4.1. Estimating Drainage-Wide Species Richness
Most efforts to estimate fish species richness emphasize the “stream reach” scale, typically a stream segment of several hundred meters or less, for the purposes of studying community structure, assessing the effects of environmental perturbations, identifying cost-effective sampling strategies, and ensuring data quality. Other studies address fish species richness at regional, continental, or global scales in order to identify large-scale biodiversity patterns and processes that create them [
43,
44]. Although species lists have been compiled for many drainage basins (e.g., [
45]), relatively few studies [
23,
24,
25] have explicitly addressed the problem of estimating fish species richness at the scale of the entire watershed, and no studies have addressed this issue at the scale of multiple contiguous basins, despite its potential importance. The drainage basin or watershed is a highly appropriate unit for conservation because of the physical connectivity of lotic ecosystems combined with the vagility of fish and other aquatic organisms. Accurately and efficiently estimating species richness over multiple, contiguous watersheds is important when designing conservation areas, assessing the large-scale benefits of conservation programs, and identifying potential diversity “hotspots”. From a theoretical standpoint, metacommunity theory emphasizes the need to characterize biological communities at multiple scales to understand how the dispersal of interacting species influences local and regional biodiversity [
46].
Assessing fish species richness at a drainage basin or larger scale necessitates the effective sampling of numerous individual belonging to differently behaving species that are distributed among different habitats. The basic problem in such large-scale surveys is not sampling all individuals within a habitat, hence missing some species by chance, combined with the problem of not sampling all habitats, some of which may support unique species for ecological reasons [
19]. Regarding catchment scale sampling, Smith et al. [
23] identified practical problems that are related to differences in gear efficiency across diverse habitats, the species selectivity of different gears, and the need to identify an appropriate level of sampling effort. They believed that using a variety of gear types helped to address these problems. Multiple gears types were not used in this study because abundant coarse and fine woody debris made it difficult to deploy nets and because electrofishing was the historical method of choice due to its overall effectiveness and convenience. The effects of gear bias and related sampling issues on the accuracy of the species richness estimates presented herein is impossible to quantify; however, they were probably greater for large streams where greater habitat volume made it harder to catch fish and to adequately sample all habitats.
Unlike stream-reach scale surveys that encompass habitat diversity on spatial scales of pools, riffles, runs, and smaller, drainage basin or multiple drainage basin surveys must consider larger scale habitat heterogeneity. The most important source of this heterogeneity observed herein was stream width (or its correlate stream order), which was associated with changes in species composition that mainly were caused by the addition of species with increasing stream size. The influence of stream size is supported by previous research showing that it and related measures (e.g., drainage basin area) create environmental heterogeneity that strongly affects fish richness and composition [
47,
48,
49]. Secondary to stream size, and accounting for less variance, were differences in species composition among drainage basins. I did not attempt to identify reasons for differences among drainages basins but habitat, stream-size related factors, and proximity to source pools in larger watercourses likely played a role, as explained more later.
The strata in the stratified design were stream order (a convenient representation of stream size) and drainage basin to account for the major factors affecting species composition in the study area. The result was a mix of higher and lower order stream sites that are located on the main stems and tributaries of all five SRS drainages, with sites sampled randomly within each stratum to the extent possible. The stratified design also spaced the sample sites relatively evenly over the stream networks on the SRS, thereby resembling the spatially spread (hyperdispersed) sampling strategy that was used by Rosensweig et al. [
19] to efficiently estimate the species richness of butterflies over 110 North American ecoregions. They found that estimates were more accurate when sample sites were spread uniformly than when clustered or selected randomly. The samples in the stratified design were drawn from a data set collected over the shortest time possible to minimize the influence of species changes over time (i.e., species-time relationship) on the results. Research in central European streams showed that the number of species was more strongly influenced by spatial scale than by time, although the latter had a significant effect, especially in larger streams with greater connectivity to other watercourses that facilitated colonization [
50].
The level of effort associated with the stratified design was much less than the total level of effort expended at the Savannah River Site, expressed as the number of sites sampled (32 vs. 70) and the number of individuals collected (12,390 versus 39,087). It was superior to the random selection of 32 sites, which detected fewer species and almost always resulted in the underestimation of species richness. Ideally, the stratified approach would have been evaluated further by generating multiple stratified designs for statistical analysis, but the number of appropriately located sites sampled with sufficient effort was inadequate for this. Despite this limitation, the results suggest that the efficiency of species richness estimation can be maximized by using a sampling design that accounts for key spatial factors that influence species richness, as was originally hypothesized.
I restricted the stratified design to sites that were sampled with multiple (4–7) electrofishing passes because multiple-pass electrofishing usually results in the detection of more species than single pass electrofishing. For example, Meador et al. [
51] reported that as little as 40% of estimated total species richness was collected on the first pass, and Kimmel and Argent [
52] reported that a second electrofishing pass usually added at least one species. Paller [
21], working with multiple-pass data collected from some of the streams in the present study, found that the number of species increased asymptotically with the number of passes. There are two reasons that more passes collect more species. One is a passive sampling effect—more passes collect more individuals, hence are more likely to detect rare species. The other, less well-documented reason, is that some species may be more likely to appear during a second or third electrofishing pass because of their behavior combined with gear and/or observer biases. I observed that some benthic species (e.g., madtoms Noturus spp. and/or darters Etheostoma and Percina spp.) were often missed on the first pass, perhaps because they initially remained on the bottom where they were hard to see. Others have reported that darters and cyprinids [
52] or cyprinids and centrarchids [
51] were most often missed on the first pass.
When restricted to the first two passes, species richness estimates that were based on the stratified design fell slightly short of the benchmark 70 species. Although not analyzed, it is probable that one-pass sampling would produce larger underestimates. The reduced efficacy of two-pass sampling was at least partly due to passive sampling effects—4211 individuals were collected by two-pass sampling as compared with 12,390 by multiple-pass sampling. It may be possible to reduce this effect by increasing the time spent shocking during each pass, increasing the number of electrofishers [
52], or increasing sample site length since all would result in the collection of more individuals.
Reducing the number of electrofishing passes compromised the accurate estimation of SRS-wide species richness. However, first-order stream sites could be eliminated without such compromise, thereby decreasing the total effort. Modifying the stratified design by eliminating six first order stream sites (20% reduction in number of sites) resulted in no change in the number of species collected from the unmodified design (60), a slight reduction in the number of individuals collected (10,411 as compared with 12,390), and no reduction in the accuracy of species number estimates. First-order stream habitats within the study area did not support species absent from higher order sites nor did they usually support large numbers of individuals. Thus, eliminating them did not affect the inclusion of habitats with unique species nor the number of individuals collected—two factors that are largely responsible for the SAR. Similarly, Smith and Jones [
24] recommended allocating most sampling effort to third order streams when sampling for watershed-level species richness in Great Lakes streams. However, eliminating or reducing the sampling effort in first or second order streams is inappropriate when these streams support species that differ from those in higher order streams.
Of the six species richness estimators that were examined in this study, the two that produced the most accurate estimates of species richness were the nonparametric Chao 2 and first-order jackknife. Both displayed the desirable features of approaching the benchmark more quickly than the species accumulation curve and stabilizing near the benchmark without greatly overshooting it. In contrast, the other estimators followed the species accumulation curve closely (thereby exhibiting little predictive power) and/or underestimated or overestimated species richness. The variance around all species richness estimates was high when using the stratified design. Although it could be decreased by sampling more sites, large variances are typical of species richness estimators that extrapolate beyond the limits of the data [
18].
Various estimators have been used by researchers to assess richness at the stream reach scale, including the jackknife, bootstrap, Chao, and Michaelis-Menton [
49,
53,
54,
55]. Glowacki and Penczak [
56] found that no estimator was consistently accurate, but preferred the homogenous model of Chao and Lee [
57], followed by the first-order jackknife at the reach scale. Estimators that were used at the watershed scale include the first-order jackknife for watersheds in the Great Lakes region [
24] and the Chao 2 for a small watershed in Alabama [
25]. In the latter case, the Chao 2 method produced over-estimates at sites with high numbers of uniques and low numbers of duplicates. The Chao and jackknife families of estimators are dependent on the number of species that are found in only one sample (uniques) or two samples (duplicates) [
13,
18]. Since these variables depend on relative abundance patterns, sampling efficiency, and number of samples, it is impossible to predict the success of these estimators using other data sets. Furthermore, all of the estimators compensate only for the incomplete sampling of individuals and not for the incomplete sampling of habitats. Thus, when sampling at the drainage basin scale, they must be paired with a sampling design that adequately represents all habitats that may support unique species.
The SRS data showed that samples from as few as 25 sites (stratified design minus first-order streams), which represented about 85% of the total species, could provide a basis for the accurate estimation of total species richness across five contiguous watersheds occupying 780 km
2. For comparison, 9–25 sites and 17–49 sites were needed to represent 80% and 90%, respectively, of the species in individual watersheds (24–433 km
2) in the Great lakes region [
24], and a mean of 8.4 and 15.5 sites were needed to represent 80% and 90%, respectively, of the species in the Little Choctawhatchee River watershed (416 km
2) in southeastern Alabama [
25]. These results suggest that sampling about 25–30 sites can provide an adequate basis for estimating species number in individual, small watersheds or contiguous, similar small watersheds assuming sufficient reach-scale sampling at each site. Reach scale sampling at the SRS was relatively intensive—4–7 electrofishing passes within a reach averaging 50 channel widths in length—as compared with one electrofishing pass within a reach averaging 30 channel widths in the other studies that are mentioned above. Reach-scale sampling effort will need to vary with conditions but must be sufficient to collect most species present, and reductions in reach-scale effort will likely require the addition of more sites to assess species richness at the watershed-scale. Additional requirements for accurate watershed-scale species richness assessments are an adequate sampling design that represents all habitats, and the analysis of the data with an appropriate species richness estimator. In all cases, a SAC should be constructed, and, if necessary, extrapolated to assess sampling adequacy by determining if a majority of the species have been collected (e.g.,
Figure 4, see [
13]).
4.2. Comparing Species Richness Among Reservations
An important reason for estimating species richness is to identify high diversity areas of possible conservation interest. All of the measures of species richness indicated that the SRS supported relatively high fish diversity for the Sand Hills ecoregion. Species richness and density curves were substantially steeper for the SRS than for FBN and FBR, and species richness estimates for the SRS were higher than for these reservations. Alpha diversity was significantly higher at the SRS than at all other reservations, including state, federal, and private reservations that were established and managed for conservation related purposes. SRS diversity was exceptional at all spatial scales, including stream reach, drainage, and across drainages. This was not simply a result of the large size of the SRS (780 km2), which was only slightly greater than the other reservations (650 and 737 km2, respectively, for FBR and FBN). Greater instream habitat diversity, less disturbed land coverage, and higher proportions of forested land coverage were likely contributing factors. These features are associated with fish assemblages that exhibit high biotic integrity and support many species, including those that are sensitive to disturbance.
The assemblage of fish communities is the result of local abiotic and biotic factors that select or exclude potential colonists based on their adaptive traits and regional factors that affect the ability of colonists to reach local habitats [
58]. Habitat features, such as those described previously, are important locally; the size of regional source pools and the distance from them are important regionally. The immediate source pools for the tributaries under study were the major sixth or seventh order rivers that they were connected to: Savannah River for SRS streams, Apalachicola/Chattahoochee River for FBN streams, and Cape Fear and by Pee Dee Rivers for FBR streams. Diversity differences among these large rivers could create diversity differences among their tributaries; however, the total number of species in the Savannah (149) and Apalachicola/Chattahoochee (153) River systems is almost identical [
45]. The 119 species in the Cape Fear River system [
45] is about 20% lower, which seems unlikely to fully explain why species richness in FBR drainages was 38% lower than in SRS drainages. However, distance from the source pool (connectivity) can also affect colonization rates. Connectivity can increase richness for catostomids, cyprinids, and darters due to immigration from larger streams [
59,
60], and such proximity-related effects can occur up to 20 km from the confluence with mainstem rivers [
61]. The average straight-line distance of the SRS sample sites from the Savannah River was 13.6 km, as compared with 25.9 km and 25.4 km for FBN and FBR, respectively, suggesting that this factor could have contributed to the greater diversity of SRS tributaries.
Estimated species richness for individual SRS drainages ranged from 57–71. This can be compared with other North American drainages, despite differences in sampling methods (especially number of sites and individuals sampled, see [
9]) that weaken the comparison. Species richness was 19 in the Virgin River, a tributary of the Colorado River, (123 collections from 76 sites [
62]), and 23 in the Canadian River basin, Texas (eight collections from eight sites [
63]). Many species at both locations were non-native, and both locations were in the western United States where fish communities are relatively depauperate due to biogeographic factors and habitat instability. Forty species were collected from the Pine River drainage in Michigan (48 samples from 12 sites [
64]) and 45 species were collected from the Broad River in Georgia (49 samples from 45 sites) [
65]. The preceding numbers represent species observed, rather than estimated total species richness. Watershed-level species richness estimates derived from Chao 2 or first-order jackknife estimators were 23–58 in nine Great Lakes drainages [
24] and 43–59 (depending upon season) in the Little Chocktawhatchee River in Alabama [
25]. Estimated species richness (second-order jackknife) at a single site ranged from 24–47 in Illinois streams that are located in the relatively species rich Mississippi River basin [
5]. SRS streams compare favorably to these estimates but are less than the 117 species estimated for the Ivinhema River basin in the Mato Grosso do Sul State of Brazil [
53] and about the same or slightly greater than estimated for headwater streams of the Paraguay and Parana basins in the Pantanal region of Brazil (50–53 species) [
49] or the Meghna river estuary in Bangladesh (53 species, [
66]). Thus, biodiversity in SRS streams appears relatively high—greater than in other streams within the Sand Hills ecoregion, higher than in many North American streams, and within the range observed in high diversity Neotropical and Indomalayan biogeographical realms. Upper Three Runs, the SRS stream that supported the greatest fish species richness (71) is also known for its exceptional diversity of aquatic insects, over 550 species, which is one of the highest values worldwide [
67]. These data suggest that SRS streams, especially Upper Three Runs, represent potential “biodiversity hotspots”, at least for the Sand Hills ecoregion, and show how drainage-wide diversity assessments can identify streams of conservation interest.