Accepted Article
Received Date : 28-Jul-2014 Accepted Date : 13-Mar-2015 Article type
: Standard Paper
Global patterns and predictors of fish species richness in estuaries Rita P. Vasconcelos*, Sofia Henriques, Susana França, Stéphanie Pasquaud, Inês Cardoso, Marina Laborde, Henrique N. Cabral MARE - Marine and Environmental Sciences Centre, Faculdade de Ciências, Universidade de Lisboa, Campo Grande, 1749-016 Lisboa, Portugal
*corresponding author: e-mail [email protected]
Running headline Global predictors of fish diversity in estuaries
Abstract (1) Knowledge on global patterns of biodiversity and regulating variables is indispensable to develop predictive models. (2) The present study used predictive modelling approaches to investigate hypotheses that explain the variation in fish species richness between estuaries over a worldwide spatial extent. Ultimately, such models will allow assessment of future changes in ecosystem structure and function as a result of environmental changes. (3) A comprehensive worldwide database was compiled on the fish assemblage composition and environmental characteristics of estuaries. Generalized Linear Models were used to quantify how variation in species richness among estuaries is related to historical events, energy dynamics, and ecosystem characteristics, whilst controlling for sampling effect. (4) At the global extent, species richness differed among marine biogeographic realms and continents, and increased with mean sea surface temperature, terrestrial net primary productivity, and the stability of connectivity with marine ecosystem (open versus temporarily open estuaries). At a smaller extent (within marine biogeographic realm or continent) other characteristics were also
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/1365-2656.12372 This article is protected by copyright. All rights reserved.
Accepted Article
important in predicting variation in species richness, with species richness increasing with estuary area and continental shelf width. (5) The results suggest that species richness in an estuary is defined by predictors that are spatially hierarchical. Over the largest spatial extents species richness is influenced by the broader distributions and habitat use patterns of marine and freshwater species that can colonize estuaries, which are in turn governed by history contingency, energy dynamics and productivity variables. Species richness is also influenced by more regional and local parameters that can further affect the process of community colonization in an estuary including the connectivity of the estuarine with the adjacent marine habitat, and, over smaller spatial extents, the size of these habitats. In summary, patterns of species richness in estuaries across large spatial extents seem to reflect from global to local processes acting on community colonization. The importance of considering spatial extent, sampling effects and of combining history and contemporary environmental characteristics when exploring biodiversity is highlighted.
Key-words coastal, community, alpha-diversity, meta-analysis, review, predictive models, species distribution models.
Introduction Estuaries are among the most biologically productive and valuable ecosystems worldwide (Costanza et al. 1997), yet are threatened by an increase in human activities which may affect ecosystem health, functions and services (Barbier et al. 2011). As in marine environments, human impacts have depleted species and habitats, degraded water quality and accelerated species invasions (Lotze et al. 2006). This loss of marine biodiversity is increasingly impairing ecosystem services, namely food provision, water quality maintenance and recovery from perturbations (Worm et al. 2006). Understanding biodiversity patterns and their underlying processes is therefore important when investigating the consequences of biodiversity loss to ecosystem function and services. Biodiversity is viewed as the variety of life, embracing variation from genes to ecosystems, and is commonly measured as species richness (Purvis & Hector 2000). Global patterns of biodiversity in estuaries have been largely unexplored, despite the value of estuarine ecosystems and the existing knowledge on variability of fish communities within estuaries and their environmental drivers (see
This article is protected by copyright. All rights reserved.
Accepted Article
reference works Whitfield 1998; Blaber 2000; Elliott & Hemingway 2002; Able & Fahay 2010). As transitional systems, estuaries establish links with marine and freshwater ecosystems (Beck et al. 2001; Attrill & Power 2002), and persistent environmental fluctuations place considerable physiological demands on the species that utilize them (e.g Elliott & Quintino 2007). Fish communities in estuaries include estuarine resident species, and also marine and freshwater species which enter estuaries eitheras migrants or stragglers, as well as diadromous and amphidromous species (e.g. Elliott et al. 2007). The fish communities in estuaries tend to be dominated by a few persistent and abundant core species (e.g. Magurran & Henderson 2003), and their species richness is lower than in the adjacent marine environment (e.g. Martino & Able 2003). The main environmental driver of the structure of fish communities within estuaries is the longitudinal salinity gradient (e.g. Sosa-López et al. 2007; Whitfield et al. 2012).
The scarcity of knowledge on global richness patterns and predictors for estuarine ecosystems contrasts remarkably with adjacent marine and freshwater ecosystems (Oberdorff, Guégan & Hugueny 1995; Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013). Investigation of the environmental correlates of fish richness among estuaries is fragmented, and it has generally been based on region-specific approaches, without standardization for sampling effects and at mismatched spatial extents. This knowledge gap occurs for most taxa in estuaries, except for mangroves (Polidoro et al. 2010) and macroinvertebrates (Attrill, Stafford & Rowden 2001).
The present study aims to identify patterns of variation in fish species richness among estuaries distributed across large geographical extents, and to disentangle the relative importance of underlying predictors. Several hypotheses to explain spatial variation of species richness patterns in estuaries were considered, some of which have been repeatedly identified across taxa and ecosystems (Gaston 2000).
The evolutionary history hypothesis states that evolutionary processes have resulted in different faunal composition of oceanic basins, leading to differences in species richness (e.g. Floeter et al. 2008; Tittensor et al. 2010). The evolutionary processes that have generated these differences affect rates of speciation and adaptation to available habitats. Accordingly, we hypothesize that across large
This article is protected by copyright. All rights reserved.
Accepted Article
extents, estuaries with faunas possessing different evolutionary histories will fundamentally differ in species richness, as can be seen across small spatial extents for estuaries (e.g. Monaco, Lowery & Emmett 1992; Pease 1999; Harrison & Whitfield 2006), and globally for coastal and freshwater ecosystems (Oberdorff, Guégan & Hugueny 1995; Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013).
A latitudinal gradient in species richness is a ubiquitous ecological pattern, with numerous explanatory mechanisms proposed (Willig, Kaufman & Stevens 2003; Gaston 2007). Among these, the species-energy hypothesis advocates that metabolic rate increases with temperature and promotes higher speciation rates leading to greater diversity. Additionally it is possible that species distribution ranges are largely set by thermal tolerance, with more species tolerant of warm temperatures (Gaston 2000). Based on evidence for estuaries (Pease 1999; Harrison & Whitfield 2006; Nicolas et al. 2010; França, Costa & Cabral 2011) and coastal and freshwater systems (Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013) we hypothesize that, across large extents, species richness in estuaries decreases with latitude and increases with temperature. It is possible that differences in species richness between estuaries at large spatial extents relate to primary productivity, as previously supported for freshwater fish (Oberdorff, Guégan & Hugueny 1995). This productivity-richness hypothesis suggests a positive effect of primary productivity on species richness by allowing larger populations to persist, thereby reducing extinction risk and supporting a higher diversity of niche specialists (Willig, Kaufman & Stevens 2003; Tittensor et al. 2010). The hypothesis of Species-Area Relationships (SAR) assumes a change in species numbers with increasing area, and has been termed one of the few laws in ecology. Essentially, larger sampled areas support higher number of individuals and therefore increase the probability of encountering additional species, as well as increased chances of finding environmental heterogeneity and species that differ in their niches (Pihl et al. 2002; Scheiner 2003; Guilhaumon et al. 2008; Guilhaumon et al. 2012). We consider the possibility that species richness in estuaries can increase with habitat area (Horn & Allen 1976; Monaco, Lowery & Emmett 1992; Nicolas et al. 2010). Additionally, since assemblages in estuaries include marine, freshwater and estuarine species, then differences in species richness between widely distributed estuaries may also relate to habitat area in the adjacent
This article is protected by copyright. All rights reserved.
Accepted Article
coastal and freshwater ecosystems (Nicolas et al. 2010). It is known that species richness in those ecosystems is dependent upon habitat area (Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013). Species richness in estuaries is commonly dominated by marine species (e.g. Whitfield 1999; Franco et al. 2008), and therefore colonization processes and isolation should influence species richness. We hypothesize that patterns of species richness in estuaries across large spatial extents relate to connectivity between estuaries and the marine ecosystem. Species richness has been shown to be higher in those estuaries that are permanently open to the marine environment (Pease 1999; Harrison & Whitfield 2006), and that have a greater mouth size (Horn & Allen 1976; Monaco, Lowery & Emmett 1992; Nicolas et al. 2010). By contrast lower percentages of marine species have been found in fish communities of microtidal estuaries (Young & Potter 2003).
Finally, fish assemblages in estuaries are naturally faced with environmental stress, particularly related to salinity (Elliott & Quintino 2007; Whitfield et al. 2012). Thus it is possible that salinity is linked with patterns of species richness at large spatial extents. Increases in river flow in an estuary can lead to decreased suitability of estuarine habitat for marine species and consequently decreased species richness (Whitfield & Harrison 2003), while hyperhaline areas within estuaries have also been shown to have lower species richness (Whitfield et al. 2012).
This study relies on a comprehensive database on fish assemblages and environmental characteristics of estuaries worldwide, based on studies at single estuary scale. Predictor variables were identified to address the hypotheses above and a statistical modelling approach was applied first at a global extent, and then at a smaller geographical extent (each marine biogeographic realm and continent). The results should improve our ability to predict biodiversity at the estuary scale, which is paramount in view of continued environmental change and biodiversity loss. Materials and Methods Database A database was built combining published data on: (a) fish species assemblages in estuaries; and (b) historical and environmental characteristics of those estuaries (details in Appendix S1 and S2). In this database, a sample consists of the total list of fish species sampled in a given estuary and study.
This article is protected by copyright. All rights reserved.
Accepted Article
The obtained database has 786 samples for 430 estuaries worldwide (Fig.1; details in Appendix S1 and S2). To evaluate the evolutionary history hypothesis two factors were included in the database: marine biogeographic realm - the largest spatial unit of the Marine Ecoregions of the World system (Spalding et al. 2007) and hereafter designated as realm (Arctic, Temperate Southern Africa, Temperate Northern Atlantic, Temperate Northern Pacific, Temperate South America, Temperate Australasia, Tropical Atlantic, Tropical Eastern Pacific, Western Indo-Pacific, Central Indo-Pacific); and continent (Africa, Europe, North America, South America, Asia, Oceania). Contemporary environmental conditions and habitat distribution are poor predictors of marine faunal breaks (Keith et al. 2013), which supports the use of a biogeographic classification scheme to evaluate the evolutionary history hypothesis.
To address the latitude-hypothesis, absolute latitude was determined (º; at estuary mouth), whilst for the energy-hypothesis temperature was determined (ºC; annual mean sea surface temperature outside the estuary mouth). Regarding the productivity-hypothesis, and as comparable data for estuaries are rare, primary productivity was evaluated for the adjacent marine and terrestrial -3
ecosystems using chlorophyll a concentration (mg.m ; annual mean outside the estuary mouth) and net primary productivity (gC.m-2.day-1; annual mean around the estuary), respectively.
The SAR-hypothesis was evaluated with three variables, namely: estuary area (km2), continental shelf width as a measure of habitat area of the adjacent marine ecosystem (m; minimum distance to 2
continental shelf limit), and drainage basin area (km ) as a measure of habitat area of the adjacent freshwater ecosystem. The ecosystem-connectivity hypothesis was investigated using three variables, namely: estuary type (open, temporarily open), estuary mouth width (m) and tidal regime [microtidal (4m)].
Finally, to explore the hypothesis of habitat suitability we considered two variables: estuary salinity type [regular (40), hyperhaline (typically with areas >40)]; and annual river flow (m3.s-1; mean annual river flow).
This article is protected by copyright. All rights reserved.
Accepted Article
To deal with incompletely sampled assemblages, species richness can be estimated based on species accumulation curves, species abundance distributions or richness estimators (Magurran 2004). However, most studies in the database published total species richness which excluded the use of these methods. To control for sampling effects, sampling effort was used as a covariate in statistical analysis, following Dunn et al. (2009) as sustained by Gotelli and Colwell (2011). Specifically, sampling effort was determined as the total number of samples. Furthermore, to detect sampling bias due to sampling gear selectivity, the number of fishing gear types was determined.
Data analysis Generalized Linear Models (GLMs) were fitted to quantify the variation of fish species richness at estuary scale across large extents as a function of historic and environmental characteristics of estuaries. Species richness was the response variable, and predictor variables were both continuous (latitude, temperature, chlorophyll a, net primary productivity, estuary area, continental shelf width, drainage basin area, estuary mouth width, river flow) and categorical (realm, continent, estuary type, tidal regime, salinity type). In addition, to account for sampling effects, total sample number was included as an obligatory continuous covariate, and number of fishing gear types as a candidate continuous covariate. Covariates were fourth root transformed to reduce right skewness and the effect of extreme observations (Clarke & Warwick 2001; Zuur, Ieno & Smith 2007). GLMs are an extension of linear models which incorporate non-normal distributions of the response variable and transformations of dependent variables to linearity. Poisson GLMs allow dealing with the heterogeneity and non-negative values of count data (Zuur, Ieno & Smith 2007) but preliminary analysis revealed over-dispersion (model variance exceeds the mean), therefore negative binomial was used since its variance is a function of the mean (Ver Hoef & Boveng 2007) (packages stats and MASS; R). GLMs were used as there was no spatial autocorrelation in model residuals, measured with Moran’s I tests (Dormann et al. 2007) (packages ncf and spdep; R).
To evaluate the effects of spatial extent, GLMs were fitted hierarchically: (1) models were fitted to the whole database; (2) the database was partitioned by realm and models were fitted to each data subset (nine realms; one realm had n=2 and was excluded from analysis); (3) the whole database was partitioned by continent and models were fitted to each data subset (six continents).
This article is protected by copyright. All rights reserved.
Accepted Article
GLMs were built with stepwise selection: predictor variables were sequentially added to and dropped from a model until Akaike information criterion (AIC) or residual deviance were not reduced by adding variables. GLMs were evaluated based on explanatory power (explained deviance), 2
goodness-of-fit (AIC) and predictive capacity (r of linear model of predicted versus observed species richness), while accounting for parsimony (low number of variables). The contribution and significance of each predictor variable in a GLM were tested by sequentially adding variables to the model (package stats; R).
Pair-wise associations between continuous variables were evaluated using Pearson’s correlations (package stats; R). Aiming at a robust analysis, several candidate predictors were excluded from GLM analysis due to high or moderate correlation with other predictors (Appendix S3 - Table S1), namely: latitude (r=0.74 with temperature), chlorophyll a (r=0.45 with continental shelf width), estuary mouth width (r=0.79 with estuary area), drainage basin area (r=0.72 with estuary area), and river flow (r=0.86 with drainage basin area and r=0.84 with estuary area). The variable “number of sampling gear types” was excluded from GLM analysis as it revealed very low and inconsistent relationships with species richness in preliminary analysis.
One-way analysis of variance and Tukey’s honest significant difference post-hoc tests were used to test differences in mean species richness among sites with different categorical predictor variables (marine biogeographic realm, continent, estuary type, salinity type and tidal regime) (package stats; R). All statistical analysis were run in R software (R Development Core Team 2008), and a significance level of 0.05 was considered.
Results Estuaries included in the analysis were distributed from tropical to temperate latitudes (1º to 59º), from cold to warm mean sea surface temperature (3 ºC to 31 ºC), in regions with low to high terrestrial net primary productivity and along coasts with low to high chlorophyll a concentration (Appendix S3 Fig. S1). Substantial variation was present in estuary mouth width, estuary area, drainage basin area and river flow among estuaries across sampled realms and continents. Most estuarine systems were
This article is protected by copyright. All rights reserved.
Accepted Article
open (71%), and temporarily open estuaries were mostly located in Tropical Eastern Pacific, Temperate Southern Africa and Temperate Australasia realms. In general, estuaries with larger area had wider mouth/s, larger drainage basin area and higher river flow (Appendix S3 - Table S1). Estuaries were located in coastal regions with very narrow (e.g. Iberian Coast) to wide continental shelf (e.g. North Sea) (Appendix S3 - Fig. S1), and chlorophyll a concentration outside the estuary generally increased with continental shelf width (Appendix S3 - Table S1). Most estuaries were microtidal (69%) and regular regarding salinity type (89%) (Appendix S3 - Fig. S1).
Fish species richness varied notably among estuaries (mean 29, maximum 214). GLMs fitted at the global extent attained high explanatory power (explained deviance 63%) and predictive capacity (r2=0.48) (Table 1). In this global model, the variable realm explained a high proportion of variance (13%), followed by temperature, estuary type, continent, and net primary productivity (explained deviance 4%-1%) (Table 1). Furthermore, sampling effort accounted for 39% of explained deviance, and the remaining statistically significant predictor variables (estuary area and continental shelf width) had low explained deviance (
Received Date : 28-Jul-2014 Accepted Date : 13-Mar-2015 Article type
: Standard Paper
Global patterns and predictors of fish species richness in estuaries Rita P. Vasconcelos*, Sofia Henriques, Susana França, Stéphanie Pasquaud, Inês Cardoso, Marina Laborde, Henrique N. Cabral MARE - Marine and Environmental Sciences Centre, Faculdade de Ciências, Universidade de Lisboa, Campo Grande, 1749-016 Lisboa, Portugal
*corresponding author: e-mail [email protected]
Running headline Global predictors of fish diversity in estuaries
Abstract (1) Knowledge on global patterns of biodiversity and regulating variables is indispensable to develop predictive models. (2) The present study used predictive modelling approaches to investigate hypotheses that explain the variation in fish species richness between estuaries over a worldwide spatial extent. Ultimately, such models will allow assessment of future changes in ecosystem structure and function as a result of environmental changes. (3) A comprehensive worldwide database was compiled on the fish assemblage composition and environmental characteristics of estuaries. Generalized Linear Models were used to quantify how variation in species richness among estuaries is related to historical events, energy dynamics, and ecosystem characteristics, whilst controlling for sampling effect. (4) At the global extent, species richness differed among marine biogeographic realms and continents, and increased with mean sea surface temperature, terrestrial net primary productivity, and the stability of connectivity with marine ecosystem (open versus temporarily open estuaries). At a smaller extent (within marine biogeographic realm or continent) other characteristics were also
This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/1365-2656.12372 This article is protected by copyright. All rights reserved.
Accepted Article
important in predicting variation in species richness, with species richness increasing with estuary area and continental shelf width. (5) The results suggest that species richness in an estuary is defined by predictors that are spatially hierarchical. Over the largest spatial extents species richness is influenced by the broader distributions and habitat use patterns of marine and freshwater species that can colonize estuaries, which are in turn governed by history contingency, energy dynamics and productivity variables. Species richness is also influenced by more regional and local parameters that can further affect the process of community colonization in an estuary including the connectivity of the estuarine with the adjacent marine habitat, and, over smaller spatial extents, the size of these habitats. In summary, patterns of species richness in estuaries across large spatial extents seem to reflect from global to local processes acting on community colonization. The importance of considering spatial extent, sampling effects and of combining history and contemporary environmental characteristics when exploring biodiversity is highlighted.
Key-words coastal, community, alpha-diversity, meta-analysis, review, predictive models, species distribution models.
Introduction Estuaries are among the most biologically productive and valuable ecosystems worldwide (Costanza et al. 1997), yet are threatened by an increase in human activities which may affect ecosystem health, functions and services (Barbier et al. 2011). As in marine environments, human impacts have depleted species and habitats, degraded water quality and accelerated species invasions (Lotze et al. 2006). This loss of marine biodiversity is increasingly impairing ecosystem services, namely food provision, water quality maintenance and recovery from perturbations (Worm et al. 2006). Understanding biodiversity patterns and their underlying processes is therefore important when investigating the consequences of biodiversity loss to ecosystem function and services. Biodiversity is viewed as the variety of life, embracing variation from genes to ecosystems, and is commonly measured as species richness (Purvis & Hector 2000). Global patterns of biodiversity in estuaries have been largely unexplored, despite the value of estuarine ecosystems and the existing knowledge on variability of fish communities within estuaries and their environmental drivers (see
This article is protected by copyright. All rights reserved.
Accepted Article
reference works Whitfield 1998; Blaber 2000; Elliott & Hemingway 2002; Able & Fahay 2010). As transitional systems, estuaries establish links with marine and freshwater ecosystems (Beck et al. 2001; Attrill & Power 2002), and persistent environmental fluctuations place considerable physiological demands on the species that utilize them (e.g Elliott & Quintino 2007). Fish communities in estuaries include estuarine resident species, and also marine and freshwater species which enter estuaries eitheras migrants or stragglers, as well as diadromous and amphidromous species (e.g. Elliott et al. 2007). The fish communities in estuaries tend to be dominated by a few persistent and abundant core species (e.g. Magurran & Henderson 2003), and their species richness is lower than in the adjacent marine environment (e.g. Martino & Able 2003). The main environmental driver of the structure of fish communities within estuaries is the longitudinal salinity gradient (e.g. Sosa-López et al. 2007; Whitfield et al. 2012).
The scarcity of knowledge on global richness patterns and predictors for estuarine ecosystems contrasts remarkably with adjacent marine and freshwater ecosystems (Oberdorff, Guégan & Hugueny 1995; Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013). Investigation of the environmental correlates of fish richness among estuaries is fragmented, and it has generally been based on region-specific approaches, without standardization for sampling effects and at mismatched spatial extents. This knowledge gap occurs for most taxa in estuaries, except for mangroves (Polidoro et al. 2010) and macroinvertebrates (Attrill, Stafford & Rowden 2001).
The present study aims to identify patterns of variation in fish species richness among estuaries distributed across large geographical extents, and to disentangle the relative importance of underlying predictors. Several hypotheses to explain spatial variation of species richness patterns in estuaries were considered, some of which have been repeatedly identified across taxa and ecosystems (Gaston 2000).
The evolutionary history hypothesis states that evolutionary processes have resulted in different faunal composition of oceanic basins, leading to differences in species richness (e.g. Floeter et al. 2008; Tittensor et al. 2010). The evolutionary processes that have generated these differences affect rates of speciation and adaptation to available habitats. Accordingly, we hypothesize that across large
This article is protected by copyright. All rights reserved.
Accepted Article
extents, estuaries with faunas possessing different evolutionary histories will fundamentally differ in species richness, as can be seen across small spatial extents for estuaries (e.g. Monaco, Lowery & Emmett 1992; Pease 1999; Harrison & Whitfield 2006), and globally for coastal and freshwater ecosystems (Oberdorff, Guégan & Hugueny 1995; Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013).
A latitudinal gradient in species richness is a ubiquitous ecological pattern, with numerous explanatory mechanisms proposed (Willig, Kaufman & Stevens 2003; Gaston 2007). Among these, the species-energy hypothesis advocates that metabolic rate increases with temperature and promotes higher speciation rates leading to greater diversity. Additionally it is possible that species distribution ranges are largely set by thermal tolerance, with more species tolerant of warm temperatures (Gaston 2000). Based on evidence for estuaries (Pease 1999; Harrison & Whitfield 2006; Nicolas et al. 2010; França, Costa & Cabral 2011) and coastal and freshwater systems (Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013) we hypothesize that, across large extents, species richness in estuaries decreases with latitude and increases with temperature. It is possible that differences in species richness between estuaries at large spatial extents relate to primary productivity, as previously supported for freshwater fish (Oberdorff, Guégan & Hugueny 1995). This productivity-richness hypothesis suggests a positive effect of primary productivity on species richness by allowing larger populations to persist, thereby reducing extinction risk and supporting a higher diversity of niche specialists (Willig, Kaufman & Stevens 2003; Tittensor et al. 2010). The hypothesis of Species-Area Relationships (SAR) assumes a change in species numbers with increasing area, and has been termed one of the few laws in ecology. Essentially, larger sampled areas support higher number of individuals and therefore increase the probability of encountering additional species, as well as increased chances of finding environmental heterogeneity and species that differ in their niches (Pihl et al. 2002; Scheiner 2003; Guilhaumon et al. 2008; Guilhaumon et al. 2012). We consider the possibility that species richness in estuaries can increase with habitat area (Horn & Allen 1976; Monaco, Lowery & Emmett 1992; Nicolas et al. 2010). Additionally, since assemblages in estuaries include marine, freshwater and estuarine species, then differences in species richness between widely distributed estuaries may also relate to habitat area in the adjacent
This article is protected by copyright. All rights reserved.
Accepted Article
coastal and freshwater ecosystems (Nicolas et al. 2010). It is known that species richness in those ecosystems is dependent upon habitat area (Tittensor et al. 2010; Parravicini et al. 2013; Tisseuil et al. 2013). Species richness in estuaries is commonly dominated by marine species (e.g. Whitfield 1999; Franco et al. 2008), and therefore colonization processes and isolation should influence species richness. We hypothesize that patterns of species richness in estuaries across large spatial extents relate to connectivity between estuaries and the marine ecosystem. Species richness has been shown to be higher in those estuaries that are permanently open to the marine environment (Pease 1999; Harrison & Whitfield 2006), and that have a greater mouth size (Horn & Allen 1976; Monaco, Lowery & Emmett 1992; Nicolas et al. 2010). By contrast lower percentages of marine species have been found in fish communities of microtidal estuaries (Young & Potter 2003).
Finally, fish assemblages in estuaries are naturally faced with environmental stress, particularly related to salinity (Elliott & Quintino 2007; Whitfield et al. 2012). Thus it is possible that salinity is linked with patterns of species richness at large spatial extents. Increases in river flow in an estuary can lead to decreased suitability of estuarine habitat for marine species and consequently decreased species richness (Whitfield & Harrison 2003), while hyperhaline areas within estuaries have also been shown to have lower species richness (Whitfield et al. 2012).
This study relies on a comprehensive database on fish assemblages and environmental characteristics of estuaries worldwide, based on studies at single estuary scale. Predictor variables were identified to address the hypotheses above and a statistical modelling approach was applied first at a global extent, and then at a smaller geographical extent (each marine biogeographic realm and continent). The results should improve our ability to predict biodiversity at the estuary scale, which is paramount in view of continued environmental change and biodiversity loss. Materials and Methods Database A database was built combining published data on: (a) fish species assemblages in estuaries; and (b) historical and environmental characteristics of those estuaries (details in Appendix S1 and S2). In this database, a sample consists of the total list of fish species sampled in a given estuary and study.
This article is protected by copyright. All rights reserved.
Accepted Article
The obtained database has 786 samples for 430 estuaries worldwide (Fig.1; details in Appendix S1 and S2). To evaluate the evolutionary history hypothesis two factors were included in the database: marine biogeographic realm - the largest spatial unit of the Marine Ecoregions of the World system (Spalding et al. 2007) and hereafter designated as realm (Arctic, Temperate Southern Africa, Temperate Northern Atlantic, Temperate Northern Pacific, Temperate South America, Temperate Australasia, Tropical Atlantic, Tropical Eastern Pacific, Western Indo-Pacific, Central Indo-Pacific); and continent (Africa, Europe, North America, South America, Asia, Oceania). Contemporary environmental conditions and habitat distribution are poor predictors of marine faunal breaks (Keith et al. 2013), which supports the use of a biogeographic classification scheme to evaluate the evolutionary history hypothesis.
To address the latitude-hypothesis, absolute latitude was determined (º; at estuary mouth), whilst for the energy-hypothesis temperature was determined (ºC; annual mean sea surface temperature outside the estuary mouth). Regarding the productivity-hypothesis, and as comparable data for estuaries are rare, primary productivity was evaluated for the adjacent marine and terrestrial -3
ecosystems using chlorophyll a concentration (mg.m ; annual mean outside the estuary mouth) and net primary productivity (gC.m-2.day-1; annual mean around the estuary), respectively.
The SAR-hypothesis was evaluated with three variables, namely: estuary area (km2), continental shelf width as a measure of habitat area of the adjacent marine ecosystem (m; minimum distance to 2
continental shelf limit), and drainage basin area (km ) as a measure of habitat area of the adjacent freshwater ecosystem. The ecosystem-connectivity hypothesis was investigated using three variables, namely: estuary type (open, temporarily open), estuary mouth width (m) and tidal regime [microtidal (4m)].
Finally, to explore the hypothesis of habitat suitability we considered two variables: estuary salinity type [regular (40), hyperhaline (typically with areas >40)]; and annual river flow (m3.s-1; mean annual river flow).
This article is protected by copyright. All rights reserved.
Accepted Article
To deal with incompletely sampled assemblages, species richness can be estimated based on species accumulation curves, species abundance distributions or richness estimators (Magurran 2004). However, most studies in the database published total species richness which excluded the use of these methods. To control for sampling effects, sampling effort was used as a covariate in statistical analysis, following Dunn et al. (2009) as sustained by Gotelli and Colwell (2011). Specifically, sampling effort was determined as the total number of samples. Furthermore, to detect sampling bias due to sampling gear selectivity, the number of fishing gear types was determined.
Data analysis Generalized Linear Models (GLMs) were fitted to quantify the variation of fish species richness at estuary scale across large extents as a function of historic and environmental characteristics of estuaries. Species richness was the response variable, and predictor variables were both continuous (latitude, temperature, chlorophyll a, net primary productivity, estuary area, continental shelf width, drainage basin area, estuary mouth width, river flow) and categorical (realm, continent, estuary type, tidal regime, salinity type). In addition, to account for sampling effects, total sample number was included as an obligatory continuous covariate, and number of fishing gear types as a candidate continuous covariate. Covariates were fourth root transformed to reduce right skewness and the effect of extreme observations (Clarke & Warwick 2001; Zuur, Ieno & Smith 2007). GLMs are an extension of linear models which incorporate non-normal distributions of the response variable and transformations of dependent variables to linearity. Poisson GLMs allow dealing with the heterogeneity and non-negative values of count data (Zuur, Ieno & Smith 2007) but preliminary analysis revealed over-dispersion (model variance exceeds the mean), therefore negative binomial was used since its variance is a function of the mean (Ver Hoef & Boveng 2007) (packages stats and MASS; R). GLMs were used as there was no spatial autocorrelation in model residuals, measured with Moran’s I tests (Dormann et al. 2007) (packages ncf and spdep; R).
To evaluate the effects of spatial extent, GLMs were fitted hierarchically: (1) models were fitted to the whole database; (2) the database was partitioned by realm and models were fitted to each data subset (nine realms; one realm had n=2 and was excluded from analysis); (3) the whole database was partitioned by continent and models were fitted to each data subset (six continents).
This article is protected by copyright. All rights reserved.
Accepted Article
GLMs were built with stepwise selection: predictor variables were sequentially added to and dropped from a model until Akaike information criterion (AIC) or residual deviance were not reduced by adding variables. GLMs were evaluated based on explanatory power (explained deviance), 2
goodness-of-fit (AIC) and predictive capacity (r of linear model of predicted versus observed species richness), while accounting for parsimony (low number of variables). The contribution and significance of each predictor variable in a GLM were tested by sequentially adding variables to the model (package stats; R).
Pair-wise associations between continuous variables were evaluated using Pearson’s correlations (package stats; R). Aiming at a robust analysis, several candidate predictors were excluded from GLM analysis due to high or moderate correlation with other predictors (Appendix S3 - Table S1), namely: latitude (r=0.74 with temperature), chlorophyll a (r=0.45 with continental shelf width), estuary mouth width (r=0.79 with estuary area), drainage basin area (r=0.72 with estuary area), and river flow (r=0.86 with drainage basin area and r=0.84 with estuary area). The variable “number of sampling gear types” was excluded from GLM analysis as it revealed very low and inconsistent relationships with species richness in preliminary analysis.
One-way analysis of variance and Tukey’s honest significant difference post-hoc tests were used to test differences in mean species richness among sites with different categorical predictor variables (marine biogeographic realm, continent, estuary type, salinity type and tidal regime) (package stats; R). All statistical analysis were run in R software (R Development Core Team 2008), and a significance level of 0.05 was considered.
Results Estuaries included in the analysis were distributed from tropical to temperate latitudes (1º to 59º), from cold to warm mean sea surface temperature (3 ºC to 31 ºC), in regions with low to high terrestrial net primary productivity and along coasts with low to high chlorophyll a concentration (Appendix S3 Fig. S1). Substantial variation was present in estuary mouth width, estuary area, drainage basin area and river flow among estuaries across sampled realms and continents. Most estuarine systems were
This article is protected by copyright. All rights reserved.
Accepted Article
open (71%), and temporarily open estuaries were mostly located in Tropical Eastern Pacific, Temperate Southern Africa and Temperate Australasia realms. In general, estuaries with larger area had wider mouth/s, larger drainage basin area and higher river flow (Appendix S3 - Table S1). Estuaries were located in coastal regions with very narrow (e.g. Iberian Coast) to wide continental shelf (e.g. North Sea) (Appendix S3 - Fig. S1), and chlorophyll a concentration outside the estuary generally increased with continental shelf width (Appendix S3 - Table S1). Most estuaries were microtidal (69%) and regular regarding salinity type (89%) (Appendix S3 - Fig. S1).
Fish species richness varied notably among estuaries (mean 29, maximum 214). GLMs fitted at the global extent attained high explanatory power (explained deviance 63%) and predictive capacity (r2=0.48) (Table 1). In this global model, the variable realm explained a high proportion of variance (13%), followed by temperature, estuary type, continent, and net primary productivity (explained deviance 4%-1%) (Table 1). Furthermore, sampling effort accounted for 39% of explained deviance, and the remaining statistically significant predictor variables (estuary area and continental shelf width) had low explained deviance (
