Introduction

In wetlands, hydrology is an important driver of ecosystem change because it maintains both the abiotic and biotic components that contribute to the defining features and function of wetland habitats (Mitsch and Gosselink 2007). Many wetlands undergo ecosystem-scale changes as anthropogenic agents, including accelerating rates of land-use and climate change (Batzer and Sharitz 2006), as well as restoration efforts, alter their hydrology. To predict the ecosystem effects of hydrologic changes and to direct adaptive assessment efforts, it is important to understand the complex relationships between hydrology and biotic assemblages. Because biotic assemblages can actively change wetland hydrology and physicochemistry through feedbacks (Mitsch and Gosselink 2007), hydrologic alterations by restoration efforts could modify these relationships if assemblage changes occur.

Bioassessment tools permit interpretation of the direction of environmental change and are useful for developing management targets to inform and assess restoration efforts (EPA 2011a). Because diatom assemblages are sensitive to environmental changes, including hydrologic changes, they can be used to develop robust inference models for early detection of shifts due to habitat degradation or restoration projects (Davis et al. 1996; Stevenson 1998). Sensitivity to hydrologic changes have been shown by the immediate recovery of diatom production in desiccated periphyton (benthic algae; Browder et al. 1994) upon rehydration (Thomas et al. 2006), and the influence of hydroperiod on the relative abundance of desiccation-resistant species inhabiting the periphyton matrix (Gottlieb et al. 2005). Periphyton is an important indicator of wetland degradation and restoration progress because it undergoes rapid and measurable structural, functional, and compositional responses to environmental changes (especially hydrology and water quality; Browder et al. 1994) at a large range of spatial scales (McCormick and Stevenson 1998; Gaiser 2009).

Diatom-based assessments conducted over long time periods should be used to inform wetland adaptive assessment tools about the rates, causes, and functional consequences of change to improve future management strategies (Gaiser and Rühland 2010). Because diatoms persist in the fossil record, they provide quantitative measures of past conditions that give ecologically-based context for current and future management plans (Smol and Stoermer 2010). Along with long-term monitoring, reconstructing the hydrologic history of wetlands is important, as past settings can provide hydrologic targets for restoration (Marshall et al. 2009), particularly in the face of changes in rainfall and inundation due to climate change (Mulholland et al. 1997). In addition to assessment over time, the inherent spatial attribute of diatom-based models provide an opportunity to examine responses at multiple spatial scales. The degree of consistency of species responses to ecosystem properties and the distribution of environmental gradients among hydrologically connected wetland subbasins determine whether models should be basin-specific (e.g., Gaiser et al. 2006) or regional.

Wetland assessment programs have lagged behind other aquatic systems, such as lakes and streams (EPA 2011b). Robust diatom-based inference models will provide a means for examining both the drivers and biological feedbacks of regulatory variables at a range of spatial scales. Most diatom-based assessments of wetlands have addressed water quality parameters such as nutrients (Pan and Stevenson 1996; Cooper et al. 1999; Gaiser et al. 2006; Lane and Brown 2007; Velinsky et al. 2011; La Hée and Gaiser 2012), salinity (Taffs 2001; Hicks and Nichol 2007; Wachnicka et al. 2010, 2011), dissolved oxygen (Zhang et al. 2011), and pH (Owen et al. 2004; Taffs et al. 2008). Some studies have employed diatoms to assess wetland vegetation (Huvane and Cooper 2001; Gaiser et al. 2005a), and several have employed diatom-based assessment of wetland hydrologic characteristics such as hydroperiod (Gaiser et al. 1998; Mackay et al. 2011). To date, inference models that reflect biological feedbacks, such as periphyton characteristics, to hydrologic conditions have not been developed. Reconstructions of ecosystem properties using regional inference models were successful for several wetlands (Gaiser et al. 1998; Taffs 2001; Owen et al. 2004), while other studies used or advocated basin-specific models (Cooper et al. 1999; Gaiser et al. 2006; La Hée and Gaiser 2012).

In the Florida Everglades, restoration goals, targets, and assessment tools are at the forefront of scientifically-sound adaptive management of the multi-billion dollar, landscape-scale Comprehensive Ecosystem Restoration Plan (C & SF Project 1999). Restoration efforts propose to redirect unused freshwater to areas that historically had greater periods of inundation, as well as to areas that would benefit cities and farmers. Without robust inference models, however, changes to hydrology could result in unintended consequences for wetland structure and function (e.g., Surratt et al. 2012). While development of diatom response models to water quality parameters such as periphyton total phosphorus (TP) concentrations (McCormick et al. 1996; Pan et al. 2000; Gaiser et al. 2006; La Hée and Gaiser 2012) have allowed paleoecological reconstructions of historical environments (e.g., Cooper et al. 1999; Slate and Stevenson 2000), hydrologic reconstructions have been impeded by the lack of robust diatom-based inference models (but see Sanchez et al. 2012). Also, diatom responses to periphyton characteristics reflective of biological feedbacks and responses to hydrologic changes have not been defined. If specific diatom assemblages inhabit periphyton of differing characteristics (e.g., biovolume), diatoms could be used to infer past wetland conditions even though the characteristics of the periphyton itself are not conserved in monitoring or paleoecological records. The numerous biogeochemical processes attributed to periphyton make the maintenance of native periphyton structure an important aspect of Everglades restoration (Hagerthey et al. 2011). Examination of periphyton along hydroperiod gradients in a marl-based wetland area found little effect of hydroperiod on periphyton biomass (Wachnicka, unpublished data), but patterns across the long hydrologic gradients present in the larger Everglades landscape, including the hydrologically different (but connected) Water Conservation Areas (WCAs) and Everglades National Park (ENP) (Light and Dineen 1994), are expected. Extensive studies have been conducted on the periphyton and paleoecological indicators of the historically P-enriched and hydrologically-managed WCA-2A (e.g., McCormick et al. 1996, 1998; Cooper et al. 2008). However, a quantitative evaluation of diatom sensitivity to periphyton characteristics occurring across the broader hydrologic gradients represented in the Greater Everglades could elucidate landscape-scale responses to hydrologic change.

In this study, we examined the diatom assemblages of periphyton samples that were gathered in 2006 as part of a landscape-scale monitoring program across the Everglades. Our first objective was to quantify diatom assemblage patterns and associations with habitat characteristics. We used the term ‘habitat characteristics’ to include both the abiotic physicochemical factors of the environment and the biological features of the periphyton mat that influence, and are influenced by, the diatoms embedded in the mat matrix. We hypothesized that hydroperiod, periphyton biovolume, and periphyton TP concentration would be associated with the differences in diatom assemblages across the Everglades. Our second objective was to test whether the responses of diatom assemblages across the Everglades were consistent throughout the landscape. We hypothesized that a spatially extensive dataset would enable detection of basin-specific responses, if they exist, but also provide an opportunity to develop landscape-scale models. Our final objective was to evaluate the predictive capacity of diatom-based inference models. In addition, we mapped inferences relative to actual patterns to illustrate the utility of visualizing landscape-scale patterns for application in restoration planning and management.

Methods

Site Description

Located in Florida, USA, the Everglades is an expansive wetland (>6,000 km2) encompassing a mosaic of environmental conditions resulting from natural gradients and differences in biogeochemistry, as well as anthropogenic compartmentalization and eutrophication (Davis and Ogden 1994). Owing to the limestone bedrock underlying much of the southern Everglades, periphyton assemblages can contain an abundance of calcium carbonate precipitates that allows the formation of thick, calcareous mats. However, areas of higher nutrient levels contain organic, filamentous films that are easily disrupted (McCormick and O’Dell 1996; McCormick et al. 2001; Gaiser et al. 2005b, 2011). This study focuses on calcareous communities occurring throughout the spatially complex WCAs and ENP, exclusive of soft-water regions (the Arthur R. Marshall Loxahatchee National Wildlife Refuge, also known as WCA-1) with a very different algal assemblage and suite of environmental stressors (Harvey and McCormick 2009; Gaiser et al. 2011; Hagerthey et al. 2011). The Everglades has a subtropical climate with a distinct wet and dry season, the duration of which are controlled by climate variability and water management and can have significant influences on variability in species abundances and ecosystem properties.

Sample Collection and Processing

Periphyton samples were collected during the 2006 wet season (September through December) as part of the Monitoring and Assessment Program of the Comprehensive Everglades Restoration Plan (RECOVER 2004). A total of 86 sites were included in this study (refer to Fig. 5; see Online Resource for exact locations). Generalized random-tessellation stratification (Stevens and Olsen 2004) was used to choose a spatially balanced set of sampling locations. The landscape was divided into 800 m × 800 m grids and a representative sample of these was drawn as primary sampling units (PSU); three sampling sites were randomly selected from the samplable habitat in each PSU (Philippi 2005). Samplable habitat included all locations where vegetation was not too dense for our sampling device to enclose 1 m3 of the water column and less than 1 m deep; primarily wet prairies and sloughs met these conditions (Gunderson 1994).

At each sampling location, water depth was measured, and water samples were taken for measurement of pH and conductivity (μS cm−1). Other hydrologic variables, including hydroperiod (days flooded) and days since dry (number of days since flooding of the marsh surface after the latest drying event when water levels were <5 cm), were estimated by calibration to nearby continuous water level gauges using digital elevation models provided by the Everglades Depth Estimation Network (EDEN) (http://sofia.usgs.gov/eden/stationlist.php). The sampling device used to delineate the sampling area was a 1 m3 enclosure with mesh sides and open on the top and bottom (Jordan et al. 1997). Visual assessment for aerial cover (percent of the surface of the enclosed area covered by periphyton) was conducted before all periphyton within the enclosure was collected and measured for biovolume using a perforated graduated cylinder. Extraneous plant matter, animals, and other debris were on average less than 0.5 % of the volume of the sample and did not affect biovolume measurements. If no benthic, epiphytic, or metaphytic periphyton was present, flocculent detritus from the benthos was collected (Troxler and Richards 2009; Pisani et al. 2011).

Periphyton samples were taken back to the laboratory and frozen before further processing. Animals, plant matter, and other debris were removed, and subsamples were taken for the measurement of dry weight (g m−2) by drying at 80 °C to constant weight, ash-free dry mass (g m−2) by combustion at 500 °C for 1.5 h, chlorophyll a mass (μg m−2) and chlorophyll a concentration (μg g−1 dry weight) by fluorometry (Welschmeyer 1994), total periphyton P (μg g−1 dry weight) by colorimetry after dry combustion (Solorzano and Sharp 1980; EPA 1983), and diatom species composition analysis. Organic content was calculated as the ratio of ash-free dry mass to total dry mass expressed as a percent. Periphyton P concentrations were used as a metric of P availability because periphyton P has a strong correlation with P load (Gaiser et al. 2004, 2005b, 2006). See Online Resource for full dataset of all habitat characteristics mentioned above. Diatom samples were cleaned of calcite and organic matter using strong acids and chemical oxidizers (Hasle and Fryxell 1970), and then permanently affixed to glass slides using Naphrax®. A minimum of 500 valves were counted and identified per slide (Weber 1973) using a compound light microscope at 1,000× magnification. Identifications were made to the lowest taxonomic level possible (variety or forma) using a database of South Florida diatom taxa (http://fce.lternet.edu/data/database/diatom) and other references (e.g., Slate and Stevenson 2007).

Data Analysis

Species abundances were relativized by the maximum abundance achieved by each species over all samples to reduce the differential impact of common and uncommon species on inferences. All species were included in the analyses after checking for outliers more than two standard deviations from the mean Sørenson distance measure (McCune and Grace 2002). Three categories of habitat characteristics were analyzed: hydrology, periphyton abundance, and periphyton quality (Table 1). Hydrology included days since dry (DSD), hydroperiod (HYPER), and water depth (DEPTH). Periphyton abundance included periphyton biovolume (PBIOV), aerial cover (AERCO), chlorophyll a mass (CHLMA), dry weight (DRYWT), and ash-free dry mass (AFDM). Periphyton quality included periphyton organic content (ORGCO), periphyton total phosphorus (TP), chlorophyll a concentration (CHLCO), water column pH (pH), and water column conductivity (CONDU). We used the term ‘periphyton quality’ to include measurements of periphyton and water column attributes because of the intimate relationship between water quality (nutrient and ion concentrations) and the composition and function of periphyton mats, including both the influence of the water column on mats and the influence of mats on the surrounding water column (Gottlieb et al. 2005; Thomas et al. 2006; Hagerthey et al. 2011). Each habitat characteristic was transformed to bring skewness (a metric to assess normality) closest to zero, including square root, arcsine square root, tenth root, and log transformations (McCune and Grace 2002) (Table 1). Samples more than two standard deviations from the mean Euclidean distance measure were considered outliers and removed from the dataset; subsequent analysis of species confirmed the absence of outlier species. Samples were categorized a priori into three wetland subsets based on landscape pattern and management practices: Shark River Slough (SRS), Water Conservation Areas 2A and 2B (WCA-2), and Water Conservation Areas 3A and 3B (WCA-3).

Table 1 Summary statistics of habitat characteristics (before transformations) and transformations used to bring skewness of values closest to zero
Full size table

To examine diatom assemblage patterns associated with gradients of habitat characteristics, we used non-metric multidimensional scaling (NMS) ordination (Kruskal and Wish 1978; Minchin 1987) with the Bray-Curtis dissimilarity measure (Bray and Curtis 1957). Because our data are observational, we did not attempt to identify causal relationships between assemblages and habitat characteristics. The statistical program DECODA (Database for Ecological Community Data; Minchin 1990) was used to run ordinations and to fit vectors of maximum correlation of habitat characteristics with assemblage difference. To detect whether assemblages had consistent responses to habitat characteristics throughout the landscape or if there were distinct regional differences, analysis of similarity (ANOSIM) was conducted using Primer-E statistical software (version 6.0) to determine differences in diatom assemblages among and within the three wetland subsets (Clarke and Gorley 2006). Following the results of the ANOSIM, separate ordinations of wetland subsets were conducted to determine any subset-specific assemblage patterns. To test the null hypothesis that species responses to habitat characteristics are not associated among subsets, we used Kendall’s coefficient of concordance with correction for tied ranks, W c (Zar 1996).

We used the Bio-Env (Biota-Environment) procedure in Primer to determine the habitat characteristics most correlated with differences in diatom assemblages (Clarke and Ainsworth 1993). Bio-Env determines the correlation (Spearman) between the ranked dissimilarities of all possible pairs of samples based on the assemblage matrix and the ranked dissimilarities based on subsets of the environmental (habitat characteristics) matrix (Clarke and Warwick 2001). Rank correlation ρ = 1 indicates complete agreement, whereas ρ = −1 indicates complete opposition (Clarke and Warwick 2001; Clarke and Gorley 2006). Because Bio-Env attempts to match all possible combinations of habitat characteristics to explain differences in the assemblage data, correlations could be found by chance. We tested the null hypothesis that the biota and the environment are not linked by using the BEST (Biota Environment STepwise) permutations test (999 permutations) in Primer to determine statistical significance (Clarke et al. 2008).

For the selected habitat characteristics, weighted averaging calibration models were constructed using C2 software to obtain species optima and tolerance values (Juggins 2005). The weighted averaging models were tested by simulating prediction errors using the bootstrapping resampling method repeated 1000 times (Efron 1982; Birks et al. 1990). The resulting Root Mean Square Error of Prediction (RMSEP) and bootstrapped (boot) r 2 were used as more conservative estimates of model fit compared to the apparent Root Mean Square Error (RMSE) and r 2 (Efron 1983; Birks et al. 1990; ter Braak and Juggins 1993; Birks 1995). For variables that were previously transformed, we present the upper and lower limits of tolerance values because of transformation-generated scale changes. For the same reason, we present RMSE and RMSEP values in transformed units. Finally, observed, diatom-inferred, and the difference between the observed and diatom-inferred values were mapped using ArcGIS version 10 (ESRI 2011). Interpolation techniques were not applied due to the inaccuracies that could result from the use of a limited number of sampling points across a large area.

Results

Assemblage Patterns and Habitat Characteristics

We observed 59 species of diatoms representing 21 genera. The most abundant species among all samples were Mastogloia cf. lacustris (also reported as Mastogloia smithii and M. smithii var. lacustris; Gaiser et al. 2010), Encyonema evergladianum, Fragilaria synegrotesca, and Brachysira neoexilis, making up 36 %, 24 %, 12 %, and 9 % of all diatom valves counted, respectively. Mastogloia cf. lacustris was the most abundant species in WCA-3, SRS, and WCA-2 (38 %, 35 %, and 34 % of all valves counted, respectively). There were 15 species common to all wetland subsets that contributed 95 % of the total relative abundance, indicating high overlap of species among regions.

Differences among sites in the relative abundance of species generally grouped samples by wetland subset in the NMS ordination, though overlap was also common (Fig. 1a). A three-dimensional solution resulted in the greatest decrease in minimum stress level (stress = 0.183) and was rotated by degree increments until the relative distribution of the vectors of habitat characteristics could be clearly depicted (Fig. 1a and b). The total rotation was 260°, which also maximized the association between hydroperiod and Axis 1. The percentages of variance represented by Axis 1, 2, and 3 for the rotated ordination were 23 %, 29 %, and 20 %, respectively. Periphyton biovolume, dry weight, AFDM, and TP had the largest magnitude of association with the ordination axes, as indicated by the length of the vectors (refer to Table 4 for exact correlation values). The relative placement and angles of the vectors showed patterns in diatom assemblages were correlated with hydrology (hydroperiod, days since dry, and water depth), periphyton quantity (periphyton biovolume, AFDM, dry weight, aerial cover, and chlorophyll a mass), and periphyton quality (TP, organic content, chlorophyll a concentration, and pH).

Fig. 1
figure 1

Non-metric multidimensional scaling ordination plot (multivariate representation of dissimilarities between samples based on rank order differences in assemblage composition). Three-dimensional ordination shown in two-dimensional plots for easier visualization: Axis 2 vs. 1 (a) and Axis 3 vs. 1 (b). Plots rotated 260° (maximizing association between hydroperiod and Axis 1) to show relative distribution of environmental vectors. Wetland subsets: Shark River Slough (SRS), Water Conservation Areas 2A and 2B (WCA-2), and Water Conservation Areas 3A and 3B (WCA-3)

Full size image

The habitats included in this study ranged from short (min = 39 days) to very long (max = 333 days) hydroperiod sites; periphyton biovolumes were as low as 0 mL m−2 (4 WCA-3 sites and 1 SRS site) to as high as 14,000 mL m−2 (Table 1). Most habitat characteristics were positively skewed; hydroperiod, however, was negatively skewed, indicating a greater number of long hydroperiod sites relative to short hydroperiod sites in our samples (Table 1). Habitat characteristics within the same category (hydrology, periphyton abundance, and periphyton quality) were correlated (p < 0.05), except for conductivity; conductivity was only correlated with pH (Table 2). Organic content and other metrics of periphyton quality were correlated with metrics of periphyton abundance. SRS sites exhibited a negative relationship between hydroperiod and periphyton biovolume, while longer hydroperiod sites in WCA-2 and WCA-3 tended to have mats of higher periphyton biovolume but less organic content (Table 3).

Table 2 Pearson’s correlation matrix of habitat characteristics (prior to transformations). Sample size = 86
Full size table
Table 3 Pairwise correlations between three habitat characteristics describing hydrology (HYPER), periphyton abundance (PBIOV), and periphyton quality (ORGCO) of wetland subsets: SRS (Shark River Slough) WCA-2 (Water Conservation Areas 2A and 2B), and WCA-3 (Water Conservation Areas 3A and 3B)
Full size table

Landscape-Scale Patterns

Relative abundances of diatom species varied consistently among the three wetland subsets according to ANOSIM (Global R = 0.26; p < 0.01). Pairwise tests showed that SRS and WCA-2 were the most different (R = 0.52), while SRS and WCA-3 were most similar (R = 0.17), but differences between all pairs were statistically significant (p < 0.01). Ordinations performed for the subsets (stress = 0.17, 0.03, and 0.16; SRS, WCA-2, and WCA-3, respectively) showed that SRS assemblages had similar associations with habitat characteristics as the combined dataset (Fig. 2a), except for responses to pH (Table 4). The WCA-3 assemblage also showed similar associations as those of the combined dataset (Fig. 2b), except associations with days since dry, hydroperiod, chlorophyll a mass, conductivity and pH were not statistically significant. The WCA-2 assemblage was only associated with TP but, more importantly, had similar placement of vectors relative to each other as in the combined dataset (Fig. 2c). Kendall’s test of concordance between subset-specific species responses to habitat characteristics rejected the null hypothesis that species responses were not associated (W c  = 0.76, χ 2 = 27.4, p < 0.01). Therefore, concordant associations of assemblage patterns among subsets allowed development of landscape-scale metrics using the entire dataset.

Fig. 2
figure 2

Non-metric multidimensional scaling. Three-dimensional ordination of wetland subsets: SRS (a), WCA-2 (b), and WCA-3 (c). Plots rotated to show relative distribution of environmental vectors

Full size image
Table 4 Maximum correlation to vectors of habitat characteristics from three-dimensional NMS of the entire dataset and each subset separately
Full size table

Weighted Averaging Models and Maps

Hydroperiod, days since dry, water depth, periphyton biovolume, TP, and chlorophyll a concentration were selected by the Bio-Env procedure as the subset of habitat characteristics that formed the most correlated (ρ = 0.27) ranked set of dissimilarities with that of the assemblage matrix. The procedure was repeated with forced exclusions because of high correlations between hydroperiod and days since dry, hydroperiod and water depth, and between TP and chlorophyll a concentration (refer to Table 2). In the second run, hydroperiod, periphyton biovolume, and TP were selected as the subset to reach the highest rank correlation (ρ = 0.25), with no improvement with the addition of any number of the remaining variables. Because the observed rank correlation could not be obtained after 999 permutations using randomized environmental matrices, we rejected the null hypothesis that there is no link between the assemblages and the environment (p < 0.001). The variance in assemblage composition explained by hydroperiod, periphyton biovolume, and TP was 12.4 %, 13.6 %, and 17.9 %, respectively (p < 0.005 for all).

Because hydroperiod, periphyton biovolume, and TP best matched assemblage patterns, these three metrics best informed inference models for assessment applications. Our TP weighted averaging models were comparable to previous models generated from Everglades diatom assemblages (Gaiser et al. 2006; La Hée and Gaiser 2012). Thus, we do not reiterate the results in this study. We evaluated the performance of weighted averaging models constructed for hydroperiod and periphyton biovolume by comparing diatom-inferred values with observed values (Fig. 3). Diatom-inferred hydroperiod was correlated with observed values with r 2 = 0.63 (RMSE = 35 days) and boot r 2 = 0.19 (RMSEP = 55 days); diatom-inferred periphyton biovolume was correlated with observed values with r 2 = 0.56 (RMSE = 19 square root mL m−2) and boot r 2 = 0.41 (RMSEP = 23 square root mL m−2) (Table 5).

Fig. 3
figure 3

Diatom-inferred vs. observed hydroperiod (a) and periphyton biovolume (b). Diatom-inferred values calculated using weighted averaging models with bootstrapping cross validation method. Weighted averaging method with the lowest RMSEP (Root Mean Square Error of Prediction) chosen. Strength of model prediction indicated by r 2 of trendline (apparent and bootstrapped r 2 values (boot r 2) shown). Dashed line indicates 1:1 line of 100 % correlation. Axes scaled to back-transformed values for periphyton biovolume

Full size image
Table 5 Performance of weighted-averaging models by linear regression of diatom-inferred values against observed values of entire dataset and each subset separately
Full size table

Hydroperiod optima of diatom species ranged from 99 to 303 days and tolerances ranged from 8 to 81 days above and below optima; periphyton biovolume optima ranged from 0 to 6,000 mL m−2 and tolerances ranged from as low as 0 mL m−2 (where flocculent matter from the benthos, rather than periphyton, was sampled) and as much as 11,000 mL m−2 (Table 6). Pinnularia microstauron was associated with shorter hydroperiod sites, while Encyonema sp. 02 was associated with longer hydroperiod sites; Eunotia naegelii was associated with low periphyton biovolume, while Nitzschia serpentiraphe was associated with high periphyton biovolume, though it had a wide tolerance range of 850 to 8,700 mL m−2 (Fig. 4).

Table 6 Frequency, maximum abundance, and weighted-average optima and tolerance ranges of diatom species for hydroperiod (HYPER) and periphyton biovolume (PBIOV)
Full size table
Fig. 4
figure 4

Relative abundances of diatom species indicating preferences (optima and tolerance ranges) for a range of hydroperiod (a–f) and periphyton biovolume (g–l). Second order polynomial trendlines. Note different y-axis scales for each taxon

Full size image

Maps of observed values, diatom-inferred values, and differences (observed minus inferred values) for hydroperiod (Fig. 5a, b and c) and periphyton biovolume (Fig. 5d, e and f) allowed visualization of the spatial distribution of these important habitat characteristics and the accuracy of diatom-based inferences. The highest predicted values of hydroperiod were in WCA-3, while the northern part of SRS had the lowest predicted values. In SRS, very long hydroperiod conditions (>280 days) were underestimated and shorter hydroperiod conditions (<190 days) were overestimated. Inferred periphyton biovolume estimates were very similar to observed values, with the highest predicted values in SRS, and lower values (<5,000 mL m−2) in WCA-2 and WCA-3. Differences between observed and diatom-inferred values were generally no more than one standard deviation of the observed variability in hydroperiod and periphyton biovolume (Fig. 5c and f, respectively).

Fig. 5
figure 5

Observed, diatom-inferred, and difference (observed minus diatom-inferred values) maps of hydroperiod (a, b, c) and periphyton biovolume (d, e, f), respectively. Black values in difference maps (c and f) indicate values within one standard deviation of observed values. Standard deviations of hydroperiod and periphyton biovolume are 61 days and 3200 mL m−2, respectively. Patterning delineates wetland subset boundaries. Color version available online

Full size image

Discussion

Hydroperiod, periphyton biovolume, and periphyton TP content were the habitat characteristics most strongly associated with diatom assemblage structure in the Everglades. The underlying geology and geomorphology influence the natural hydropatterns of the Everglades (McVoy et al. 2011), while an extensive system of canals, levees and water control structures control the present flow of water (Sklar et al. 2002). Alterations to the hydrology of the Everglades influence periphyton biovolume and nutrient content, especially when canal inputs increase delivery of minerals or P and encourage the development or disintegration of periphyton communities (Iwaniec et al. 2006; Hagerthey et al. 2011; McCormick et al. 2011). Periphyton biovolume has been used as a metric of productivity (e.g., Ewe et al. 2006; Iwaniec et al. 2006), but has not been analyzed as a habitat characteristic for diatom assemblages. Because mats of high and low abundance appear to support distinct diatom assemblages, assemblage composition is useful for inferring the abundance of periphyton in given wetland locations in the past (before drainage and management) to inform restoration target-setting. A known relationship between assemblage composition and periphyton biovolume would also aid restoration assessment because obtaining measurements of periphyton biovolume in the field is simple compared to the additional processing required to obtain measurements of chlorophyll a or nutrients in the laboratory. Because certain diatom species (e.g., Eunotia camelus, Gomphonema gracile, and Stenopterobia curvula) are associated with flocculent detrital material but absent or infrequent at sites with periphyton (zero minimum biovolume optimum and tolerance), their relative abundances can indicate the absence of periphyton. Thus, diatoms are comprehensive indicators of periphyton abundance that can provide inferences about a range of wetland environments that may or may not support periphyton.

While the relationship between phosphorus concentrations and diatom assemblage structure is well-documented (McCormick et al. 1996, 1998; Pan et al. 2000; Gaiser et al. 2006; Cooper et al. 2008; La Hée and Gaiser 2012), this study shows the important effects of hydroperiod and periphyton biovolume on diatom composition. We have yet to uncover the exact mechanisms driving diatom assemblage composition, but feedbacks among hydroperiod, periphyton biovolume, and composition could shape wetland structure and function. Since short hydroperiod areas tend to have high periphyton abundance (Gottlieb et al. 2006), the periphyton mats in these areas could favor desiccation-resistant species capable of surviving annual drying. In turn, desiccation-resistant diatoms and other algal species produce biomass, including anti-desiccative mucilage (Hoagland et al. 1993), that contribute to high biovolume periphyton mats, while also playing a role in controlling nutrient and mineral concentrations in the surrounding water column (Gottlieb et al. 2005; Thomas et al. 2006; Hagerthey et al. 2011). Further investigations are needed to understand how changes in hydrology affect these feedbacks.

The responses of individual diatom species, especially species with narrow tolerance ranges, had a key role in determining predictive relationships of diatom assemblages with hydroperiod and periphyton biovolume. For example, Fragilaria synegrotesca preferred long hydroperiod habitats; this agreed with previous findings by Gottlieb et al. (2005) and Gaiser et al. (2011). The diatom species with the longest hydroperiod optimum was Fragilaria delicatissima, a planktonic species (Patrick and Reimer 1966). Pinnularia microstauron was associated with shorter hydroperiod habitats; the type specimen of this species was found in soil on the roots of plants from Rio de Janeiro, Brazil, indicating its ability to survive in drier conditions (Patrick and Reimer 1966). The holotype of Caponea caribbea, which had the shortest hydroperiod optimum in our analyses, was collected by Podzorski (1985) from algal mats on the surface of a Jamaican peat swamp that had experienced a fire some months prior. Podzorski’s (1985) findings suggest that C. caribbea is adapted to surviving periodic desiccation and fire, which are frequent in short-hydroperiod wetlands; this species may even indicate time since fire. Species associated with habitats with very little periphyton, such as Eunotia camelus and Stenopterobia curvula, are acidophilic diatoms not usually found in calcium-rich habitats like calcareous periphyton mats (Patrick and Reimer 1966; Krammer and Lange-Bertalot 1988; Furey 2010). Gomphonema gracile and G. coronatum have been reported to prefer high P habitats with pH < 7 (Tobias and Gaiser 2006; Slate and Stevenson 2007).

Some of the more common diatom species in the flora of Everglades calcareous periphyton had broad tolerance ranges that were not useful for inferring habitat characteristics. Nitzschia serpentiraphe, a very common species in slough periphyton mats with a moderate (Slate and Stevenson 2007) to low TP optimum (Gaiser et al. 2006), reached its highest abundance in the thick, calcareous mats from SRS. Calcareous periphyton mats disintegrate due to a shift in community structure from a calcite-precipitating filamentous flora (species of Utricularia and cyanobacteria that provide substrates for diatom growth) to dominance by green algae and eutrophic diatoms following P enrichment (McCormick and O’Dell 1996; McCormick et al. 2001; Gaiser et al. 2005b, 2006, 2011). This phenomenon supports the association of N. serpentiraphe with high periphyton biovolume and correspondingly low to moderate TP habitats. However, N. serpentiraphe also had a wide tolerance range, which is expected because species in this genus are known to tolerate eutrophic conditions (Van Dam et al. 1994). The species with the highest periphyton biovolume optima (including N. amphibia var. frauenfeldii, N. serpentiraphe, Brachysira vitrea, and B. neoexilis) had an average tolerance range of greater than 8000 mL m−2. The wide tolerance ranges of these species restrict optima from reaching higher values, even though periphyton abundance can exceed 10,000 mL m−2. Inferences about periphyton abundance and enrichment based on diatoms with wide tolerance ranges should be made with caution because other factors may allow some species to thrive in both enriched and unenriched habitats.

We found that although the wetland subsets in our study differed in species’ relative abundances, similar assemblage responses to habitat characteristics allowed development of a landscape-scale model encompassing gradients broader than those present in individual subsets. None of the NMS ordinations captured a response to conductivity, likely because of the shorter conductivity gradient encompassed by our focus on alkaline freshwater habitats (where calcareous periphyton thrives) relative to gradients of other habitat characteristics. The lack of response by WCA-3 assemblages to metrics of inundation (days since dry and hydroperiod) may reflect how the impoundment of water for flood control in the WCAs limits the natural variability of inundation (Light and Dineen 1994; Romanowicz and Richardson 2008). In contrast, historic P enrichment in WCA-2 affected nutrient removal (Craft and Richardson 1993), peat accretion (Reddy et al. 1993), and vegetation (Urban et al. 1993). In our study, species assemblages in WCA-2 were not associated with any of the variables measured except TP. The lack of response to the other habitat characteristics could be attributed to the low sample size of the subset (n = 10). Despite these differences, hydroperiod and periphyton biovolume weighted averaging models made separately for each subset had lower boot r 2 values (especially models for WCA-2) or only slightly improved boot r 2 values (the hydroperiod models for SRS and WCA-3) (Refer to Table 5), probably because of the decrease in sample size (Birks et al. 1990; Reavie and Juggins 2011). The lack of improvement in correlation values of subset-specific models validated combining wetland subsets for landscape-scale assessment.

The weighted averaging models for hydroperiod and periphyton biovolume showed the utility of diatom assemblages for wetland assessment. Our Everglades model was not as strong as the hydroperiod model by Gaiser et al. (1998) for diatoms from surface sediments of intermittent ponds on the Atlantic Coastal Plain (r 2 = 0.81), because the Everglades model underestimates long hydroperiod conditions and overestimates short hydroperiod conditions. The limitations of the Everglades model could be a result of higher spatial and temporal variability within a wetland compared to ponds that are similar in geology, vegetation, and water source but range broadly in hydroperiod, permitting a more extreme gradient for developing inferences. Because we also recognized that unimodal-based calibration methods are prone to the ‘edge effect’ (ter Braak and Juggins 1993; Birks 1998), which biases inferred values towards the mean of the observed values, we evaluated but did not employ weighted averaging models using partial least squares regression (ter Braak and Juggins 1993) to conserve parsimony and because improvements were not statistically significant.

Visualizing the weighted averaging models showed that for both hydroperiod and periphyton biovolume, most of the differences between observed and diatom-inferred values were within the variability (one standard deviation) of the observed values. Because wetlands are spatially complex, we emphasize the utility of maps to summarize biotic responses to the environment by visualization of biological patterns across a landscape in a way that integrates spatial and environmental heterogeneity. Maps can complement tools already developed for restoration assessment (Gaiser 2009), to visualize compliance or deviation of current conditions from reference or restoration targets.

We emphasize the utility of analyzing diatom assemblage composition as a way to develop predictive models about the environment that coarser metrics may not always capture, especially in dynamic systems such as wetlands. Unlike one-time direct measurement of physical or chemical conditions such as periphyton biomass or water quality, assemblage composition is less prone to sampling variability because it integrates environmental conditions over larger spatial and temporal scales (Stevenson 1998). The relationship between diatom assemblages and hydrology is especially important in the Everglades because hydrologic manipulation is the basis of current restoration plans. As advances in diatom taxonomy continue to develop, it is imperative to make identifications to the species level in order to reconcile any current taxonomic discrepancies with future datasets (Julius and Theriot 2010). This effort would facilitate the combination of datasets for analyses at larger spatial and temporal scales. While assessment based on the presence or absence of a few indicator species can be less time consuming and costly, this method is prone to non-detection resulting from variable species detectability or inadequate sampling (MacKenzie 2005). Gottlieb (2003) found that long and short hydroperiod assemblages differed in species relative abundances, rather than in the presence or absence of species. Likewise, we found that wetland subsets had differing species relative abundances, not species identities, which permitted us to develop inference models at the landscape scale. The laborious process involved in species identification and enumeration is worthwhile, considering the wealth of information it provides about how the species within assemblages respond, as a whole, to the environment and should be incorporated in assessments of other wetland systems.

Sound ecologically-based assessment of change in wetlands requires an understanding of how assemblages respond to ecosystem parameters to make interpretations about past conditions, develop targets for the future, and aid assessment. This study found diatom assemblage composition has predictable relationships with two very important habitat characteristics of wetland ecosystem function: hydroperiod and periphyton biovolume. Wetland management programs should use this information to infer landscape-scale biological responses to hydrologic change and to inform restoration targets, especially in the Everglades where hydroperiod regimes no longer follow natural spatial and temporal patterns (Romanowicz and Richardson 2008). This study quantitatively described the hydroperiod and periphyton biovolume optima and tolerance ranges of Everglades diatom taxa. This autecological information should be used to futher investigate the poorly described flora of tropical karstic wetlands (La Hée and Gaiser 2012). Finally, this study found landscape-scale models are possible in the Everglades because diatom assemblage patterns were consistent across wetland subsets. Spatially-explicit visualizations of these models are useful for assessing the performance of diatom-based inferences over a large area. Landscape-scale diatom-based models from the Everglades should be used as an example for other large wetlands around the world with high hydrologic connectivity, such as boreal wetlands (Spence et al. 2011), the Pantanal (Alho et al. 1988), and the Okavango Delta (McCarthy et al. 2000). Although most tropical wetlands have received inadequate attention to date, advancements made by monitoring, restoration, and assessment efforts for the subtropical Everglades can guide newly developing wetland protection programs.