............................. 49 Comparison with Other Studies ................................ ................................ .............. 50 Future Research and Conservation Implications ................................ .................... 54 5 REGIONAL ANALYSIS ................................ ................................ ........................... 57
6 Background ................................ ................................ ................................ ............. 57 Results ................................ ................................ ................................ .................... 59 Discussion ................................ ................................ ................................ .............. 60 Conclusion ................................ ................................ ................................ .............. 62 LIST OF REFERENCES ................................ ................................ ............................... 71 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 80
7 LIST OF TABLES Table page 2 1 Descriptions and summary statistics of environmental variables. ....................... 27 2 2 Trophic traits used to measure trophic functional diversity. ................................ 28 2 3 Reproductive traits used to measure reproductive function al diversity. .............. 29 2 4 Habitat traits used to measure habitat functional diversity. ................................ 30 3 1 Differences in mean tro phic SES FDis between South Carolina level III ecoregions. ................................ ................................ ................................ ......... 35 3 2 Differences in mean reproductive SES FDis between South Carolina level III ecoregions. ................................ ................................ ................................ ......... 35 3 3 Pearson correlation coefficients between South Carolina SES FDis metrics. .... 35 3 4 Mean (and standard deviation) SES FDis measurements acr oss South Carolina level III ecoregions. ................................ ................................ .............. 35 5 1 Pearson correlation coefficients between SALCC functional diversity metrics. .. 66 5 2 Pearson correlation coefficients between SALCC functional diversity metrics and environmental indicators. ................................ ................................ ............. 66 5 3 Summary of differences between South Carolina (loca l) and SALCC (regional) functional diversity analyses. ................................ .............................. 66
8 LIST OF FIGURES Figure page 2 1 Map of South Carolina Stream Assessment samplin g site s, river basins, and ecoregions ................................ ................................ ................................ ......... 31 3 1 Trophic SES FDis acr oss South Carolina sampling points. ................................ 36 3 2 Reproductiv e SES FDis across South Carolina sampling points. ....................... 37 3 3 Habitat SES FDis across South Carolina sampling points. ................................ 38 3 4 Trophic SES random forest variable importance plot ................................ ........ 39 3 5 Reproductive SES random forest variable importance plot ............................... 40 3 6 Habitat SES random forest va riable importance plot ................................ ......... 41 3 7 Comparisons between random forest predictions and actual data. .................... 42 5 1 Map of SALCC boundaries and HUC8s. ................................ ............................ 67 5 2 Trophic SES FDis across SA LCC HUC8s ................................ ......................... 68 5 3 Reproduct ive SES FDis across SALCC HUC8s. ................................ ................ 69 5 4 Habi tat SES FDis across SALCC HUC8s. ................................ .......................... 70
9 Abs tract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science FUNCTIONAL DIVERSITY OF SOUTHEASTERN UNITED STATES FISH COMMUN I TIES By Joshua M. E pstein August 2016 Chair: Ben Baiser Major: Wildlife Ecology and Conservation Species exhibit a variety of traits and trait combinations (i.e., associated with behavior, feeding reproduction ) some of which are unique and essential to ecosystem functio ning, and some which are redundant. is not always mirrored by traditional measures of species richness. Here we take a trait based ap proach to explore patterns of functional diversity in fish communities across South Carolina (l ocal), and the South Atlantic Landscape Conservation Cooperative ( SALCC; regional) We obtained species presence/absence data f rom the South Carolina Department of Natural Resources (local), and the MARIS and USGS Nonindigenous Aquatic Species databases (r egional), and scored trophic reproductive and habitat traits from the literature W e measured standardized functional diversity using t he functional dispersion metric for 323 wadeable streams (local) and 121 sub basins (regional) in the south eastern U nit ed S tates We also determined the environmental variables driving functional diversity measurements th rough R andom Forest modeling and Pearson correlations In general, we found high functional diversity throughout the eastern highland and Appalachian regi ons and l ower functional diversity in the Coastal Plain areas. We identified buf fer scale forest cover, elevation, ecoregion, and
10 conductivity as important predictors of different components of functional diversity across South Carolina, however, n one of o ur correlations between regional functional diversity estima tes and environmental variables were strong or significant. Our analysis provides foundation al knowledge of functional diversity patterns across the southeastern United States and highlights the variables to consider when including functional diversity assessments in conservation planning for these systems.
11 CHAPTER 1 INTRODUCTION Southeastern United States Fish Communities are parti cularly vulnerable to anthropogenic and climate induced influences because alterations related to ri ver hydrology, morphology, land use, and aquatic species introduction s affect both the aquatic environment and its associated drainage basin ( Allan 1995; Du dgeon et al. 2006 ; Carpenter et al. 2011). Additionally, aquatic species are sensitive to temperature and chemical fluctuations (Williamson et al. 2008), and increases in hydrologic variability and flooding (Rich ter et al. 1997; Wang et al. 200 1) can lead to changes in landscapes surrounding aquatic habitats, potentially resulting in erosion and habitat destruction ( Richter et al. 1997; Wang et al. 2001; Williamson et al. 2008). Despite having the richest freshwater fish fauna and highest number of endemic species in North America (Burkhead 2013 ), freshwater systems of t he s outheastern United States are susceptible to habitat loss and fragmentation, which are reducing the ir ecological integrity and threatening biodiversity (Jelks et al. 2008). Currently, the region has one of the greatest numbers of imperiled freshwater fish es in the nation, including many cyprinid s, ictalurid s, and percids (Jelks et al. 2008). The continued and future conservation of these species is threatened by climate change, and increas es in damming, channelization, land development, and ground water pumping associated with regional population growth (Warren et al. 2000 ; Jelks et al. 2008 ). To evaluate the impacts to biodiversity associated with these actions many freshwater fish resear chers have focused on quantifying species richness, or taxonomic diversity (Wang et al. 2001; Lessard & Hayes 2003; Daufresne & Boet 200 7; Macedo et al. 2014), however, this
12 may not be sufficient for a complete assessment of biodiver sity and ecosystem func tioning Functional Diversity From a basic ecological context, species represent a combination of evolutionary traits (i.e., associated with feeding, reproduction and habitat ), some of which are unique and essential to ecosystem functioning, while others are redundant (Fonseca & Ganade species richness (Schweiger et al 2007; Flynn et al 2009). In fact, species richness can remain relatively unchanged while functional div ersity decreases in some assemblages (Schweiger et al 2007) Additionally it is the functional attributes of species which primarily res pond to environmental changes and disturbance (Nystrom et al. 2000; Hooper et al. 2005; Suding et al. 2008 ; Swenson et al. 2012 ) and trait differentiation between species is what drives specific ecosystem services and functions (Tilman 2001; Cadottee 2011) These realizations have led to an expansion of diversity measurements beyond the taxonomic level to include quantit ative measures of functional diversity when assessing biodiversity and ecosystem function of a region or within ecological communities (Walker et al 1999; Petchey & Gaston 2002; Vill ger et al. 2008 ; Schmera et al 2009). A majority of functional diversit y researchers have focused on plants (Cowling et al. 1994; D az & Cabido 2001; De Bello et al. 2006; Swenson et al. 2012) due to the substantial documentation of their traits (Weiher et al. 1999 ; Cornelissen et al. 2003 ), the ability to easily monitor thei r changes across large scale environmental gradients (Swenson et al. 2012), and known links be tween specific traits and ecosystem functions ( Westoby et al. 2002; Kattge et al. 2011) More recent ly, functional diversity researchers
13 have entered the aquatic realm to analyze et al. 2014), marine (Gifford et al. 2009; Wiedmann et al. 2014), and estuarine (Mouillot et al. 2007; Vill ger et al. 2012) communities. Studies of fish functional diversity in lakes, rivers, an d streams have also become prevalent with applications focusing on two major themes: 1) exploring relationships between functional d iversity and species 2012; Vill ger et al 2013), and 2) identifying how the environment structures functional ase et al. 2012; Teresa & Casatti 2012; Buisson et al. 2013; Hitt & Chambers 2014). Rather than focusing solely on species, these studies give insight into how overall trait diversity, a recently overlooked component of biodiversity, changes spatially and under various environmental conditions in freshwater systems. Functional Diversity and Species Richness With the recent realization that assessing multiple biodiversity components is important, a s ubset of freshwater fish researchers have analyzed the cong ruence between observed species richness and functional diversity patterns. These studies ger et al. 2013) have all revealed similar patterns. I n the Grijalva Basin in Mexico, functional dive rsity of fish assemblages gene rally increased with species richness from uplands to lowlands (Pease et al. 2012). Additionally functional diversity measured as functional richness increased with species richness across 25 large river basins in Europe (Vil l ger et al. combinations in 124 lakes in Southern Finland, and also revealed that species richness and functional diversity are highly correlated across lakes. Taking a different approach
14 to define areas of high conservation priority in the Lower Colorado River Basin, Strecker et al. (2011) additionally considered phylogenetic diversity in defining areas of high conservation priority. The overlap among taxonomic, phylo genetic, and functional diversity under different conservation scenarios was analyzed, and it was revealed that 75% of catchments were high conservation priorities for all three biodiversity measurements due to the representation of all species, functional groups, and phylogenetic histories (Strecker et al. 2011) However, there are large portions of the landscape that were inconsistent in predicting high values for all three diversity measurements sugges ting that some communities may naturally exhibit dif fe rences in these three components of bio diversity (Strecker et al. 2011). Due to regional and spatial differences between multiple components of biodiversity ( Devictor et al. 2010; Strecker et al. 2011), and the need to understand ecosystem processes in a changing environment (Stuart Smith et al. 2013), it is well acknowledged that multiple dimensions of biodiversity should be considered when designing robust conservation strategies and managing biodiversity ( Devictor et al. 2010; Strecker et al. 2011; Stu art Smith et al. 2013). Functional Diversity and the Environment M any freshwater fish researchers have determine d, predict ed, or analyzed the major environmental or landscape influences of functional diversity, or sought to determine how functional diversi ty changes with climate 2010; Teresa & Casatti 2012; Buisson et al. 2013; Hitt & Chambers 2014; Pease et al. 2014) Buisson et al. (2013) measured functional diversity across 83 French river basins, and related these measurem ents to projected species range shifts in response to climate change. The species expected to suffer the greatest habitat losses due to
15 climate change were not always the species displaying the most unique combination of traits (Buisson et al. 2013), empha sizing that the most functionally unique species might not always be the most vulnerable Furthermore both local and regional scale environmental variables may be important drivers of functional diversity changes (Buisson et al. 2013). Similarly, in the G rijalva Basin, environmental variables at multiple scales structure d fish assemblages : L atitude, elevation, and climate were the major regional drivers of both functional and taxonomic diversity, whereas substratum type and presence of pool habitats domina te d at the local scale (Pease et al. 2014). Pool et al. (2010) also monitored the major drivers of functional diversity, but additionally distinguished between drivers of native vs. non native functional diversity in the Colorado River Basin. Variables ass ociated with hydrologic alteration, watershed land use, and climate explain ed a majority of the variation driving overall functional diversity. Greater dam density a nd upstream storage capacity were associated with non native functional diversity, while up stream land protection, fewer dams, and higher variation in precipitation support ed high levels of native diversity (Pool et al. 2010). Water chemistry and waterbody size have also been shown as important pre dictors of functional diversity ; a cross Southern Finland lakes, lake surface area, depth, total phosphorus, pH, and color were identified as important functional diversity drivers et al. 2009). I n the upper Paran River B asin in southeastern Brazil, Teresa and C asatti (2012) explored the relationships between functional patterns and both forest cover and mesohabitat type, which were important predictors of fish community structure. They revealed that taxonomic and functional diversity increase d in deforested meso habitats (Teresa & Casatti 2012) however, this was due to differences in species evenness.
16 Taking a temporal approach to a nalyze differences in taxonomic and functional diversity in the Guyandotte River Basin before and after exposure to mining, it was re vealed that both diversity measures were lower in fish assemblages when mining was present due to the adverse hydrological, physical, and chemical alterations (Hitt & Chambers 2014). These studies have provided foundation al knowledge of the important envir onmental variables impacting functional diversity patterns in different freshwater systems This information can be used to guide future research in assessing fish functional responses to disturbance and environmental variation Our Research Study South Ca rolina hosts a diverse fish fauna due to the variety of aquatic habitats across s ecoregions These include high gradient forested, Blue Ridge streams, large r Piedmont rivers, and low elevation blackwater channels of the Coastal Plain which are often associated with wetlands (SCDNR 2015). One hundred and forty six native freshwater fish species permanently or seasonally depend on these freshwater habitats for survival communities (SCDNR 2015) These communities are predominantly threatened by habitat degradation due to urbanization and agriculture, nonnative species introductions, and climate change ( Warren et al. 2000 ; Jelks et al. 2008; SCDNR 2015). In our study, we took a trait based approach and measure d trophic, reproductive, and habitat freshwater fish functional diversity across South Carolina wadeable streams All of o ur fish sampling data comes from a statewide evaluation of aquatic natural resources conducted by the South Carolina Department of Natural Resources (SCDNR) between 2006 and 2011 with standardized fish sampling protocols (Scott et al. 2009). To our knowledge, this is the first attempt to measure freshwater fish
17 functional diversity across an entire state in the continental United States using fish sampling data from a systematic sampling regime, in which stream reaches were randomly selected from a stratified list. Because this is a part of a state wide natural resources assessment, t he SCDNR dataset also c ontains a substantial num ber of environmental predictor variables with which we used R andom F orest (RF) regression models to explore relationships with standardized functional diversity estimates These include variables associated with instream habitat, geomorphology water qual i ty, network connectivity, and landscape cover at the 120 m buffer scale.
18 CHAPTER 2 M ETHODS Study Site South Carolina, located in the South Atlantic portion of the United States, is approximately 31,000 square miles, and includes portions of three major U.S. provinces : the Coastal Plain, Piedmont, and Blue Ridge Mountains The state also include s portions of the Broad, Catawba / Wateree, Saluda, Pee Dee, Santee, Ashepoo Combahee Edisto (ACE), and Savannah drainage basins, and the Piedmont, Blue Ridge, Middl e Atlantic Coastal Plain, and Southeastern Plain level III ecoregions ( Omernik 1987; Scott et al. 2009). The data for our study were collected from 397 stream reach sampling points within 324 wadeable freshwater st reams throughout South Carolina (Figure 2 1). Fish Sampling Data The fish sampling data used in our study come from a larger state wide evaluation of aquatic natural resources in South Carolina conducted by the SCD NR between 2006 and 2011. Fish sampling locations were selected at random from a pro babilistic framework designed by the SCDNR, and stream reaches were selected from a stratified list of ecobasins and stream si zes across the state (Scott et al. 2009 ). To ensure independence among samples (i.e., to account for spatial autocorrelation) sam ple locations were selected as to not share more than half of the drainage area of any downstream site. Fish es were collected from spring to fall via backpack electrofishers and dipnets, identified and released according to the standard protocols use d by SCDNR personnel for sampling fish es in wadeable streams (Scott et al. 2009).
19 A total of 101 fish species were collected throughout the state during the sampling period. Environmental Variables In addition to sampling fish es a suite of environmental and l andscape data were collected or quantified within stream reaches at all sampling locations following the protocols described in the South Carolina Stream Assessment Standard Operating Procedures (Scott et al. 2009) Instream variables included stream chann el width, depth, velocity, and inorganic and organic substrates (Table 2 1) Stream channel width was measured at distances of 0, 25, 50, 75, and 100 meters along a given sample reach, and t hese measurements were averaged to get the average transect width (Scott et al. requires two people to move upstream along the sample reach while moving in a random and subs trate were measured 50 times throughout this transect. Substrates were classified based on the following ; In organic substrate s were classified based on particle size, while organic substrates were classified by size and composition as either fine particula te organic matter (FPOM), coarse particulate organic matter (CPOM), fine woody debris (FWD), large woody debris (LWD), or aquatic vegetation (AV ; Scott et al. 2009 ). Water quality data included water temperature, diss olved oxygen, conductivity, pH and tur bidity (Table 2 1) all of which were measured using well calibrated portable meters (Scott et al. 2009) Additionally, temperature logger data (Table 2 1) was received through the deployment of temperature loggers within the sample reach at each site, whi ch were set to record the daily water temperature for 12 months (Scott et al. 2009). Geomorphological measurements incl uded slope, bankfull width, bankfull
20 depth sinuosity, and channel type (Table 2 1) Bankfull width and depth were measured according to the methods described by Harrelson et al. (1994) Sinuosity was characterized as a binary variable; a stream was either sinuous or not sinuous (Marion et al. 2015) Channel type was described as one of the following: not channelized with distinct single ch annel, channelized, or not channelized with swamplike and brai ded characteristics (Marion et al. 2015). L and cover within 120 m riparian buffer s was quantified by SCD N R personnel using ArcGIS 10.0 as was done in Marion et al. (2015) for the South Carolin a C oastal P lain (Table 2 1) Catchments encompassed the entire drainage area upstream of sites and r iparian buffers were designated as 120 m land areas adjacent to stream networks in defined catchments (Marion et al. 2015). T he percent of open water cover urban cover, forest cover, deciduous/mixed forest cover, evergreen forest cover, grass/shrub cover, agricultural cover, rowcrop cover, barren cover, and wetland cover was determined at this buffer scale Additionally, st ream network attributes were determ ined by the SCDNR (Marion et al. 2015) and included distance to the HUC8 main stem, and the number of dams, impoundments, ro ads, and tributaries both upstream and downstream of the main stem (Table 2 1) In this study we explored the relationships between environmental predictor variables and freshwater fish functional diversity in a four step process: 1. Determined trophic, reproductive, and habitat traits associated with the sampled fish species. 2. Measured standardized functional diversity values for each se t of traits across sites. 3. Removed correlated environmental predictor variables and fit preliminary random forest (RF) model s for each set of standardized functional diversity measurements
21 4. Further removed the least important predictor variables in the pre liminary RF models and fit final RF models. Fish Traits Data We created a database detailing traits related to trophic reproductive, and habitat ecology for each of the 101 fish species sampled across South Carolina stream reach sampling points ( For dat abase see: http://ufdc.ufl.edu/l/IR00007844/00001 ) Based on the Virginia Tech Fish Traits Database ( Frimpong & Angermeier 2009) we designated and coded 11 trophic traits (Table 2 2) as binary descriptors for each species. I f a fish species was found to h ave a given trait (i.e., be nthic feeding species), we assigned this as a 1 and a species without this trait we assigned a 0. The same was done for reproductive and habitat traits; for the reproductive category we designated 19 traits (Table 2 3) and for th e habitat category we designated 25 traits (Table 2 4 ). We have obtained data on specific trait attributes for these sampled fish species using the Virginia Tech FishTraits Database ( Frimpong & Angermeier 2009 ) primar y literature, USFWS reports, state or regional fish identification texts and expert opinion (For a complete list of trait data sources see: http://ufdc.ufl.edu/l/IR00007844/00001 ) Functional Diversity Metric Based on the sampled species and their associated traits, we measured trophic repr oductive, and habitat functional diversity separately for fish communities across stream sampling p oints. We removed s ites from our analysis that were dry during the sampling period or in which less than two species of fish were collected from our analysis due to the inability to measure functional diversity ( i.e., totaling 13 sites). We additionally removed four sites from our habitat functional diversity analysis due to a substantial amount of missing trait data for the present fish species. We calculated
22 f unctional diversity using the functional dispersion metric ( FDis; Lalibert & Legendre 2010 ) (R Core Development Team 2012) We chose the FDis metric because it is uncorrelated with species richness, is not stro ngly influe nced by outliers, ( Lalibert & Legendre 2010 ) and it can be calculated using any distance or dissimilarity measure ( Anderson et al. 2006) We measured FDis using Gower distance s measured from our trait data matrices which can handle any trait type and mi ssing trait data (Gower 1971) FDis measures the average distance of all species in a community from the community centroid in multidimensional trait space ( Lalibert & Legendre 2010 ) If all species have the same value or classification for each trait th en they will all lie exactly on the centroid and the average distance from the centroid wi ll be zero (i.e., there is no functional diversity ). On the other hand, when there is large variation in traits, species will be displaced fr om the community centroid and functional diversity will be high We compared observed FDis values to null distributions of FDis values to calculate standard effect size (SES) FDis which standardize s functional diversity measurements to remove potential bias associated with decre asing variance surrounding functional diversity measurements as species richness increases (Swenson 2014 ). To generate our null distributions of FDis values, w e model to randomize our trophic, reproductive, and habitat trait da ta matrices (Swenson 2014). In doing this, we shuffle d the names of the species on our trait data matrices and generated 999 random FDis values for each of our stream reach sampling points We then subtracted the mean of each null distribution from the obs erved FDis value for each stream reach and divided this by the standard deviation of the null distribution to
23 generate SES FDis values (Equation 2 1 ). Positive SES FDis value s are associated with higher functional diversity than expected compared to the n ull distribution, whereas negative SES FDis value s are associated with lower functional diversity compared to the null distribution (Swenson 2014) Rather than using raw measures of functional diversity we used these SES FDis estimates as the response var iable in our statistical models to deter mine environmental drivers of functional diversity After we measured SES FDis for our three trait categories, we conducted a one way ANOVA between sampling points to determine if mean SES FDis values were different among South Carolina level III ecoregions (Omernik 1987). We then used to determine which level III ecoregions had significantly different mean SES FDis values (2 1) Random Forest We used RF regression to identify the environmental variables that are most i nfluential in d riving South Carolina freshwater fish trophic reproductive, and habitat SES FDi s patter n s using functions in the v 3.3.2 (R Core Development Team 2012) RF is a machine learning technique that can be used for both classification and regression and uses an algorithm to reveal the major predictor variables d riving a response variable (Breiman 2001). The model is an extension of bagging methods (Breiman 1996 ; Liaw & Wiener 2002 ) in which an ensemble of trees is built, and each tree is independently constructed using a subset of the original dataset. RF operate s using two levels of randomness; first, by generating n tree trees from randomly selected bootstrap samples from the original data, and second, by splitting the
24 tree at each node based on the best split from among a random sample of the predictor variables m try (Liaw & Wiener 20 02) Approximately one third of the original data called the of bag (OOB) data is held back from inclusion in RF construction to determine model prediction error (i.e., mean squared error (MSE) for regression) and variable im portance. Variable importance is determined during the construction of each tree based on the mean decrease in accuracy (%IncMSE). For each variable, t his metric measures the difference in MSE estim ates between the original OOB data and the OOB data in whi ch a given variable is randomly permuted ( Breiman 2001; Liaw & Wiener 2002). If there is a large difference in error estimates when a given variable is randomly permuted, this variable is important. Based on this, e ach generated tree gets a vot e on which p redictor variable is the most important for splitting observations based on and the final most important variable has the highest a verage number of votes among the entire ensemble of trees (Liaw & Wiener 2002). The SCDNR database contained missing environ mental data for some sites, and we imputed m issing val ues in these predictor variables using the proximity matrix from our RF models (i.e., using the rfImpute function ) The proximity matrix measures the similarity between observations by measuring the fra ction of trees in which two observations are assigned to the same terminal node during RF construction. Observations that are more similar to one another will have proximities close to 1, and will be placed in the same terminal node more often than those t hat are different from one another and have proximities closer to 0 (Liaw & Wiener 2002). Missing values for continuous predictors are determined using the weighted average of the non missing observations, where the weights are the RF proximities; missing values for categorical
25 predictors are imputed as the value from the category with the largest average proximity (Breiman 2003). Because normalization is not a model assumption, we did not transform or standardize our data for RF, nor did we remove outliers because they have little to no influence in the RF algorithm ( Breiman 2001) Additionally we screened our environmental variables with a two step process. First we eliminated highly correlated ( r > 0.70) environmental and landsca pe variables based on Pearson correlation coefficients because RF models have known biases for selecting highly correlated variables as important ( Strobl et al. 2008 ). Average velocity and average depth were highly correlated with the standard deviation of velocity and depth, respectively ( r = 0.85 and r = 0.72). The percent of rowcrop cover at the buffer scale was highly correlated with the percent of agricultural cover at the buffer scale ( r = 0.77), and the percent of forest cover at the buffer scale was highly correlated with the percent of deciduous/m ixed forest cover at the buffer scale ( r = 0.85 ) and wetland cover at the buffer scale ( r = 0.76) Therefore, we removed the standard deviations of velocity and depth, and buffer variables associated with rowcrop cover, deciduous/mixed forest cover and wetland cover from our analysis. Next, we ran a preliminary RF model for each set of SES FDis measurements and removed the least important variables to maximize prediction accuracy as was done in Oliveira et al. (2012) and Marion et al. (2015) We removed all variables with a %IncMSE less than 10%, and then ran the RF model s again using the remaining variables. Using the randomForest function, we ran three RF models in which trophic, reproductive, and habit at SES FDis estimates were separately analyzed as the response variable. For these models we built 5,000 trees as i t has been shown that
26 using many trees is necessary to get stable estimates of variable importance and proximity ( Grimm et al. 2008; Liaw & Wiener 2002) We split the tree s at each node based on the best split from among the number of predictor variables divided by three As recommended, we ran our models with double and half of this value (Liaw & Wiener 2002), and the total variance explained was either unchanged or reduced Next, w e determined v ariable importance based on the %IncMSE We then determined model fit using th e percent of variance explained (pseudo R 2 ), which is calculated as one minus the MSE of the OOB data divided by the varian ce of the response variable observations (Equation 2 2 ) We also analyzed linear regression models relating observed and RF predicted measurements (Liaw & Wiener 2002). (2 2 )
27 Table 2 1. Descriptions and summary statistics of environmental variables. Abbreviation Description Mean Range Ecoregion Categorical (Blue Ridge, Piedmont, Southeastern Plains or Middle Atlantic Coastal Plain) NA NA Elevation Average catchme nt elevation (m) 321.24 3 1101 Velocity_M Average velocity (m/s) 0.11 0 0.45 Velocity_SD Standard deviation of velocity (m/s) 0.08 0 0.48 Depth_M Average depth (m) 0.24 0.04 0.64 Depth_SD Standard deviation of depth (m) 0.13 0.03 0.32 Width_M Average width (m) 3.93 0.38 14.58 Per_AV Percent aquatic vegetation 0.07 0 4.38 Per_FPOM Percent fine particulate organic matter 0.05 0 1.00 Per_CPOM Percent coarse particulate organic matter 0.20 0 0.88 Per_FWD Percent fine woody debris 0.10 0 0.49 Per_LWD P ercent large woody debris 0.11 0 0.40 Per_BR Percent bedrock 0.03 0 0.44 Per_Sand Percent sand 0.26 0 0.80 Avg_Daily_Mean_T emp Average daily mean temperature (C) 23.43 19.05 29.85 Avg_Daily_ Var Average daily mean temperature variance 1.53 0.15 6.58 A vg_Daily_ SD Average daily mean temperature standard deviation 1.19 0.39 2.56 Temp Temperature (C) 20.97 7.92 33.34 DO Dissolved oxygen (mg/L) 6.62 0 13.79 Cond Conductivity (S/cm) 102.32 11.00 708.00 pH pH 6.90 4.85 8.96 Turbidity Turbidity (NTU) 10 .52 0.84 245.10 Slope Slope 0.02 0 6.10 Wbkf Bankfull width (ft) 34.52 3.88 162.1 Dbkf Bankfull depth (ft) 3.27 0.30 10.20 Sinuosity Binary (sinuous or not sinuous) NA NA Channel_Type Categorical (not channelized with distinct single channel, channeli zed, or not channelized with braided characteristics NA NA Calculated_ WS_Area Catchment area (km 2 ) 32.26 0.17 154.13 BP_06_OPENWATER Percent of buffer under open water 0.01 0 0.12 BP_06_URBAN Percent of buffer under urban cover 0.07 0 0.79 BP_06_DECIDU OUS/MIXED Percent of buffer under deciduous/mixed forest cover 0.24 0 0.96 BP_06_EVERGREEN Percent of buffer under evergreen forest cover 0.17 0 0.69 BP_06_GRASSSHRUB Percent of buffer under grass/shrub cover 0.10 0 0.40 BP_06_ROWCROP Percent of buffer under rowcrop cover 0.06 0 0.58 BP_06_AGRICULTURE Percent of buffer under agricultural cover 0.13 0 0.84 BP_06_WETLAND Percent of buffer under wetland cover 0.27 0 0.90 BP_06_BARREN Percent of buffer under barren cover 0 0 0.09 BP_06_FOREST Percent of buffer under forest cover 0.41 0 1.00 DIST_MS Distance to HUC8 main stem (m) 11,258 323 70,880 DS_DAMS_MS Number of downstream dams on main stem 0.26 0 5 DS_IMPOUND_MS Number of downstream impoundments on main stem 0.42 0 4 DS_ROADS_MS Number of downst ream roads on main stem 3.16 0 12
28 Table 2 1. Continued Abbreviation Description Mean Range DS_TRIBUTARIES Number of downstream tributaries on main stem 4.35 0 33 US_DAMS_MS Number of upstream dams on main stem 0.42 0 4 US_IMPOUND_MS Nu mber of upstre am impoundments on main stem 0.72 0 8 US_ROADS_MS Number of upstream roads on main stem 3.54 0 14 US_TRIBUTARIES Number of upstream tributaries on main stem 3.22 0 19 Table 2 2. Trophic traits used to measure trophic functional diversity. Trait Co de Trait NONFEED Adults do not feed BENTHIC Benthic feeder SURWCOL Surface or water column feeder ALGPHYTO Algae or phytoplankton MACVASCU Any part of macrophytes and vascular plants DETRITUS Detritus or unidentifiable vegetative matter INVLVFSH A quatic and terrestrial invertebrates, and larval fish FSHCRCRB Larger fishes, crayfishes, crabs, frogs, etc. BLOOD Blood EGGS Eggs OTHER Other diet components distinct from the preceding classes Note : Trophic traits obtained from Frimpong, E. A., & An germeier, P. L. (2009). Fish traits: a database of ecological and life history traits of freshwater fishes of the United States. Fisheries 34 487 495
29 Table 2 3. Reproductive traits used to measure reproductive functional diversity. Trait Code Trait GUARD Guarder NONGUARD Non guarder BEARER Bearer OPNSSPWNR Open substratum spawner BRDHDR Brood hider SCHOOSER Substratum chooser NSTSPWNR Nest spawner PELAGO Pelagophils LPELAGO Litho pelagophils LITHORGS Lithophils (rock, gravel, s and) LITHOSM Lithophils (silt, mud) PHYTOLITHO Phytolithophils PHYTO Phytophils PSAMMO Psammophils SPELEORR Speleophils (rock cavity, roof) SPELEO CG Speleophils (cavity generalist) SPELEOBB Speleophils (bottom burrowers) POLYPHIL Polyphils ARIADON O Ariadnophils Note : Reproductive traits obtained from Frimpong, E. A., & Angermeier, P. L. (2009). Fish traits: a database of ecological and life history traits of freshwater fishes of the United States. Fisheries 34 487 495
30 Table 2 4. Habitat trait s used to measure habitat functional diversity Trait Code Trait MUCK Muck substrate CLAYSILT Clay/ silt substrate SAND Sand substrate GRAVEL Gravel substrate COBBLE Cobble substrate BOULDER Boulder substrate BEDROCK Bedrock substrate VEGETAT Aqua tic vegetation DEBRDETR Organic debris or detrital substrate LWD Large woody debris PELAGIC Open water PREFLOT Lotic and lentic systems but more often lotic PREFLEN Lotic and lentic systems but more often lentic LARGERIV Medium to large river SMALLR IV Stream to small river CREEK Creek SPRGSUBT Spring or subterranean water LACUSTRINE Lentic systems POTANADR Potamodromous or anadromous LOWLAND Lowland elevation UPLAND Highland elevation MONTANE Mountainous physiography SLOWCURR Slow current MO DCURR Moderate current FASTCURR Fast current Note: Habitat traits obtained from : Frimpong, E. A., & Angermeier, P. L. (2009). Fish traits: a database of ecological and life history traits of freshwater fishes of the United States. Fisheries 34 487 495
31 Figure 2 1. Map of South C arolina Stream Assessment sampling sites, river basins, and ecoregions. Note : Map Obtained from the South Carolina Department of Natural Resources (SCDNR) South Carolina Stream Assessment.
32 CHAPTER 3 RESULTS Functional D iversity Patterns SES FDis measurements varied across the South Carolina landscape within level III ecoregions Our ANOVA results depict that South Carolina ecoregion had a significant effect on m ean trophic [F (3,380) = 29.04 p = 2.2x10 16 ] and reproduct ive [F (3,380 ) = 15.37 p = 1 .9 x10 9 ] SES FDis measurements, but not on habitat measurements [F (3,376) = 0.21 p = 0.89 ] For the trophic category, all ecoregions were significant ly different from one another ( p < 0.05) expect the Blue Ridge with the Pied mont ( p = 0.97 ) and Southeastern Plains ( p = 0.10 ) and the Southeastern Plains with the Middle Atlantic Coastal Plain ( p = 0.06 ; Table 3 1 ) For the repr oductive category, all ecoregions were significantly different from one another ( p < 0.05 ) except the Piedmont with the Blue Ridge ( p = 0.29 ) and Middle Atlantic Coastal Plain ( p = 0.05 ; Tab le 3 2 ) Our t rophic, reproductive, and habitat SES FDis measurements were uncorrelated with o ne another across sampling points (Table 3 3 ). We show ed that t rophi c SES FDis decreased from the Blue Ridge to th e Middle Atlantic Coastal Plain r egion (Figure 3 1 ), and m ean SES FDis was greatest in the Blue Ridg e region ( mean = 0.63 standard deviation = 0.40 ) and lowest in th e Middle Atlantic Coastal Plain region ( mean = 0 .4 0 standard deviation = 0.94 ; Table 3 4 ) Reproductive SES FDis increased from the Blue Ridge to the Southeastern Plains region (Figure 3 2 ), and m ean SES FDis was greatest in the Southeastern Plains region ( mean = 0.76, standard deviation = 0.62 ) and lo west in the Blue Ridge region (mean = 0.19 standard deviation = 0.74 ; Table 3 4 ). We also showed that h abita t SES FDis generally decreased from the Blue
33 Ridge to th e Middle Atlantic Coastal Plain region (Figure 3 3 ), and m ean SES FDis was greatest in the Piedmont region ( mean = 1.16 standard deviation = 0.98 ) and lowest in th e Middle Atlantic Coastal Plain region ( mean = 2.55 standard deviation = 1.51 ; Table 3 4 ). Random Forest Regression Models We built RF regression models to determine the main pre dictors of trophic, reproductive, and habitat SES FDis For our trophic SES FDis model, the perc entage of variance explained was 18.05 % ( MSE = 0.622 ). In this model, our results indicate that the most important variables were buffer scale forest cover (pos itive association with SES FDis ) elevation (positive association with SES FDis ) and ecoregion (categorical variable) with 47.64, 43.90, and 38.11 %IncMSE, respectively ( Figure 3 4) For our reproductive SES FDis model, the percentage of variance explaine d was 24. 4 5 % ( MSE = 0.347 ). In this model, our results indicate that the most im portant variables were elevation (negative association with SES FDis ) and conductivity (negative association with SES FDis ) with 37.17 and 33.56 %IncMSE respectively (Figure 3 5) Our habitat SES model only contained three variables after r emoving those below 10 %IncMSE. These consisted of the number of downstream dams on the main stem buffer scale evergreen forest cover, and the number of downstream tributaries on the main st em with 25.63, 21.14, and 13.45 %IncMSE in the final model respectively ( Figure 3 6 ). This model did not explain any o f the overall variation We used linear regression s to relate our observed and RF predicted measurements for each trait category to evalu ate model fit (Figure 3 7 ). Our trophic, reproductive, and habitat models explaine d 18.0 % ( R 2 = 0.180 residual standard error = 0.7897 on 382 df and F statistic = 85.19), 24.4 % ( R 2 = 0.244 residual standard error = 0.5898 on 382 df and F statistic = 124. 7 0 ), and 0% ( R 2
34 = 1.17 x 10 3 residual standard error = 1.449 on 374 df and F statistic = 0.561 ) of the overall variation respectively. Our trophic and reproductive models were significant ( p < 0.05) but our habitat model was not ( p = 0. 45 )
35 Table 3 1. Differences in mean trophic SES FDis between South Carolina level III ecoregions. Ecoregion 1 Ecoregion 2 Mean Difference Standard Error Middle Atlantic Coastal Plain Blue Ridge 1.03* 0.44 Piedmont Blue Ridge 0.14 0.43 Southeastern Plains Blue Ri dge 0.75 0.43 Piedmont Middle Atlantic Coastal Plain 0.89* 0.14 Southeastern Plains Middle Atlantic Coastal Plain 0.28 0.14 Southeastern Plains Piedmont 0.61* 0.12 Note: The mean difference is significant at the p < 0.05 level. Table 3 2. Differ ences in mean reproductive SES FDis between South Carolina level III ecoregions. Ecoregion 1 Ecoregion 2 Mean Difference Standard Error Middle Atlantic Coastal Plain Blue Ridge 0.69 0.36 Piedmont Blue Ridge 0.47 0.35 Southeastern Plains Blue Ridge 0.94* 0.35 Piedmont Middle Atlantic Coastal Plain 0.22* 0.11 Southeastern Plains Middle Atlantic Coastal Plain 0.26* 0.12 Southeastern Plains Piedmont 0.48* 0.10 Note: The mean difference is significant at the p < 0.05 level. Table 3 3 Pearson corre lation coefficients between South Carolina SES FDis metrics. Trophic Reproductive Habitat Trophic 1.00 0.0 3 0.14 Reproductive 0.03 1.00 0.1 3 Habitat 0.14 0.1 3 1.00 Note: No significant correlations. Table 3 4 Mean (and standard deviation) SES F Dis measurements across South Carolina level III ecoregions. Trophic Reproductive Habitat Blue Ridge 0.63 (0.40 ) 0.19 (0.74 ) 1.35 (1.65 ) Piedmont 0.49 (0.68 ) 0.28 (0.69 ) 1.16 (0.98 ) Southeastern Plains 0.1 2 (0.81 ) 0 .76 (0.6 2 ) 2.42 (1.49 ) Middle A tlantic Coastal Plain 0.4 0 (0.95 ) 0.51 (0.6 0) 2.55 ( 1.51 )
36 Figure 3 1. T rophic SES FDis across South Carolina sampling points. Warm colors indicate high SES FDis and cool colors indicate low SES FDis. No sampling points were in the Southern Coastal P lain.
37 Figure 3 2. R eproductive SES FDis across South Carolina sampling points. Warm colors indicate high SES FDis and cool colors indicate low SES FDis. No sampling points were in the Southern Coastal Plain.
38 Figure 3 3 H abitat SES FDis across South Carolina sampling points. Warm colors indicate high SES FDis and cool colors indicate low SES FDis. No sampling points were in the Southern Coastal Plain.
39 Figure 3 4. Trophic SES random forest variable importance plot. Variables with the largest %IncMSE are the most important drivers of SES FDis. Trophic SES FDis was positively associated with BP_06_FOREST and elevation and this model explained 18.05% (MSE = 0.622) of the overall variation
40 Figure 3 5. Reproductive SES random forest variable importan ce plot. Variables with the largest %IncMSE are the most important drivers of SES FDis. Reproductive SES FDis was negatively associated with elevation and conductivity and this model explained 24.45% (MSE = 0.347) of the overall variation
41 Figure 3 6. Habitat SES random forest variable importance plot. Only three variables remained after removing unimportant variables from our initial RF model. This model did not explain any of the variation driving habitat SES FD is.
42 Figure 3 7. Comparisons between random forest predictions and actual data. A) Trophic SES FDis ( R 2 = 0.180 p < 0.05 ) B) Reproductive SES FDis ( R 2 = 0.244 p < 0.05 ) C) Habitat SES FDis ( R 2 = 1.17 x 10 3 p = 0.45 )
43 CHAPTER 4 DISCUSSION Drivers of Functional Diversity Environmental variables associated with hydrologic alteration (Pool et al. 2010) instream habitat (Teresa & Casatti 2012; Pease et al. 2014), land cover (Pool et al. 2010; Teresa & Casatti 2012) water chemistry and climate (Pool et al. 2010; Pease et al. 2014) have been shown to be important drivers of freshwater fish functional diversity in various geographic locations. Across South Carolina, we found that 120 m buffer scale forest cover, el evation, and ecoregion were important for trophic SES FDis, and elevation and conductivity were important for repro ductive SES FDis Trophic SES FDis For our trophic SES FDis model, buffer scale forest cover was the most important predictor, and had a posi tive relationship with SES FDis Forest cover has intuitive links with the success and health of freshwater ecosystems in the temperate Southeast because reduced riparian forest cover is associated with degraded stream ecosystems. Streams with reduced fore sts exhibit raised water temperature (Johnson & Jones 2000), stream narrowing (Sweeney et al. 2004), and increased nutrient and sediment input (Lowrance et al. 1997; Jones et al. 2001), all of which may have detrimental impacts on stream ecosystems. Alloch thonous nutrient input is also reduced, which is vital for the proper functioning of stream processes (Gessner 2002). The effects of deforestation are often exacerbated when the cleared land is developed for agriculture or urbanization due to increases in nutrient and sedimentation input All of the degradation associated with forest removal can induce stream homogenization ( Rahel 2000; Scott & Helfman 2001), which inhibits the opportunity for unique trophic
44 niches which, when present and filled, support fu nctionally unique species. Subsequently, s treams with cleared forest cover a re often characterized by more cosmopolitan species with shared generalist traits suitable for inhabiting these reduced stream systems (Jones et al. 1999; Burcher et al. 2008). As an example of forest importance for trait diversity, snags (i.e., submerged wood substrates) are a unique habitat structure use d by fish es that are provided by fallen forest trees or branches (Benke et al 1984). These snags not only serve a s refuge, but a lso support high invertebrate density (Benke et al. 1984), and provide food availability for fish es Aside from snags, forests and trees could support a number of other traits (i.e., through the presence of terre strial invertebrates leaf litter, shade, c o ol water, etc.) The use of these unique habitat types could support higher levels of trophic functional diversity compared to where they are absent with minimal forest cover. Elevation was ranked as the second most important variable for trophic SES FDis. This pattern of trophic SES FDis (Figure 3 1) displays an increasing functional diversity trend moving from the Middle Atlantic Coastal Plains region (i.e., low elevation) to the Blue Ridge region (i.e., high elevation ). Elevation often implies greater hab itat complexity; at higher elevations streams contain a variety of habitats including riffles, runs, pools and waterfalls, and a variety of substrate types whereas lowland streams have less of these structures and are more depositional (Wallace et al. 19 92; Scott & Helfman 2001; SCDNR 2015) Additionally, the Blue Ridge physiographic regi on has more intact forest ( Wallace et al. 1992 ; SCDNR 2015), and forest cover is associated with more heterogeneous habitats and st ream fish assemblages (Scott & Helfman 2001) The diverse and specialized habitats of the highlands are used by a myriad of
45 biota (Scott & Helfman 2001) which create the opportunity for diverse trophic interactions, and functionally unique combinations of trophic traits. For these reasons, we speculate that high trophic functional diversity was linked with high elevation. Ecoregion was ranked as the third most important predictor of trophic functional diversity. Because the level simila r explanations account for both of these patterns Again, our trophic map depicts clear cut differences in functional diversity based on ecoregion (and by default, from e levation that likely account for the observed functional differences. These include land use and geological differences, which can influence differences in species assemblages and habitats (Hocutt & Wiley 1986; Utz et al. 2010). Anthropogenic activity has historically been greater in Coastal Plain systems than Blue Ridge systems including deforestation, agriculture, and infrastructure development during the 18 th and 19 th centuries (Marion et al. 2015), and these activities persist (Hardison et al. 2009; Utz et al. 2010). Actions associated with development and land use modifications have negative impacts on stream fish communities and habitats including increased runoff (Wang et al. 2001), hydrological alterations (Paul & Meyer 2001; Walsh et al. 2005), chan nelization, and temperature (Kaushal et al. 2010) and chemical (Paul & Meyer 2001; Wang et al. 2001; Walsh et al. 2005) fluctuations. These actions lead to the simplification of stream ecosystems which have reduced resources available, and species unable t o thrive under these modifications are often replaced by cosmopolitan generalists (Jones et al. 1999; Scott & Helfman 2001; Helms et al. 2011), which exhibit similar traits suitable for consuming these available resources. Therefore, streams with
46 more seve re anthropogenic modifications would have lower trophic functional diversity. The Blue Ridge on the other hand, is much less disturbed compared to the Coastal Plain with more intact forest and less urbanization (SCDNR 2015), and these systems would theref ore display higher functional diversity estimates. Additionally, the Appalachian highlands of the Blue Ridge are geologically older compared to the Coastal Plains due to their exclusion from sea level inundations and glacial events ( Hocutt & Wiley 1986 ; St range & Burr 1997 ). This long period of geologic stability has likely contributed to the high levels of endemism and specialism present (Mayden 1987 ; Burkhead et al. 1997 ) as well as different mesohabitat structures (Wallace et al. 1992; Scott & Helfman 2 001). habitat heterogeneity is due to elevation, but also because these streams have been free from major past geologic disturbances (Hocutt & Wiley 1986). T herefore, there are many groups of taxa that have simultaneously persisted in these habitats throughout time and again, this may have allowed for diverse trophic interactions and functionally unique trait combinations to develop. Reproductive SES FDis Elevation was the most important predictor of reproductive SES FDis, but the dir ec tion of the relationship was opposite of the trophic SES FDis trend (i.e., positive for trophic and negative for reproductive). This pattern of reproductive SES FDis (Figure 3 2) displays a decreasing functional diversity trend moving from the Middle Atl antic Coastal Plains region (i.e., low elevation) to the Blue Ridge region (i.e., high elevation). This was surprising because we would expect the reproductive pattern to follow the trophic pattern due to the complexity of substrates and habitats availabl e for nesting and spawning at higher elevations. However, we believe that the reproductive traits that we
47 considered for our analysis could be responsible for this trend. Our reproductive traits correspond to the reproductive guilds described by Balon (197 5) which define the mode of reproduction through the characterization of spawning, preferred substrate, and degree of parental care. By nature of the structure of these guilds, any given species can only be associated with up to three traits, many of which must occur together (i.e., phytophils and ariadnophils can only occur with nest spawning species). Contrastingly, a given species can eat a combination of different things (i.e., detritus, plants, small fish, invertebrates), and thus fulfill multiple trop hic traits, and there are no underlying restrictions on which trophic traits can occur together. This idea could have limited the potential for diversity of reproductive traits in the highlands because many highland species are reproductively specialized, for example, to spawn on rocky bottoms (Scott & Helfman 2001), and therefore the breadth of potential trait combinations capturing this specialization is narrow. These highland species might reproduce in functionally unique ways that can simply not be capt ured by these reproductive guilds, and including other reproductive traits in our analysis, such as egg size, female age at maturation, maximum fecundity, or the use of visual or acoustic cues might provide greater opportunity for differenc es in traits to be represented. If our reproductive traits are indeed simply characterizing reproductive fish guild s, then our measurem ents might not be providing valuable information regarding the overall e cological integrity of these systems H owever at this time we c annot be certain that our pattern is not due to an actual trend representing state wide reproductive trait diversity in freshwater streams Conductivity had a negative association with reproductive SES FDis and was found to be the second most important va riable assessed Conductivity is an indicator of
48 salinity and ion concentration in water Reproduction in fishes is sensitive to conductivity, as sperm motility can be negatively affected by changes in osmotic concentration (Alavi & Cosson 2006). Conductiv ity can also be used to detect total dissolved solid s in the aqueous environment (TDS ; Wang & Yin 1997 ), which often constitute nutrient input in str eams. Conductivity has been shown to increase after deforestation (Likens et al. 1970) urbanization (Wang & Yin 1997 ; Paul & Meyer 2001 ) and with sewage pollution (Daniel et al. 2002) all of which permit nutrients to enter streams in different ways which can be harmful to fish. The added nutrients can also cause algal blooms, which can ultimately lead to anox ic conditions. Ultimately, reduced sperm motility and an excess of nutrients could result in reduced functional diversity if species no longer have the appropriate spawning conditions Functionally unique spawners m ight be replaced by generalist spawners ( often nonindigenous species ; Scott & Helfman 2001 ; Helms et al. 2011 ) which are able to tolerate and reproduce under these modified conditions, and these species likely share simi lar reproductive traits (i.e., sediment tolerant polyphils ; Helms et al. 2011 ) Based on this we would expect that lower conductivity should be evident in streams with high reproductive functional diversity. Habitat SES FDis O ur habitat SES RF model did not explain any of the overall variation driving functional diversity across South Carolina First, although we considered many habitat components as environmental predictors, we did not include fine scale mesohabitat variables in our models, such as the numbers of riffles, pools, and runs, which have been shown to be important in the functional composition of fish communities (Teresa & Casatti 2012; Pease et al. 2014) and are largely present in the highlands ( Wallace et al.
49 1992; Scott & Helfman 2001). These presence of these habitat features might be functionally important because they provide the opportunity for unique habitat niches, which might increase local functional diversity. Also we speculate that the lack of variation explained could partially be due to the incompleteness of our habitat traits database which contained fa r more missing data than our tro phic and reproductive databases This was due to our inability to determine all 25 habitat traits for all of our sampled species. Our methodology is able to deal with missing trait valu es, however, these diminish the accurac y of variable selection during the splitting of trees at each node in our RF models, potentially reducing the total amount of explained variation. Site Comparisons To explore the variables identified by our RF models as important predictors of functional d iversity we id entified two sites with high and low trophic SES FDis estimates T he South Pacolet R iver in the Piedmont region has high trophi c SES FDis (FD is = 1.1 71) and the t ributary to Waccamaw River in the Middle At lanti c Costal Plains region has low trophic SES FDis (FD is = 3.449 ). The South Pacolet River located in the Broad River Basin was characterized by high forest cover (9 7 .1 %) and high elevation (976 m ). The nine species sampled in this stream reach were Moxostoma rupiscartes Nocomis leptoce phalus Semotilus atromaculatus Notropis chlorocephalus (endemic) Noturus insignis Lepomis auritus Clinostomus funduloides and Etheostoma thalassinum (endemic) These species represented a diverse array of trophic traits; o nly two species shared the s ame trait combination (i.e., N. leptocephalus and S. atromaculatus ) Otherwise, the remaining species two of which are endemic, exhibited a unique comb ination of trophic traits. For example, only one species was characterized as a consumer of eggs (i.e., C. funduloides ) Additionally, all traits were
50 represented except the NONFEED ( nonfeeding adults), BLOOD (or parasitic lampreys that feed mainly on blood), and OTHER categories expressing the variety of present trait combinations The tributary to the Wac camaw River in the Pee Dee River Basin, on the other hand, had much lower forest cover (19.6%) and elevation ( 11m). Six of the species sampled all shared the same exact combination of trophic traits (i.e., Amia calva Umbra pygmaea Micropterus salmoides Acantharchus pomotis Aphredoderus sayanus and Chaenobryttus gulosus all were characterized as benthic, surface/water column, invertebrate/larval fish, and large fish feeders in our database). The remaining three species ( Lepomis macrochirus Centrarchus macropterus and Gambusia holbrooki ) had trait combinations that deviated from this categorization slightly Because the tributary to th e Waccamaw River contains primarily sunfishes (i.e., five species) representing great overlap in trait similarity funct ional diversity for this stream was low. This simple comparison provides an example of two streams with very different tro phic SES FDis estimates that had contrasting values for the three most important trophic SES random forest predictors (i.e., Buffer sc ale forest cover, elevation, and ecoregion), thus exemplifying their importance. Comparison with Other Studies Forest cover, elevation, and geographic region have all been shown as important influences on functional diversity in other studies. Burcher et a l. (2008) an alyzed forest changes with in southern Appalachian streams in Virginia and North Carolina and highlighted that l ow er functional composition (i.e., more generalist species with shared trophic, reproductive, and habitat traits) was ob served for fi sh assemblages in streams with low fo rest cover Additionally, forest c over was identified as an important local scale influence on functional structure in the Grijalva Basin in Mexico ; greater canopy
51 cover and forested catchments were associated with spec ies with larger eggs and smaller clutch sizes, alluding to an association between forest cover and reproduction (Pease et al. 2012). Teresa and Casatti (2012) revealed a negative association between functional diversity and forest cover; functio nal diversi ty actually increased in deforested stream mesohabitats in the Paran River basin, south easte rn Brazil, which is attributed to differences in species evenness between sites. Pease et al. (2012) revealed that elevation, latitude, and climate factors were t he most important influences of the observed gradients in functional and taxonomic structure of fish communities in the Grijalva Basin. Differences in elevation, latitude, and climate are inherent between the lowlands in the n orth and highlands in the sout h of the Grijalva Basin additionally portraying the importance of not just elevation, but overall geographic region as an influence on functional diversity To our know ledge, conductivity has not been identified as an important driver of functional divers ity prior to our analysis however, other water chemistry variables such as total phosphorus and pH have been Compared with other studies, ours provides valuable insights into functional diversity patterns across South Carolina for thre e reasons. First, we use highly standardized fish sampling data collected by the SCDNR to determine fish presence/ absence across wadeable South Carolina streams Regional or national databases are often used to determine fish community structure ( Strecker et al. 2011; Pool et al. 2012 ) and may contain information from multiple sampling efforts or combine sampling data from multiple studies. When collecting presence/ absence data from these databases it is important to consider potential biases (i.e., due to gear bias, targeted species, sampling period, etc.) which could potentially yield inaccurate functional
52 diversity measurements. Under ideal circumstances, thorough depiction of fun ctional diversity requires a unified sampling effort for the entire fish fauna (Pease et al. 2012; Buisson et al. 2013; Hitt & Chambers 2014) to ensure all species and their traits are detected Second, most studies combine many different types of traits (i.e., feeding, reproduction, habitat, locomotion, life history) into one functional diversity metric to determine an overall fun ctional diversity estimate for the community. However, we measure d functional diversity separately for trophic, reproductive, and habitat traits to explore spatial differences across South Carolina be cause different sets of traits may be important for different ecological processes and ecosystem functions In doing this, we showed that all three measurements of functional diversity across the state were uncorrelated with one another, and therefore, the re is likel y value in considering each measurement when conducting biodiversity assessments and making conservation decisions. To our knowledge we are the first to do this for freshwater fish. We also standardized our functional diversity measurements by c omparing observed values to generated null distributions (Swenson 2014) This is often not done or mentioned, but ensures that the range of possible functional diversity estimates are equivalent along the species richness continuum. And finally, we consi der a vast array of environmental predictors both at the local and landscape scale whereas other studies have consider ed just local (Er s et al. 2009) o r landscape scale variables (Poo l et al. 2012). It has been shown that environmental variables influenc e fish communities differently depending on the spatial scale of assessment (Wiens 2002; Fausch et al. 2002; Wang et al. 2003), and therefore,
53 many studies have expressed that considering multiple scales is important in assessing both functional and taxono mic diversity of fish communities (Pease et al. 2012; Buisson et al. 2013; Marion et al. 2015). Additionally, both upstream and downstream processes ( Strecker et al. 2011), as well as those impacting both the instream habitat and the surrounding landscape or riparian zone (Dudgeon et al. 2007; Strayer & Dudgeon 2010 ), may be important drivers of biodiversity patterns. Collecting the breadth of data to capture all of these potentially influential variables can be difficult depending on the availability of ti me and resources for a research study. However, o ur analysis encompasses a majority of these factors natural resources assessment and we therefore feel con f ident that we have included a substantial amount of t he potentially import ant functional diversity drivers S ome variables that are missing and that should be considered in future work are finer scale habitat variables (i.e., the numbers of riffles, runs, and pools) and land use variables at the catchment sc ale. Some key limitations should be considered when assessing these results Our analysis does not include relative abundances in our functional diversity estimates. Because it is often difficult to determine accurate relative abundance estimates for fish only a subset of studies have incorporated these ( Pease et al. 2012; Teresa and Casatti 2012; Hitt & Chambers 2014) Abundance weighted functional diversity measurements may better represent community functional structure because the ecological effects of species are often proportional to their abundances ( Daz & Cabido 2001 ; Stuart Smith et al. 2013). Additionally, most studies of freshwater fish measure multiple functional diversity metrics to determine if similar patterns are revealed from each (Er s et al.
54 2009) Often functional evenness, functional richness, and functional dispersion or divergence are measured to capture multiple components of the filled community functional trait space ( Pool et al. 2012; Teresa & Casatti 2012; Hitt & Chambers 20 14 ). We have yet to explore these other functional diversity metrics to compare with our functional diversity estimates obtained from the functional dispersion metric. Future Research and Conservation Implications As the human population increases and the Earth continues to war m, persistent land use and climate changes will continue to have detrimental effects on freshwater aquatic ecosystems of the southeastern United States ( Warren et al. 2000 ; Jelks et al. 2008) This is alarming due to the number of fre shwater fish species that are already imperiled in a region characteristic of the highest diversity in the nation (Jelks et al. 2008), and it will be important to continue monitoring the status of these fish communities. Specifically a cross South Carolina, 57 freshwater fish species are listed as species of greatest conservation need Continued urban growth has increased deforestation and the spread of non native species, and the entire state is threatened by non po int pollution and impoundments (SCDNR 2015 ). Additionally, c limate change is expected to warm the state with unknown consequences for many fish species, including the Eastern brook tro l ocated in the Blue Ridge region (SCDNR 2015). With these exacerbating threat s, it will be increasingly important to develop conservation strategies that address multiple facets of biodiversity rather than s olely focusing on the taxonomic level. Our analysis has identified the areas of highest and lowest functional diversity acros III ecoregions and elevational gradient, both of which were important functional diversity predictors based on our models. State wide
55 geographic differences between trophic and reproductive SES FDis portray that it might be benefic ial to consider both measures of functional diversity separately in conservation planning as a means to explore how specific ecosystem services and f unctions are driven by these trait categories (i.e., how instream nutrient cycling is driven by fish trophi c traits) Our RF models have also identified forest cover as important for trophic SES FDis South Carolina forests are rapidly being converted to residential land and timber and over 13 million acres of forests covering nea rly two thirds of the state are at ri sk for development (SCDNR 2015). For this reason it will be critical to focus conservation efforts on forest and riparian vegetation protection The SCDNR has already implemented a number of habitat protection programs, one of which is the Forest Legacy Program, which identifies environmentally important forests and works to preserve them (SCDNR 2015). The program should consider functional diversity measurements in their selection of forests to protect functionally unique South Carolina stream sys tems. Additionally, conductivity and catchment area have been identified as important predictor s of reproductive functional diversity and should be co nsidered when developing conservation strategies to protect reproductively unique stream communities Ou r study provides a foundation for understanding functional diversity patterns for freshwater fish communities across wadeable South Carolina streams. We have successfully identified state wide functional patterns and determined the key drivers of these pat terns, however, we hope that this work will be expand ed in the following ways. First, our functional diversity estimates can be used as a baseline for comparison with future estimates of functional diversity across the st ate to assess how specific land use
56 and climate changes have impacted particular streams and river basins. Next, the s ame analysis should be conducted in other states across the s outheastern United States to determine if similar patterns of functional diversity exist, and if the same enviro nmental drivers come out as predominant With this knowledge regional conservation efforts can be synchronized based on the most severe threats, and the most appropriate conservation strategies can be developed to reflect the pres ervation of functionally u nique communities Future research should also assess the overlap between taxonomic, phylogenetic, and func tional diversity to determine whether or not they are in alignment across the state (Strecker et al. 2011) The same variables identified as importan t in our functional diversity analysis might be different from those identified in taxonomic or phylogenetic analyses, and this needs to be explored to collectively address all of these biodiversity components when aiming to conserve the integrity of fish communities across the state Additionally measuring functional beta diversity could provide more direct insight on species turnover and homogenization. Because non native species are a significant threat, determining drivers of both native and non native functional diversity, as was done in Pool et al. (2010) cou ld highlight the environmental and landscape variables to target in counteracting this problem. We recognize that there is more knowledge to be gained, and feel that our identified South Carolina functional diversity patterns and drivers have paved the way for many of these subsequent analyses.
57 CHAPTER 5 REGIONAL ANALYSIS Background The South Atlantic Landscape Conservation Cooperative (SALCC) is part of a network of Landscape Conservation Coope ratives (LCC s ) whose goal is to inform natural resource management decisions through large scale landscape conservatio n efforts. These 22 LCCs cover North America and the U.S. Pacific and Caribbean islands, and consist of partnerships among federal agencie s, regional organizations, states, tribes, NGOs, universities and other entities w ithin the given geographic area ( LCC Network 2015). The SALCC covers portions of six states including Virginia North Carolina, South Carolina, Georgia, Florida, and Alabama and extends 200 miles into the Atlantic Ocean. In 2012, the SALCC began selecting natural resource indicators an d targets as specific landscape scale measures of success for natural resources. An indicator was defined by the SALCC as a metric designed to inform easily and quickly about th e conditions of a system, where as a target is a numeric goal established for an indicator. Recommendations were made to the SALCC steering comm ittee in March 2012 to select indicators for aquat ic and terrestrial ecosystems These r ecommendations were based on input from 235 experts in marine, freshwater, and terrestrial resources in the SALC C geography and five adjacent LCCs (SALCC 2015). The criteria in choosing indicators consisted of the following: ecological criteria, s uch as how well indicators capture key ecosystem elements and major landscape threats; practical criteria, such as t he ability to model and monitor; and social criteria, such as how well the indicators resonate with various audiences (SALCC 2015). After se lection, indicators have gone or
58 will go through a testing process in which they are modeled to reveal whether or not they are successful in representing the given ecosystem. ch ange in a species, group of species, or a community as a response to management or restoration actions or species invasions or extinctions (Cairns et al. 1993). To understand how an indicator affects a species or community, specific attributes of a communi ty must be chosen for measurement. Ideally a huge suite of attributes of species or the community of interest would be known, ranging from demographic parameters, functional life history traits, or high resolution occu pancy data. However, in reality the a vailable sampling data are often limited in terms of taxonometric scope or the attributes measured The objective of this work was to exa mine aquatic indicators proposed by the SALCC steering committee in addition to other potential indicators, and test t he statistical linkages bet ween these indicators suggested by cooperators and their ability to represent the ecological integrity of a given ecosystem. We chose to accomplish this task through the quantification of fish functional diversity, as was done fo r South Carolina wadeable streams (Chapter 2) due to its influence on ecosystem functioning (Tilman 2001) and ability to reveal community responses to disturbance (Nystrom et al. 2000; Hooper et al. 2005; Suding et al. 2008 ) Here we present our analysis in which we explored the statistical relationships between functional div ersity and the following indicators: natural cover near rivers and streams, urbanization, agriculture, open water, and the number of established invasive species. We measured SES FDis as described in C hapter 2 for South Carolina fish communities however, h ere w e measured SES FDis for fish communities in each of t he
59 121 sub basins (HUC8s) within the SALCC region ( Figure 5 1 ) We tallied species presence/absence from data obtained from the M ultistate A quatic R esources I nformation S ystem database ( MARIS; data period 1953 2012 http://www.marisdata.org/ ) which contains over one million fish sampling records from a compilation of nationwide sampling efforts, and the USGS Nonindigeno us Aqua tic Species (NAS) Database (Fuller 2015, http://nas.er.usgs.gov/ ), which provides detailed invasive fish species occurrence and distribution data. If a fish species was found at any time within a HUC8, even for a single occurrence, we considered this sp eci es as part of the community. Using these data, we ident ified 379 species in the SALCC region We collected traits (For database and trait data sources see: http://ufdc.ufl.edu/l/IR00007844/00001 ) and measured functional dispersion as was done in Chap ter 2 for South Carolina fish communities We explored associations between functional diversity for each trait category (i.e., trophic, reproductive, and habitat) and HUC8 scale measurements of natural cover near rivers and streams, urbanization, agriculture, a nd open water using Pearson correlations Our i ndicator variables were obtained from the Southeast Aquatic Resources Partnership (SARP; natural cover and open water) and the National Land Cov er Database (NLCD; urbanization and agriculture). We also related functional diversity to the number of established invasive species data which were obtained from the U.S. Geological Survey ( Fuller 2015 ). Res ults Trophic and reproductive SES FDis measurements were highly correlated across the SALCC but both trophic an d reproductive SES FDis were uncorrelated with habitat SES FDis (Table 5 1 ) We estimated h igh trophic and reproductive SES FDis
60 throughout the eastern hi ghland and Appalachian regions, and lower level s of SES FDis were generally found in Coa stal Pl a in and Piedmont regions (Figures 5 2 and 5 3 ). There was no clear pattern associated with habitat SES FDis across the region (Figure 5 4 ). T here were no significant or strong correlations between indic ators a nd SES FDis (Table 5 2 ). Trophic SES FDis was positiv ely associated with percent of natural cover, and negatively associated with urba nization, agriculture, open water, and number of invasive species. Reproductive SES FDis was positively associated with urbanization, agriculture, and open water, and negative ly associated with percent of natural cover and number of invasive species. Habitat SES FDis was positively associated with urbanization and open water, negatively associated with agriculture and number of invasive species, and not associated with percent of natural cover. Discussion Our correlation coefficients of SES FDis across the SALCC landscape suggest ed strong correlations betwee n trophic and reproductive SES FDis and no correlations between both trophic and reproductive SES FDis with habitat SES FD is H igh trophic and repr oductive SES FDis were estimated throughout the eastern highland and Appalachian regions (Figures 5 2 and 5 3) areas of known high fish diversity with large numbers of endemic species ( Hocutt & Wiley 1986; Burkhead et al. 1997 ; Sc ott 2006 ) Lower SES FDis was generally found in Coastal Plain and Piedmont areas wher e levels of endemism are lower and cleared forest cover, urbanization and agricultural land cover are greater ( Hardison et al. 2009; Utz et al. 2010) These landscape ch aracteristics lend support for the displayed differences in regional functional diversity ; anthropogenic land use modifications negatively alter aquatic systems and may reduce
61 the divers ity of species and un ique trait combinations through homogenization ( R ahel 2000; Scott & Helfman 2001) Contra stingly, there was no notable trend for regional habitat SES FDis A potential reason for why levels of habitat SES FDis lack a regional pattern is due to the high degree of habitat heterogeneity and complexity acro ss the entire S ALCC In this analysis we considered not just wadeable freshwater streams, as was done in our South Carolina analysis, but rather all of the aquatic systems within SALCC HUC8s. Therefore, the absence of a trend might be reflective of the spa tial diversity of aquatic s ystems across HUC8s, as there are inherent differences in habitat features and species assemblages comprising rivers, lakes, and streams (i.e., water flow, size, depth, instream vegetation, etc.), a nd there are ev en differences b etween rivers, lakes, and streams across ecoregions Contrastingly, o ur South Carolina analysis depic ts a clear trend of decreasing habitat SES FDis moving from the Blue Ridge to the Middle Atlanti c Coastal Plain region, and the identification of this tren d is likely due to only including freshwater wadeable streams in the analysis T here are distinctive differences in wadeable freshwater streams moving from the highlands, where streams are high gradient, to the lowlands, where streams are of lower gradient (SCDNR 2015) and it makes sense that our analysis captures a trend depict ing this Th e Alapaha River (HUC8 Unit 03110202), a blackwater river flowing from Georgia to Flo rida, portrayed the lowest SES FDis value s for all three sets of traits. Although thi s HUC8 did not have extremely un favorable conditions for the explored indicators, w e speculat a comm on characteristic of blackwater rivers, did not exhibit suitable conditions for diverse trophic interactions, reproduction,
62 and habitats. Consequently, 12 of the 14 species found in this HUC8 belong ed to the Centrarchidae or sunfish, family, and therefore exhibit ed almost the exact same set of trophic, reproductive, and habitat traits in our database. The other two species (i.e., Esox niger and Esox americanus vermiculatus ) were in the Esocidae family and also share d the same set of traits All three sub ba sins that portrayed the highest trophic, reproductive, and habitat functional diversity value s (USGS Cataloging Units 03010204 03120003, and 03040205 respectively ) had a high degree of undisturbed riparia n (i.e, all above 90%), and had close to zero percent agriculture and urbanization, all of which would be expected for highly functionally diverse systems due to the intact nat ural landscape Additionally, a much highe r diversity of fish families was represented in all of these HUC8s compared to the Alapaha River HUC8. Based on these contrasting HUC8s, it is evident that the indicato rs that we explored are not always useful in e xplaining our observe d functional diversity patterns Conclusion There are key differences between our local and regional analyses (Table 5 3) that should b e considered when evaluating our functional diversity patterns and results As opposed t o our local scale (i.e. South Carolina) analysis, our regional scale analysis used preexisting databases to determine the present fish species and, therefore, these fish sampling data were not standardized These databases contain ed records fro m multiple sampling eff orts and, a s discussed in chapter 4 this is not ideal for a complete assessment of the entire fish fauna. Additionally, these sampling efforts were unevenly distributed temporally; in some years a HUC8 might have been sampled once or not at all whereas i n other years it might have been sampled as many as 10 15 times. I f s ampling the entire fish fauna was not the goal of a specific sampling effort included in a
63 large national or regional database, a number of species might be excluded unintentionally. Simi larly, t hese data encompass records from over a 59 year time frame, and therefore, it might also be difficult to acc ount for potential extirpations and s pecies introductions when measuring functional diversity using large scale databases For lack of a sys tematic and informative way to exclude species based on the sampling data, we included all species ever sampled as present in our regional analy sis. We know that this is a coarse estim a te of community composition however, this is often the best approach a vailable when conducting these types of large scale regional analyse s through the assemblage of sampling efforts For these reasons, we feel confident that our South Carolina functional diversity estimates more accurately capture the complete state wide fi sh fauna. However, in this analysis we only characterized assemblages in wadeable freshwater streams and, therefore, we cannot draw conclusions regarding the functional diversity of as can be done in our regional analysi s. Although our regi onal fish sampling data contains some important limitations o ur results do display potentially informative patterns of trophic and reproductive SES FDis However it is currently unclear if these estimates of functional diversity are e ffective in providing information about the ecological integrity of regional aquatic systems the goal In order for functional diversity to be an effective surrogate for ecosystem integrity, our SES FDis estimates must pro vide useful information regarding specific indicators or environmental variables that can be associated with overall ec osystem health. In other words, significant relationships should exist between SES FDis and specific features of the landscape aquatic h abitat,
64 or the environment. Whereas we successfully identified a number of significant drivers of our local e stimates of trophic and reproductive SES FDis (Chapter 3) our regional analysis showed that our subset of environmental indicators were not signif icant drivers of our SES FDis patterns. It is possible that this lack of significance is an artifact of potentially inaccurate functional diversity estimates due to the sampling data included. However, in this analysis we only analyzed five specific indica tors due to the availability of data but there are a number of other variables to consider that are potentially significant drivers of regional functional diversity For example, in future work we plan to analyze the number of i mperiled species across SAL CC HUC8s If there happens to be a significant relationship between SES FDis and the number of imperiled fish species, we would be able to conclude that functional diversity successfully depicts the ecological integrity of fish communities across the regio n through the identification of threatened and endangered fishes This would enable the identification of areas of high conservation priority (i.e., areas where both functional diversity and the number of imperiled species are high). Additionally, we hope to consider the number of dams and impoundments, climate and water quality as indicators all of which have been shown as important deter minants of fish functional diversity ( ) and could all have important links with SES FDis across the SALCC Ultimately, if there are no significant associations between functional diversity and indicators, it is possible that these indicators simply do not reveal valua ble insight regarding fish traits across the entire SALCC region Theref ore, using functional diversity to measure indicator effectiveness might not be warranted. Perhaps assessing the effectiveness of these indicators through the analysis of a different biodiversity
65 response, such as taxonomic di versity, species distribution models or measures of beta diversity could be more effective at the regional scale and s hould be explored as a subsequent step I ncorporating the effects of multiple biodiversity components could be valuable in revealing which are impacted by specific i ndicators and thus aid in the development of robust and multi faceted regional conservation strategies (Devictor et al. 2010; Strecker et al. 2011; Stuart Smith 2013 ).
66 Table 5 1. Pearson correlation coefficients between SA LCC functional diversity metrics. Trophic Reproductive Habitat Trophic 1.0 0.91 0.48 Reproductive 0.91 1.00 0.36 Habitat 0.48 0.36 1.00 Note: All c orrelations are significant at the p < 0.05 level. Table 5 2 Pearson correlation coefficients betwee n SALCC functional diversity metrics and environmental indicators. Trophic Reproductive Habitat Percent of Natural Cover 0.13 0.03 0.00 Percent Urbanization 0.11 0.10 0.08 Percent Agriculture 0.10 0.01 0.08 Percent Open Water 0.01 0.03 0.04 Numb er of Invasive Species 0.06 0.08 0.03 Note: No significant correlations. Table 5 3 Summary of differences between South Carolina (local) and SALCC ( regional) functional diversity analyses. South Carolina SALCC Data Collection Period 2006 to 2011 1953 to 2012 Number of Sampling E fforts One Multiple Standardization Standardized Not s tandardized Fish Sampling Data Source SCDNR Stream Assessment MARIS and USGS NAS d atabases Waterbody Types Included South Carolina freshwater s treams All aquatic sys tems across SALCC HUC8s Number of Species Included 101 379
67 Figure 5 1. Map of SALCC boundaries and HUC8s. The SALCC region is outlined in red and HUC8s are outlined in blue.
68 F igure 5 2. T rophic SES FDis across SALCC HUC8s. Warm colors indicate high SES FDis and cool colors indicate low SES FDis.
69 Figure 5 3 R eproductive SES FDis across SALCC HUC8s. Warm colors indicate high SES FDis and cool colors indicate low SES FDis.
70 Figure 5 4. H abitat SES FDis across SALCC HUC8s. Warm colors indic ate high SES FDis and cool colors indicate low SES FDis.
71 LIST OF REFERENCES Alavi, S.M.H. & Cosson, J. (2006) Sperm motility in fishes (II). Effects of ions and osmolality: A review. Cell Biology International 30 1 14. Allan, J.D. (1995) Stream ecolo gy: structure and function of running waters Chapman and Hall, London. A nderson, M.J., Ellingsen, K.E. & McArdle, B.H. (2006) Multivariate dispersion as a measure of beta diversity. Ecology Letters 9 683 693. Balon, E.K. (1975) Reproductive guilds of fishes proposal and definition. Journal of the Fisheries Research Board of Canada 32 821 864. Benke, A.C., Vanarsdall, T.C., Gillesp ie, D.M. & Parrish, F.K. (1984) Invertebrate productivity in a sub tropical blackwater river The importance of habita t and life history. Ecological Monographs 54 25 63. Breiman, L. (1996) Bagging predictors. Machine Learning 24 123 140. Breiman, L. (2001) Random forests. Machine Learning 45 5 32. Breiman, L. (2003) Manual for setting up, using, and understandi ng rand om f orest V4.0. Available at : https://www.stat.berkeley.edu/~breiman/Using_random_forests_v4.0.pdf (accessed May 2016). Burcher, C.L., McTammany, M.E., Benfie ld, E .F. & Helfman, G.S. (2008) Fish assemblage responses to forest cover. Environmental Management 41 336 346. Burkhead, N.M. (2013 ) The Southeastern Fishes Council List of Freshwater Fishes. Available at: http://www.sefishescouncil.org/fishes/ (accessed May 2016). Burkhead, N.M., Walsh, S.J., Freema n, B.J. & Williams, J.D. (1997) Status and restoration of the Etowah River, an imperiled Southern Appalachian ecosystem. Aquatic fauna in peril: the Southea stern perspective (ed. by G.W. Benz & D.E. Collins) Southeast Aquatic Research Institute, Lenz Design and Communications, Decatur. Buisson, L., Grenouill et, G., Villger, S., Canal, J. & Laffaille, P. (2013) Toward a loss of functional diversity in strea m fish assemblages under climate change. Global Change Biology 19 387 400. Cadotte, M.W. (2011). The new diversity: management gains through insights into the functional diversity of communities. Journal of Applied Ecology 48 1067 1069.
72 Cairns, J. (J r.), McCormick, P.V. & Niederlehner, B.R. (1993) A proposed framework for developing indicators of ecosystem h ealth. Hydrobiologia 263 1 44. Carpenter, S.R., Stanley, E.H. & Zanden, M.J.V. (2011) ecosystems: Physical, che mical, and biological changes. Annual Review of Environment and Resources 36 75 99. Cornelissen, J.H.C., Lavorel, S., Garnier, E., Daz, S., Buchmann, N., Gurvich, D.E., Reich, P.B., Ter, Morgan, H.D., van de r Heijden, M.G.A., Pausas, J.G. & Poorter, H. ( 2003 ) A handbook of protocols for standardized and easy management of plant functional traits worldwide. Australian Journal of Botany 51 335 380. Cowling, R.M., E sler, K.J., Midgley, G.F. & Honig, M.A. (1994) Plant functional diversity, species divers ity and climate in arid and semi arid southern Africa. Journal of Arid Environments 27 141 158. D'agata, S., Mouillot, D., Kulbicki, M., Andrfout, S., Bellwood, D.R., Cinner, J.E., Cowm an, P.F., Kronen, M., Pinca, S. & Vigliola, L. (2014) Human mediat ed loss of phylogenetic and functional diversity in coral reef fishes. Current Biology 24 555 560. Daniel, M.H.B., Montebelo, A.A., Bernardes, M.C., Ometto, J.P.H.B., DeCamargo, P.B., Krusche, A.V., Ballester, M.V., Victoria, R.L. & Martinelli, L.A. (20 02) Effects of urban sewage on dissolved oxygen, dissolved inorganic and organic carbon, and electrical conductivity of small streams along a gradient of urbanization in the Piracicaba River basin. Water Air and Soil Pollution 136 189 206. Daufresne, M. & Boet, P. (2007) Climate change impacts on structure and diversity of fish communities in rivers. Global Change Biology 13 2467 2478. th, G. & Fabricius, K.E. (2000) Classification and regression trees: A powerful yet simple technique for ecologic al data analysis Ecology 81 3178 3192. De Bello, F., Lep J. & Sebasti M.T (2006) Variations in species and functional plant diversity along climatic and grazing gradients. Ecography 29 801 810. Devictor, V. Mouillot, D., Meynar d, C., Jiguet, F ., Thuiller, W. & Mouquet, N. (2010) Spatial mismatch and congruence between taxonomic, phylogenetic, and functional diversity: the need for integrative conservation strategies in a changing world. Ecology Letters 13 1030 1040. Daz, S. & Cabido, M. (20 01) Vive la diffrence: plant functional diversity matters to ecosystem processes. Trends in Ecology and Evolution 16 646 655.
73 Dudgeon, D., Arthington, A.H., Gessner, M.O., Kawabata, Z.I., Knowler, D.J., Leveque, C ., Naiman, R.J., Prieur Richard, A .H., Soto, D. Stiassny, M.L.J. & Sullivan, C.A. (2006) Freshwater biodiversity: importance, threats, status and conservation challenges. Biological Reviews 81 163 182. E & Rask, M. (2009) Characterizing functional trait diversity and trait environment relationships in fish assemblages of boreal lakes. Freshwater Biology 54 1788 1803. Fausch, K.D., Torgersen, C.E., Baxter, C.V. & Li, H.W. (2002) Landscapes to riverscapes: Bridging the gap between research and conservation of stream fishes. Bioscience 52 483 498. Flynn, D.F., Gogol Prokurat, M., Nogeire, T., Molinar i, N., Richers, B.T., Lin, B.B. & DeClerck, F. (2009) Loss of functional d iversity under land use intensification across multiple taxa. Ecology letters 12 22 33 Fo nseca, C.R. & Ganade, G. (2001) Species functional redundancy, random extinctions and the stability of ecosystems. Journal of Ecology 89 118 125. Frimpong, E.A. & Angermeier, P.L. (2009) Fish traits: a database of ecological and life history traits of freshwater fishes of the United States. Fisheries 34 487 495. Fuller, P. (2015) Nonindigenous Aquatic Species Database, Gainesville, FL. U.S. Geological Survey. A vailable at: http://nas.er.usgs.gov/ Gessne r, M.O. (2002) A case for using litter breakdown to assess functional stream integrity. Ecological Applications 12 498 510. Giffor d, D.J., Collie, J.S. & Steele, J.H. (2009) Functional diversity in a marine fish community. Ices Journal of Marine Science 66 791 766. Grimm, R., Behrens, T., Mae rker, M. & Elsenbeer, H. (2008) Soil organic carbon concentrations and stocks on Barro Col orado Island Digital soil mapping using Random Forest analysis. Geoderma 1 2 102 113. Gower, J.C. (1971) A general coefficient of similarity and some of its properties. Biometrics 27 857 871. Halpe rn, B.S. & Floeter, S.R. (2008) Functional diversit y responses to changing species richness in reef fish communities. Marine Ecology Progress Series 364 147 156.
74 Howard, R.J. & Brinson, M.M. (2009) Urban land use, channel incision, and water table decl ine along coastal plain streams, North Carolina. Journal of the American Water Resources Association 45 1032 1046. Harrelson, C.C., Rawlin s, C.L. & Potyondy, J.P. (1994) Stream channel reference sites: an illustrated guide to field technique. U.S. Fores t Service General Technical Report RM 245. Helms, B.S., Werneke, D.C., Gangloff, M.M., Hartfield, E.E. & Feminella, J.W. (2011) The influence of low head dams on fish assemblages in streams across Alabama. Journal of the North American Benthological Soci ety 30 1095 1106. Hit t, N.P. & Chambers, D.B. (2014) Temporal changes in taxonomic and functional diversity of fish assemblages downstream from mountaintop mining. Freshwater Science 33 915 926. Hocutt, C.H. & Wiley, E.O. (1986) Zoogeography of the F ishes of the Southeastern United States: Savannah River to Lake Pontchartrain The Zoogeography of North American Freshwater Fishes John Wiley & Sons, Inc. New York. Hooper, D.U., Chapin, F.S., Ewel, J.J., Hector, A., Inchausti, P., Lavorel, S., Lawton J.H., Lodge, D.M., Loreau, M., Naeem, S., Schmid, B., Setala, H. Symstad, A.J., Vandermeer, J. & Wardle, D.A (2005) Effects of biodiversity on ecosystem functioning: A consensus of current knowledge. Ecological Monographs 75 3 35. Jelks, H.L., Wals h, S.J., Burkhead, N.M., Contreras Balderas, S., Diaz Pardo, E., Hendrickson, D.A., Lyons, J., Mandrak, N.E., McCormick, F., Nelson, J.S., Platania, S.P., Porter, B.A., Renaud, C.B., Sch mitter Soto, J.J., Taylor, E.B. & Warren, M. L. (Jr.). (2008) Conservat ion status of imperiled North American freshwater and diadromous f ishes. Fisheries 33 372 407. Johnson, S. L. & Jones, J.A. (2000) Stream temperature responses to forest harvest and debris flows in western Cascades, Oregon. Canadian Journal of Fisheries and Aquatic Sciences 57 30 39. Jones, E.B.D., Helfman, G.S., Harp er, J.O. & Bolstad, P.V. (1999) Effects of riparian forest removal on assemblages in southern Appalachian streams. Conservation Biology 13 1454 1465. Jones, K.B., Neale, A.C., Nash, M.S ., Van Remortel, R.D., Wickham, J.D., Riitt ers, K.H. Predicting nutrient and sediment loadings to streams from landscape metrics: A multiple watershed study from the United States Mid Atlantic Region. Landscape Ecology 16 301 312.
75 Kattge, J., Ogle, K., Bnisch, G., Daz, S., Lavorel, S., Madin, J., Nadrows ki, K., Nllert, S., Sartor, K. & Wirth, C. (2011) A generic structure for plant trait databases. Methods in Ecology and Evolution 2 202 213. Kaushal, S.S., Likens, G.E., Jawor ski, N.A., Pace, M.L., Sides, A.M., Seekell, D., Belt, K.T., Sec or, D.H. & Wingate, R.L. (2010) Rising stream and river temperatures in the United States. Frontiers in Ecology and the Environment 8 461 466. Lalibert, E. & Legendre, P. (2010) A distance based framework for measuring functional diversity from multiple traits. Ecology 91 299 305. Landscape Conservation Cooperative Network (LCC Network) (2015) About Landscape Conservation Cooperatives. Available at: https://lccnetwork.org/about/about lccs (accessed May 2016). Les sard, J.L. & Hayes, D.B. (2003) Effects of elevated water temperature on fish and macroinvertebrate communities below small dams. River Research and Applications 19 721 7 32. Liaw, A. & Wiener, M. (2002) Classification and regression by randomForest. R News 2/3 18 22. Likens, G.E., Bormann, F.H., Johnson, M.N., Fis her, D.W. & Pierce, R.S. (1970) Effects of forest cutting and herbicide treatment on nutrient budgets in th e Hubbard Brook watershed ecosystem. Ecological Monographs 40 23 47. Lowrance, R., Altier, L.S., Newbold, J.D., Schnabel, R.R., Groffman, P.M., Denver, J.M., Correll, D.L., Gilliam, J.W., Robinson, J.L., Brinsfield, R.B., Staver, K.W. Lucas, W. & Todd, A.H. (1997) Water quality functions of Riparian forest buffers in Chesapeake Bay watersheds. Environmental Management 21 687 712. Macedo, D.R., Hughes, R.M., Ligeiro, R., Ferreira, W.R., Castro, M.A., Junqueira, N.T., Oliveira, D.R., Firmiano, K.R ., Ka ufmann, P.R., Pompeu, P.S. & Callisto, M. (2014) The relative influence of catchment and site variables on fish and macroinvertebrate richness in cerrado biome streams. Landscape Ecology 29 1001 1016. Marion, C.A., Scott, M.C. & Kubach, K.M. (2015) Mult iscale environmental influences on fish assemblage structure of South Atlantic Coastal Plain streams. Transactions of the American Fisheries Society 114 1040 1057. Mayden, R.L. (1987) Historical ecology and North American highland fishes: a research pro gram in community ecology. Community and Evolutionary Ecology of North American Stream Fishes (ed. by W.J. Matthews & D.C. Heins), University of Oklahoma Press, Norman.
76 Mouillot, D., Dumay, O. & Tomasini, J.A. (2007) Limiting similarity, niche filtering, and functional diversity in coastal lagoon fish communities. Estuarine Coastal and Shelf Science 3 4 443 456. Nystrom, M., Folke, C. & Moberg, F. (2000) Coral reef disturbance and resilience and in a human dominated environment. Trends in Ecology and E volution 15 413 417. Oliveira, S., Oehler, F., San Miguel Ayanz, J., Cam ia, A. & Pereira, J.M.C. (2012) Modeling spatial patterns of fire occurrence in Mediterranean Europe using Multiple Regression and Random Forest. Forest Ecology and Management 275 117 129. Omernik J.M. ( 1987 ) Ecoregions of the conterminous United States. Annals of the Association of American Geographers 77 118 125. Paul, M.J. & Meyer, J.L. (2001) Streams in the urban landscape. Annual R eview of Ecology and Systems 32 333 3 65. Pease, A.A., Gonzalez Di az, A.A., Rodiles Hernandez, R. & Winemiller, K.O. (2012) Functional diversity and trait environment relationships of stream fish assemblages in a large tropical catchment. Freshwater Biology 5 1060 1075. Petchey, O.L. & Gas ton, K.J. (2002) Functional diversity (FD), species richness and community composition. Ecology Letters 5 402 411. Pool, T. K., Olden, J.D., Whittier, J.B. & Paukert, C.P. (2010) Environmental drivers of fish functional diversity and composition in the L ower Colorado River Basin. Canadian Journal of Fisheries and Aquatic Sciences 67 1791 1807. Rahel, F.J. (2000) Homogenization of fish faunas across the United States. Science 288 854 856. Richter, B.D., Braun, D.P., Mendelson, M.A. & Master, L.L. (19 97). Threats to imperiled freshwater fauna. Conservation Biology 11 1081 1093. R Core Development Team. (2012) R: a language and environment for statically computing. R Foundation for Statistical Computing, Vienna. & Podani, J. (2009) A measure for assessing functional diversity in ecological communities. Aquatic Ecology 43 157 167. Schweiger, O., Musche, M., Bailey, D., Billeter, R ., Diektter, T., Hendrickx, F. & Dziock, F. (2007) Functiona l richness of local hoverfly communities (Diptera, Syrphidae) in response to land use across temperate Europe. Oikos 116 461 472.
77 Scott, M.C. (2006) Winners and losers among stream fishes in relation to land use legacies and urban development in the sou theastern US. Biological Conservation 127 301 309. Scott M.C. & Helfman G.S. ( 2001 ) Native invasions, homogenization, and the mismeasure of integrity of fish assemblages. Fisheries 26 6 15. Scott, M.C., Rose, L., Marion, C.A., Kubach, K .M., Thomaso n, C. & Price, J. (2009) The South Carolina stream assessment standard operating procedures South Carolina Department of Natural Resources, Columbia. South Atlantic Landscape Conservation Cooperative (SALC C). (2015) Indicators Roadmap. Available at: http://www.southatlanticlcc.org/indicators roadmap/ (accessed April 2016). South Carolina Departm ent of Natural Resources (SCDNR) (2015) South Carolina State Wildlife Action Plan. Available at: http://dnr.sc.gov/swap/index.html (accessed May 2016). Strayer D.L. & and Dudgeon, D. (2010) Freshwater biodiversity conservation: recent progress and future challenges. Journal of the North American Benthological Society 29 344 358 Strange, R.M. & Burr, B.M. (1997). Intraspecific phylogeography of North American highland fishes: A test of the Pleistocene vicariance hypothesis. Evolution 51 885 897. Strecker, A. L., Olden, J.D., Whittier, J.B. & Paukert, C.P. (2011) Defining conservation priorities for freshwater fishes according to taxonomic, functional, and phylogenetic diversity. Ecological Applications 21 3002 3013. Strobl, C., Boulesteix, A., Kneib, T., Augustin, T. & Zeileis, A. (200 8) Co nditional variable importance for random forests. BMC Bioinformatic 9 307. Stuart Smith, R.D., Bates, A.E., Lefcheck, J.S., Duffy, J.E., Baker, S.C., Thomson, R.J., Stuart Smith, J.F., Hill, N.A., Kininmonth, S.J., Airoldi, L., Becerro, M.A., Campbell, S.J., Dawson, T.P., Navarrete, S.A., Soler, G.A., Strain, E.M.A., Willis, T.J. & Edgar, G.J. (2013) Integrating abundance and functional traits reveals new global hotspots of fish diversity. Nature 501 539 542. Suding, K.N., Lavorel, S., Chapin, F.S., Cornelissen, J.H.C, Diaz, S. Garnier, E., Goldberg, D., Hooper, D.U., Jackson, S.T. & Navas, M.L. (2008) Scaling environmental change through the community level: a trait based response and effect framework for plants. Global Change Biology 14 1125 1140.
78 Sweeney, B.W., Bott, T.L., Jackson, J.K., Kaplan, L.A., Newbold, J.D., Standley, L.J., Hessi on, W.C. & Horwitz, R.J. (2004) Riparian deforestation, stream narrowing, and loss of stream ecosystem services. Proceedings of the National Academy of Sciences o f the United States of America 101 14132 14137. Swenson, N.G. (2014) Functional and phylogenetic ecology in R Springer Science+Business Media Publications, New York. Swenson, N.G., Enquist, B.J., Pither, J., Kerkhoff, A.J., Boyle, B., Weiser, M.D., E lser, J.J., Fagan, W.F., Forero Monta a, J., Fyllas, N., Kraft, N.J.B., Lake, J.K., Moles, A.T., Pati o, S., Phillips, O.L., Price, C.A., Reich, P.B., Quesada, C.A., Stegen, J.C., Valencia, R., Wright, I.J., Wright, S.J., Andelman, S., Jrgensen, P.M., Lac her, T.E. (Jr.)., Monteagudo, A., N zez Vargas, M.P., Vasquez Mart nez, R. & Nolting, K.M. (2012) The biogeography and filtering of woody plant functional diversity in North and South America. Global Ecology and Biogeography 21 798 808. Tere sa, F.B. & Casatti, L. (2012) Influence of forest cover and mesohabitaat types on functional and taxonomic diversity of fish communities in Neotropical lowland streams. Ecology of Freshwater Fish 21 433 442. Tilman, D. (2001) Functional diversity. Encyclopedia of Biodiversity (ed. by S.A., Levin), Academic Press New York Utz, R.M., Hilderbrand R.H. & Raesly, R.L. (2010) Regional differences in patterns of fish species loss with changing land use. Biological Conservation 143 688 699. Villger, S., Mason, N.W. & Mouillot, D. (2008) New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology 89 2290 2301. Villger, S., Miranda, J.R., Hern ndez, D.F. & Mouillot, D. (2012) Low functional beta diversity despite h igh taxonomic beta diversity among tropical estuarine fish communities. Plos One 7 e40679. Villger, S., Grenouillet, G. & Brosse, S. (2013) Decomposing functional beta diversity reveals that low functional beta diversity is driven by low functional tur nover in European fish assemblages. Global Ecology and Biogeography 22 671 681. Walker, B., Kinzig, A. & Langridge, J. (1999) Original articles: plant attribute diversity, resilience, and ecosystem function: the nature and significance of dominant and m inor species. Ecosystems 2 95 113.
79 Wallace, J.B., Webster, J.R. & Lowe, R.L. (1992) High gradient streams of the Appalachians. Biodiversity of the southeastern United States: aquatic communities. (ed. by C.T. Hackney, S.M. Adams & W.H. Martin), John W iley & Sons, New York. Walsh, C.J. Roy, A.H., Feminella, J.W., Cottingham, P.D., Groffman, P.M. & Morgan, R.P. (2005) The urban stream syndrome: current knowledge and the search for a cure. Journal of the North American Benthological Society 24 706 723 Wang, L.Z., Lyons, J. & Kanehl, P. (2001) Impacts of urbanization on stream habitat across multiple spatial scales. Environmental Management 28 255 266. Wang, L.Z., Lyons, J., Rasmusse, P., Seelbach, P., Simon, T., Wiley, M., Kanehl, P., Baker, E., N iemela, S. & Stewart, P. (2003) Watershed, reach and riparian influences on stream fish assemblages in the northern lakes and forest ecoregion, USA. Canadian Journal of Fisheries and Aquatic Sciences 60 491 505. Wang, X. & Yin, Z. (1997) Using GIS to assess the relationship between land use and water quality at a watershed level. Environment International 23 103 114. Warren, M.L. (Jr.)., Burr, B.M., Walsh, S.J., Bart, H.L. (Jr.)., Cashner, R.C., Etnier, D.A., Freeman, B.J., Kuhajda, B.R., Mayden, R. L., Robison, H.W., Ross, S.T. & Starnes, W.C. (2000) Diversity, distribution, and conservation status of the native freshwater fishes of the southern United States. Fisheries 25 7 31. Weiher, E., van der Werf, A., Thompson K., Roderick, M., Garnier, E & Eriksson, O. ( 1999 ) Challenging Theophrastus: A Common core list of plant traits for functional ecology. Journal of Vegetative Science 10 609 620. Westoby, M., Falster, D.S., Moles, A.T., Vesk, P.A. & Wright, I.J. (2002) Plant ecological strategies: Some leading dimensions of variation between species. Annual Review of Ecology and Systematics 33 125 159. Wiedmann, M.A., Aschan, M., Certain, G., Dolgov, A., Greenacre, M., Johannese n, E., Planque, B. & Primicerio, R. (2014) Functional diversity of t he Barents Sea fish community. Marine Ecology Progress Series 495 205 218. Wiens, J.A. (2002) Riverine landscapes: taking landscape ecology into the water. Freshwater Biology 47 501 515. Williamso n, C.E., Dodds, W., Kratz, T.K. & Palmer, M.A. (2008) Lakes and streams as sentinels of environmental change in terrestrial and atmospheric processes. Frontiers in Ecology and the Environment 6 247 254.
80 BIOGRAPHICAL SKETCH Joshua Epstein is a m Con servation at the University of Flor i degree in biological sciences from Rutgers University in May of 2013. There he conducted research on predicting upwelling through the analysis of foraminifera shell isotopes, and wrote his senior honors thesis on blue crab sperm limitation in the Chesapeake Bay. After graduating, Joshua worked as the oyster nursery technician for Atlantic Capes Fisheries in Cape May, N J before beginning his m his thesis Joshua measured functional diversity of freshwat er fish communities across the s outheastern United States. In Augu st of 2016, he will begin his PhD program in the School of Forest Resources and Conservation at the University of Florida, studying wetland ecology and hydrology.