Biotic and abiotic determinants of finescale dace distribution at the southern edge of their range

The factors that set range limits for animal populations can inform management plans aimed at maintaining regional biodiversity. We examine abiotic and biotic drivers of the distribution of finescale dace (Chrosomus neogaeus) in two Great Plains basins to identify limiting factors for a threatened freshwater fish population at the edge of their range.


| INTRODUC TI ON
Understanding the factors that create species' range limits is a central goal of biogeography and can inform fisheries and wildlife management. Constraints on the distribution of animal populations have been conceptualized as a series of abiotic and biotic filters operating over multiple spatial and temporal scales (Poff, 1997;Smith & Powell, 1971;Tonn, 1990). Long-term environmental change (e.g. Pleistocene glaciation) and modern disturbances such as biological invasions and habitat destruction can interact to increase extirpation risk, particularly for rare and threatened species (Brook et al., 2008). The global proliferation of invasive species threatens native biodiversity and emphasizes the importance of identifying where physical factors or biotic interactions exert more influence on populations (Clavero & García-Berthou, 2005;Louthan et al., 2015).
A key challenge to such efforts are sparse occurrence records for many species of conservation concern (Elith et al., 2006). Studies that leverage new and existing survey data in modelling efforts for these populations will be better positioned to inform their conservation and management.
Many studies employ physical climate data in species distribution modelling efforts, while applications assessing where abiotic or biotic factors will dominate are less common (Franklin, 2013;Louthan et al., 2015). The latter approach may be particularly useful in the study of climate relicts, which constitute fauna isolated due to environmental change in locally suitable areas, often near the edge of longitudinal or elevational thresholds for range-wide distributions (Hampe & Jump, 2011;Woolbright et al., 2014). Key threats to climate relict populations include limitations on available habitat and loss of genetic diversity (Hampe et al., 2005;Rabinowitz, 1981).
Studies of relicts are often challenged by their geographic rarity and lack of previous study (McDonald, 2004;Wisz et al., 2013), yet could yield insight into species-environment relationships which differ from those in a core distribution (Gross & Price, 2000;Tarkan, 2006).
Relict fish populations in the Great Plains, USA, are members of an understudied biotic community in the prairie biome (Fausch & Bestgen, 1997). Finescale dace (Chrosomus neogaeus) (Cope, 1869) is a North American freshwater fish species that occurs in the Great Plains in isolated populations at the southern edge of their range in Wyoming, South Dakota and Nebraska (Lee et al., 1980; Scott F I G U R E 1 Distribution of finescale dace (Chrosomus neogaeus) across North America and within the study area at the southern edge of their range in the (a) Belle Fourche River (n present = 16; n absent = 115) and (b) Niobrara River (n present = 13; n absent = 14) basins of Nebraska (NE), South Dakota (SD) and Wyoming (WY), USA. Locations where finescale dace were detected (black) and were not detected (white) were determined from field surveys by multiple state agencies and the authors from 2008 to 2019. The map projection is North America Albers Equal Area & Crossman, 1973) ( Figure 1). Their regional habitat associations are linked to cool-water environments (i.e. headwater streams and groundwater seeps), which may enable persistence under localized conditions more similar to cooler, wetter palaeo-climatic conditions that prevailed several thousand years ago (McPhail, 1963;Stasiak & Cunningham, 2006). The low prevalence and spatially disjunct pattern of finescale dace occurrence in the Great Plains (Hoagstrom & Berry, 2006) account for their designation as a species of conservation concern in this region (Schneider et al., 2011;South Dakota Wildlife Action Plan. Wildlife Division Report 2014-03, 2014Wyoming State Wildlife Action Plan, 2017).
The fish community associations of finescale dace are not wellunderstood, but studies in north temperate lakes indicate that native cyprinids in the genus Chrosomus are highly sensitive to introductions of non-native predators, including bass and sunfish in the family Centrarchidae and northern pike (Esox lucius) (Findlay et al., 2000;MacRae & Jackson, 2001;Tonn & Magnuson, 1982;Whittier et al., 1997). Scaling up from lakes to the landscape, Jackson and Mandrak (2002) predicted the loss of a substantial number (>25,000) of local cyprinid populations in Ontario by 2,100 due to climate-mediated northward range expansion of smallmouth bass (Micropterus dolomieu). Threats to cyprinid biodiversity from invasive predators are likely exacerbated at edge populations , highlighting the need for distributional data that can inform analyses of limiting factors.
Our objective in this study was to identify drivers of the contemporary distribution of relict finescale dace in the headwaters of two Great Plains river basins. We assessed potential abiotic and biotic pathways affecting finescale dace occurrence by fitting classification models with (a) abiotic-only and (b) combined abiotic and biotic variables. Given our location at the southern edge of their range, we hypothesized abiotic factors (e.g. stream size and temperature, channel slope) to exert a strong filtering effect on finescale dace occurrence within both river basins (Quist et al., 2005). For biotic factors, we expected that finescale dace occurrence would be positively correlated with the presence of other small-bodied native fishes, as higher native minnow richness could be indicative of habitat quality or a lack of influence from non-native species. Conversely, we expected a negative relationship between finescale dace occurrence and the presence of non-native, predatory fish species, which forage in nearshore, littoral environments. By incorporating ecologically relevant variables into our basin-specific analysis, we aim to better understand the environmental context under which limiting factors prevail for finescale dace in the Great Plains. These insights could inform broader actions to safeguard imperilled aquatic species and conserve regional biodiversity.

| Study area
Our study occurred within the Belle Fourche River and Niobrara River basins of Wyoming, South Dakota and Nebraska (Figure 1).  (Bradshaw, 1996). The Niobrara River is largely intermittent in Wyoming and becomes perennial near its confluence with Van Tassell Creek near the Wyoming-Nebraska state line. Across both basins, groundwater is an important source of surface water for fish-bearing streams and small lakes (Naus et al., 2001;Szilagyi et al., 2003).

| Fish surveys
Given a paucity of fish survey data in our region of interest, we compiled a dataset from years 2008 to 2019 from multiple state agencies (Table 1) fish community and habitats under consideration for restoration activities by managers. Given the rarity of finescale dace in the study area and lack of data, we sampled non-random lentic and lotic sites (n = 29) in historical localities to refine distributions and evaluate limiting factors (Egly & Larson, 2018). We also randomly sampled lotic sites (n = 33) within drainages with known finescale dace occurrence, translocation history or as recommended by managers for restoration potential. We jointly used these surveys and synthesized data from managers to assemble a dataset representative of contemporary basin-wide fish occurrence in the study area. While we used this 11-year snapshot of fish surveys as a representation of contemporary fish occurrence, we acknowledge that aggregating surveys over time limit our ability to examine potentially important shifts in species distributions during the study period.
We used fish community data comprised of 36 species (25 native and 11 non-native) across nine families (Table 2) to calculate native minnow richness and classify the occurrence of littoral predatory fish at study sites. Native minnow richness was calculated as the sum of unique occurrences of native, small-bodied fishes. We then subtracted finescale dace from these totals at sites where they occurred. We also calculated a binary presence/absence variable for multiple non-native, predatory fish species ( Table 2) that are known to forage in littoral, nearshore habitats and have negative effects on native cyprinid communities as introduced or invasive species (He & Kitchell, 1990;Jackson & Mandrak, 2002).

| Environmental data
We retrieved data from multiple sources to describe environmen-

| Statistical analyses
We used Random Forests classification models to examine potential relationships among abiotic and biotic factors that influence the occurrence of finescale dace in our study area. Random Forests are a type of classification and regression tree (CART) model that are generated by bootstrapping data to create a large number of decision trees, which converge on a prediction and rank covariates by their importance in explaining variation in the response (Breiman, 2001;Breiman et al., 1984). This modelling approach is well-suited for ecological studies as it assumes no underlying distribution of the data, is invariant to predictor variable transformations, can fit nonlinear relationships and is relatively insensitive to autocorrelation (Cutler et al., 2007). Finescale dace occurrence (present (1) or absent (0)) was the binary response variable, and unstandardized abiotic and biotic variables were used as predictors (Table 1). To maintain model parsimony, we used the model improvement ratio (MIR) method, which minimizes the number of retained predictors and model meansquared error while maximizing the percentage of variation explained (Murphy et al., 2010). We fit models separately for each river basin to examine potential context-specific drivers of fish occurrence and provide interpretable information for regional fisheries managers.
For each basin, we fit two groups of models. In the initial model, (a) we used only abiotic covariates as predictors of finescale dace occurrence. We then fit (b) a subsequent model with added biotic variables to examine their effects on model performance (Vezza et al., 2015). We applied the MIR procedure to each model for covariate selection. We used packages "randomForest" and "rfUtilities" in Program R for model fitting, selection and performance assessment (Evans & Cushman, 2009;R Core Team, 2019;Wiener & Andy, 2002). For datasets with excessive zeros (≥30% of the response variable), we used zero-inflated models, which combine independent ensembles of models fit with random subsets of the majority class (Evans & Cushman, 2009).
To assess model performance, we calculated the percentage of correctly classified instances (CCI) with confusion matrices (Buckland & Elston, 1993), evaluated predictive performance with area under each curve (AUC) measures from receiver operator characteristic plots (Manel et al., 2001) and evaluated cross-classification error using Cohen's Kappa statistic (K) (Cohen, 1968 estimates for out-of-bag (OOB) data (observations withheld from the training dataset), which function as cross-validated accuracy estimates (Vezza et al., 2015). Model significance (p; α = .05) was computed with a randomization test with 999 permutations (Murphy et al., 2010).
Intercorrelation among predictor variables was calculated using generalized variance inflation factor (GVIF), and highly correlated variables (GVIF > 3) were not included in the same model .
We used partial dependence plots of the marginal effect of covariates on the predicted probability of finescale dace occurrence to visualize species-environment relationships (Cutler et al., 2007).

| Random Forests classification models
Covariates supported in basin-specific classification models ex-   (Figure 3).

TA B L E 4
Random Forest classification models for finescale dace occurrence fit with fish surveys from the Belle Fourche River (n = 131) and Niobrara River (n = 27) basins of Wyoming, Nebraska and South Dakota, USA, from 2008 to 2019 and abiotic and biotic covariates (  By contrast, low-gradient stream channels and seasonal wetlands characterize the more homogenous high prairie environment of the F I G U R E 2 Partial dependence plots from Random Forests models depicting finescale dace occurrence as a response to covariates from the best supported model fit with Belle Fourche River basin surveys. Abiotic factors were the most important variables in the model, as finescale dace occurrence was negatively associated with streamflow (m 3 /s) and exhibited a nonlinear relationship to other covariates. Variable importance (range = 1.00-0.17) is listed for each covariate in descending order F I G U R E 3 Partial dependence plots from Random Forests models depicting finescale dace occurrence as a response to covariates from the best supported model from the Niobrara River basin. Finescale dace were positively associated with a biotic covariate, native fish minnow richness, exhibited nonlinear relationships to mean August water temperature (°C) and open water and wetlands (%), and were not predicted to occur at sites with littoral predators. Variable importance (range = 1.00-0.08) is listed for each covariate in descending order Niobrara River basin. In addition, the Niobrara River is experiencing the arrival of a novel non-native predator, northern pike. In this environmental context, biotic factors were important drivers of occurrence. This result adds to a body of work highlighting the utility of biotic interactions in models used for fisheries and wildlife conservation planning (Chamberlain et al., 2014;MacDougall et al., 2018).

| D ISCUSS I ON
Consistency in the relationship between August water temperature and finescale dace occurrence across both basins suggests summer thermal habitat as a potential limiting factor at the regional level.

| Drivers of finescale dace occurrence
Abiotic drivers of finescale dace occurrence in the Belle Fourche River basin suggest that geomorphic and climatic filters limit their habitat breadth. The negative relationship between finescale dace occurrence and mean annual streamflow (Figure 2) suggests an orientation towards headwaters for this population. The effect of stream size as a distributional constraint on small-bodied freshwater fishes has been documented in Wyoming (Quist et al., 2004) and other locations in North America (Taylor et al., 2006;Zorn et al., 2002). The that finescale dace inhabit low-gradient streams, bogs and lakes (Rahel, 1984;Stasiak, 1972). These patterns are likely indicative of hierarchal relationships between landscape position and important local habitat characteristics for finescale dace, such as vegetation cover (Stasiak, 1978).
The nonlinear response of finescale dace occurrence to a biotic covariate, native minnow richness, in the Belle Fourche model, may reflect less diverse fish assemblages, and historic management of finescale dace, in this part of the study area. Managed translocations, beginning in the late 1970s, established conservation populations in several headwater drainages in the Belle Fourche River basin. We documented present-day persistence at a subset of these localities.
Native fish species richness at finescale dace-occupied sites in these drainages is comparatively lower than sites in the Niobrara River basin (x Belle Fourche = 2, x Niobrara = 6). Consequently, these data may contains cold waters (<19°C) (Wehrly et al., 2003). Water temperature is known to limit the dispersal ability and geographic range of ectothermic species (Recoder et al., 2018;Turlure et al., 2010).
Future warming trends may further restrict finescale dace distribution at lower latitudes. However, groundwater-influenced headwater streams which currently support finescale dace will likely experience slower warming rates, potentially allowing them to serve as climate refugia .

| Fish community patterns in the Niobrara River
In the Niobrara River, finescale dace co-occurred with each spe- These data are indicative of novel biotic interactions that could alter or eliminate entire assemblages of native freshwater fishes.

| Conservation and management implications
We expected abiotic factors to be dominant drivers of finescale dace occurrence in our study system, as relict populations tend to occupy areas in close proximity to abiotically stressful conditions (Woolbright et al., 2014). This was supported by (a) a shared response to August water temperature in both basins, indicating that summer thermal habitat is a regional limiting factor for finescale dace in the Great Plains, and (b) the importance of abiotic drivers in the Belle Fourche River basin (Figure 2, Figure 3). The importance of biotic interactions in the Niobrara River highlights a scenario where biotic interactions may further constrain populations with naturally low prevalence. Collectively, these results suggest that potential mitigation efforts for finescale dace will benefit greatly from basinspecific management plans. Hills National Forest with perennial water and low stream gradients as two factors affecting site suitability for beaver. Areas with suitable geomorphology, vegetation and hydrology could be targeted for active beaver reintroduction (Macdonald et al., 2000) or restoration using beaver dam analogs-structures which simulate beaver dams and promote recovery of degraded streams (Pollock et al., 2014). Ecosystem benefits stemming from beaver-influenced landscapes, including increased habitat complexity, streamflow duration and invertebrate production (Kemp et al., 2012), could benefit finescale dace and the greater biotic community of the Belle Fourche River basin.
Leveraging biological data, such as the distribution and abundance of competing and predatory species, can yield insight into emergent threats to native taxa. We highlight the importance of such an approach with fish survey data from the Niobrara River basin, where biotic interactions substantially improved performance of classification models (Table 4). An apparent reduction in native minnow richness at sites with non-native littoral predators adds to previous studies examining incursions of invasive species into hotspots of native species diversity (Light & Marchetti, 2007;Stohlgren et al., 1999). Invasions of northern pike are particularly alarming, as they exhibit trophic adaptability in prey selection between fish and invertebrates, indicating their potential to alter or eliminate native fish communities as novel apex predators (Beaudoin et al., 1999;Chapman & Mackay, 1984;Sepulveda et al., 2013). Their profound impact on aquatic systems alludes to dim ecological futures for native small-bodied fishes of the Great Plains in the absence of effective predator control or containment. The Niobrara River's fragmented nature due to headwater intermittency provides opportunities for slowing northern pike invasion, but is also a challenge for the long-term fitness of native fishes (Pavlova et al., 2017). Safeguarding remaining native fish diversity in the Niobrara River will likely require measures, which both suppress predators and use hydrologic and structural fish passage barriers to control non-native colonization of remaining native fish refugia (Kruse et al., 2001).
Studies that effectively integrate new and existing datasets can inform future actions aimed at maintaining distinct assemblages of native fishes in an increasingly uncertain future of invasions and environmental change. Our results highlight the importance of basin-specific approaches due to the potential for high context dependence, as we found explanatory factors that were useful for evaluating the relative importance of biotic versus abiotic drivers. In a topographically diverse watershed such as the Belle Fourche abiotic drivers can strongly influence fish distributions, while biotic drivers are likely to be especially important in basins with recent biological invasions. In this context, novel species interactions present a more pressing risk to persistence than abiotic stressors often linked to climate relict distributions.
Conservation plans that incorporate mutual and synergistic threats will improve efforts to safeguard rare species at the edge of range-wide distributions.

ACK N OWLED G EM ENTS
We are deeply grateful to the private landowners of Crook County

CO N FLI C T O F I NTE R E S T
There is no conflict of interest declared in this article.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13227.