SEARCH

SEARCH BY CITATION

Keywords:

  • altered flow regime;
  • Arkansas River shiner;
  • fragmentation;
  • landscape change;
  • Notropis girardi ;
  • species distribution model

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Conservation efforts for threatened or endangered species are challenging because the multi-scale factors that relate to their decline or inhibit their recovery are often unknown. To further exacerbate matters, the perceptions associated with the mechanisms of species decline are often viewed myopically rather than across the entire species range. We used over 80 years of fish presence data collected from the Great Plains and associated ecoregions of the United States, to investigate the relative influence of changing environmental factors on the historic and current truncated distributions of the Arkansas River shiner Notropis girardi. Arkansas River shiner represent a threatened reproductive ecotype considered especially well adapted to the harsh environmental extremes of the Great Plains. Historic (n = 163 records) and current (n = 47 records) species distribution models were constructed using a vector-based approach in MaxEnt by splitting the available data at a time when Arkansas River shiner dramatically declined. Discharge and stream order were significant predictors in both models; however, the shape of the relationship between the predictors and species presence varied between time periods. Drift distance (river fragment length available for ichthyoplankton downstream drift before meeting a barrier) was a more important predictor in the current model and indicated river segments 375–780 km had the highest probability of species presence. Performance for the historic and current models was high (area under the curve; AUC > 0.95); however, forecasting and backcasting to alternative time periods suggested less predictive power. Our results identify fragments that could be considered refuges for endemic plains fish species and we highlight significant environmental factors (e.g., discharge) that could be manipulated to aid recovery.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Conservation efforts for threatened and endangered species are initially very reactive in nature as managers attempt to stabilize populations long enough to develop a coherent, proactive plan of action. Unfortunately, conservation and management planning never gets out of this reactive phase for many species (Wilcove, 1987; Cardillo & Meijaard, 2012; Halpern et al., 2012). Although the specifics vary on a case-by-case basis, there are two general questions that appear to keep conservation actions in this reactive stage: what is the nature of the factor(s) driving the decline or limiting the recovery of a given species and how does a given species respond to changes in availability or quality of resources (e.g., habitat, prey base, interactions with invasive species)? Species distribution models (SDMs) offer a versatile tool capable of addressing some of these concerns across a variety of terrestrial and aquatic landscapes (Elith & Leathwick, 2009). By quantitatively predicting the continuous range of a species and correlating presence with variation in environmental conditions at the landscape scale (Graham & Hijmans, 2006), SDMs can identify factors that potentially may be limiting a population. When applied judiciously, this feature of SDMs is particularly useful for quickly separating causative factors from what can be a long list of potential causes of decline (Loiselle et al., 2003; Rodríguez et al., 2007).

The utility of a tool capable of identifying likely drivers of species declines using existing data, such as SDMs, becomes apparent when compiling a list of the multiple, synergistic stressors facing ecosystems. Anthropogenic impacts such as habitat loss, degradation, and fragmentation (Cushman, 2006; Perkin & Gido, 2011), climate change (Pearson & Dawson, 2003; Van der Putten et al., 2010), changing land-use patterns (Brewer et al., 2007), competing demands for surface and groundwater (Mora et al., 2013), and introductions of nonnative species (Westhoff et al., 2011) are a few of the factors that interact to influence species abundance and distribution. Aquatic ecosystems are particularly susceptible to these multiple stressors as evidenced by the higher extinction rates of freshwater organisms compared to those seen in terrestrial counterparts (Ricciardi & Rasmussen, 1999; Revenga et al., 2000; Jenkins, 2003). The rivers and streams of the Great Plains ecoregion have experienced dramatic changes over the past 100–150 years due to changing land-cover patterns, land-use practices, and climatic shifts (Matthews, 1988; Dodds et al., 2004; Hoagstrom et al., 2011; Perkin & Gido, 2011). Under natural conditions, these aquatic systems were characterized by extremes in flows and other biotic conditions, yet supported a diverse endemic fish fauna adapted to the unique challenges of this environment (Matthews, 1988). However, anthropogenic activities have resulted in high levels of fragmentation, loss of channel complexity, reductions in stream discharge including high-flow events, and elevated temperatures resulting in new conditions, different from the prevailing extremes that formerly characterized Prairie rivers and streams (Matthews, 1988; Hall et al., 1999; Dodds et al., 2004; Hoagstrom et al., 2011; Perkin & Gido, 2011).

One group of species that has been notably impacted by the changing environmental conditions is the pelagic broadcast-spawning cyprinids reproductive guild. This reproductive ecotype represents approximately 20 species of small-bodied (<5–6 cm total length) minnows that release semi-buoyant eggs that potentially require substantial lengths of free-flowing river to successfully complete development (Williams & Bonner, 2006; Hoagstrom et al., 2011; Perkin & Gido, 2011). Thirteen of these species are considered of conservation concern (Warren et al., 2000; Jelks et al., 2008), and little information is available on the status of the other remaining species. The rapid decline of this reproductive ecotype has been attributed to a range of factors including fragmentation (Hoagstrom et al., 2011; Perkin & Gido, 2011), altered flow regimes (Hughes, 2005), and invasive species (Felley & Cothran, 1981; Pigg et al., 1999; Bonner & Wilde, 2002; Hoagstrom et al., 2011). This study examined how environmental changes across the landscape of the Great Plains and associated ecoregions impacted the persistence of members of the pelagic broadcast spawning guild. We used one of the better-studied species, Arkansas River shiner Notropis girardi, as a model and assessed how landscape changes altered the distribution of these fishes. Our general approach was to construct SDMs for two time periods relevant to the decline of Arkansas River shiner. We combined ‘ultimate’ distribution structuring variables (e.g., geology, climate) alongside functionally relevant covariates and movement-related descriptors (e.g., discharge, unfragmented river length). We used a modeling approach, MaxEnt, especially well suited for presence-only data collected using multiple sampling strategies, that was available for Arkansas River shiner. We evaluated model predictive accuracy using multiple techniques to assess model transferability across time periods.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study area

The Arkansas River shiner was originally located in the Arkansas River catchment of New Mexico, Kansas, Texas, Oklahoma, and Arkansas, USA (Fig. 1). The catchment covers several ecoregions from the Southwestern Tablelands of New Mexico and Colorado to the eastern extent of the Arkansas Valley of Arkansas. The western edge consists of a semi-arid region dominated by short grasses and rangelands with limited precipitation (approximately 40 cm annually). The majority of the region is located in the Central Great Plains, a region of significant climatic transition, characterized by highly variable precipitation (55–96 cm of rainfall annually; Woods et al., 2005). Great Plains streams were historically characterized by wide, braided channels with sand-dominated or clay beds (Matthews, 1988) and periods of intense flood or drought (Dodds et al., 2004). Today, increased groundwater pumping has resulted in many smaller tributaries being dry for a large portion of the year and main river channels often restricted to a simple, narrow thalweg (Woods et al., 2005). Much of the region has been impacted by impoundments resulting in high levels of fragmentation and altered flow regimes (Perkin & Gido, 2012). Dominant vegetation is mixed grasses and cropland that transition to an oak-savanna mix in the eastern portions of the region. The Arkansas Valley is a transition zone between more dissected regions and is much more humid than the ecoregions to the west (109–145 cm of precipitation annually).

image

Figure 1. Map of the Arkansas River catchment central USA, including main river channels. Study area enclosed by red line.

Download figure to PowerPoint

Data sources

As for many species, our knowledge of the historic distribution of Arkansas River shiner is generally limited to presence-only observations. Arkansas River shiner location records were collected from a number of sources: published literature, university theses, gray literature technical reports, museum and university collection records, museum-specimen databases and federal and state fish-community surveys. For each record, the specimen's capture location and capture date were recorded. Capture locations, provided as written descriptions, were georeferenced by locating the site on a map and converting to geographic coordinates. Records that lacked descriptions or could not be accurately located were removed from the analyses. Each location was quality assured to make certain the point was associated with the correct river, not a nearby tributary, and locations were corrected as necessary. Each location was then assigned to the geographically closest river segment within the network. Records that lacked dates or were associated with introduced populations (e.g., Pecos and Red river catchments) were omitted. Three records from the Verdigris River and two records from the lower Arkansas River fell outside the study area and were also removed from the analyses.

Species distribution models (SDMs) should be constructed around ‘functionally relevant predictors’ (Elith & Leathwick, 2009); therefore, we selected environmental covariates thought to be pertinent to the ecology of Arkansas River shiner. Environmental data were downloaded from online sources (Table 1). Distal predictors such as geology, elevation, and slope (Hynes, 1975) were used alongside those that have a direct impact on the species physiology (e.g., temperature; Fry, 1971) and factors that have been linked to the decline of Arkansas River shiner and other fish species (e.g., discharge and land-use; Cross et al., 1983; Gido et al., 2010). Stream order, maximum elevation, slope, and discharge were accessed from the NHDPlus river segment attributes (McKay et al., 2012). Discharge for each segment was derived using a step-based approach; mean annual runoff grids were calculated and excess evapotranspiration was evaluated. Flows were then adjusted, first based on reference gages, and subsequently on gages in the segment's vicinity (McKay et al., 2012). Climate data in the form of seasonal trends in precipitation and temperature, rather than daily values, were used to model the species' distribution as these better characterize the conditions relevant to the physiological constraints on the organism (Nix, 1986). These bioclimatic predictors were developed by the US Geological Survey (USGS) based on the 4-km resolution Parameter elevation Regression on Independent Slopes Model (PRISM) climate-mapping system (O'Donnell & Ignizio, 2012). River segments may cross the boundary of multiple raster cells so we calculated the mean value for the years 1990–2007 (current model) for each of the 19 bioclimatic predictors based on a weighted (length in each cell) average.

Table 1. Source and description of environmental covariates used in Maxent modeling
Environmental variableDescriptionSource
  1. a

    Current model.

  2. b

    Historic model.

Climate19 Bioclim predictors representing seasonal trends in temperature and precipitation. Resolution 4 kmO'Donnell & Ignizio (2012)
Land-useNational land cover database. Resolution 30 ma. Enhanced Historical Land-Use and Land-Cover DatabPrice et al. (2006); Fry et al. (2011)
GeologyDigitized version of 1 : 100 000–1 : 1 000 000 scale state geologic mapsStoeser et al. (2005)
Stream orderModified Strahler stream orderUSEPA, USGS (2012)
Maximum elevationMaximum elevation of each river section (cm)USEPA, USGS (2012)
SlopeSlope (m m−1) of the river sectionUSEPA, USGS (2012)
DischargeDischarge (cubic feet per second) at downstream end of river section calculated using the Extended Unit Runoff MethodUSEPA, USGS (2012)
DriftRiver length (km) available for downstream drift before meeting a barrierDam locations from US Army Corps of Engineers (2010)

Because Arkansas River shiner is thought to exhibit considerable spatial variation in resource use during its life history, a movement-related descriptor, ‘drift’, was also included. The spatial distribution of river barriers through the river network was used to calculate the variable ‘drift’ to examine the effect of river fragmentation on downstream movement of ichthyoplankton. The geographic locations of dams within the study area were downloaded from the ‘National Inventory of Dams’ (http://geo.usace.army.mil/pgis/f?p=397:1:0). The total number of dams was reduced by removing records that lacked complete information or could not be accurately located within the river network, for example, dams with no construction date or those dams more than 0.01 decimal degrees (distance measured in ArcGIS) from a river segment. We also removed dams classified as ‘off-stream’ as these would not affect ichtyoplankton drift. The remaining dams were assigned to a network segment. When multiple dams were present on the same river segment, the upstream dam was used, leaving a final dataset of 1096 barriers. The Network Analysis function within ArcMap 9.3 (ESRI, Redlands, CA, USA) was used to calculate the distance from the midpoint of each ‘river segment’ that ichtyoplankton could travel downstream through the river network until encountering a barrier or reaching the most downstream segment in the study area.

Two categorical variables were used in the analyses: land-use and geology. Digital representations of state geologic maps consisting of 213 lithology categories were downloaded (Stoeser et al., 2005). Geology was assessed as the category comprising the greatest length of channel within each segment. Current land-use data were downloaded from the 2006 version of the National Land Cover Database (Fry et al., 2011), whereas historic conditions were represented by data collected by the USGS in the 1970s and 1980s (Enhanced Historical Land-Use and Land-Cover Data: Price et al., 2006). The historic land-use categories were based on the Anderson Level II classification system (Anderson et al., 1976), whereas the NCLD used a modified version of the same system. To provide consistency between the two models, the land-use values for the two datasets were truncated into 10 broad categories prior to analyses (Online Supporting Information Table S1). Land-use was assessed as the dominant category in a 500-m buffer either side of each river segment as local land-use can have a greater influence on fish assemblage structure than catchment conditions (Stanfield & Kilgour, in press).

The historic model used the same geology, stream order, elevation, and slope variables as the current model; however, data were further processed to represent historic conditions in the study area prior to the decline of the Arkansas River shiner. The mean value for each of the 19 bioclimatic predictors was calculated from 1895 to 1989. Discharge values were again accessed from the NHDPlus. Discharge, calculated from mean annual runoff grids and adjusted based on reference gages with minimal impact by human activities, was used to ‘best represent natural flow conditions’ (McKay et al., 2012). For the historic model, we wanted to evaluate the role of the unimpacted drift potential on the historic distribution of Arkansas River shiner. Therefore, anthropogenic barriers were removed allowing the ichtyoplankton to drift uninterrupted through the river network.

Species distribution models

Current and historic models were delineated by examining the decadal variation in the number of individual river segments where Arkansas River shiner was recorded and the cumulative distribution of new river segments where the species was located (Fig. 2). The breakpoint was set at the time point when the number of river segments occupied started to decrease and the gradient of cumulative distribution leveled off. Based on this breakpoint, SDMs were constructed using species location records covering two time periods pre-1990 and 1990-2010, hereafter referred to as historic and current models, respectively.

image

Figure 2. The number of unique river segments where Arkansas River shiner presence was recorded (solid line) and the cumulative number of new river segments where Arkansas River shiner presence was recorded (dashed line).

Download figure to PowerPoint

We used a vector-based river network as the foundation for our SDMs. A 1 : 100 000 scale network for the Arkansas, Cimarron and Canadian river catchments, was downloaded from the NHDPlus website (USEPA, USGS, 2012). Arkansas River shiner inhabits main channels and larger tributaries of the Arkansas River catchment (Moore, 1944); thus, to reduce processing time, we removed first order stream segments (modified Strahler; Strahler, 1957) and those classified as ‘isolated networks’ (McKay et al., 2012). The final river network consisted of 53 617 individual river segments.

Distributions were modeled using MaxEnt (MaxEnt 3.3.3k; Phillips et al., 2006; Phillips & Dudík, 2008). MaxEnt was chosen because it is well suited to accommodate presence-only data, model performance is considered superior relative to other statistical methods (Elith et al., 2006; Peterson et al., 2007), and model performance is less sensitive to changes in sample size (Hernandez et al., 2006; Wisz et al., 2008). MaxEnt is a machine-learning tool that predicts a species' distribution by minimizing the relative entropy between presence and background probability densities in covariate space (Phillips et al., 2006; Elith et al., 2011). Rather than the traditional method where GIS derived raster layers form the environmental background to the model, environmental covariate values for each river segment were entered in a spreadsheet form using the samples-with-data approach (see Elith et al., 2011). This has great applicability for modeling aquatic species (e.g., Liang et al., in press; Dyer et al., 2013), due to the linear nature of river networks. Specifically, several individual segments with different covariate values may be present within a single raster pixel and therefore an average or single value would have to be assigned to that point. The default MaxEnt settings were applied, except for the maximum number of background points which was set to correspond with the number of individual river sections in the river network (53 617).

The relative contribution of the environmental variables was assessed within MaxEnt via percent contribution and permutation importance statistics. Percent contribution reports the relative contribution of each variable to model fit, whereas permutation importance is the loss in model predictive power with the removal of that variable. MaxEnt can be used to produce two sets of response curves to examine the shape of the relationship between an environmental covariate and the species' probability of presence. ‘Marginal’ response curves are produced where the values of the other variables are set to their average value over the presence locations whereas ‘single variable’ response curves are created based on a model of that variable with all other covariates discounted (Phillips, 2005). When highly correlated environmental variables exist in the model, examination of ‘marginal’ response curve can be misleading (Phillips, 2005). Pearson's product-moment correlations showed colinearity (r > 0.65) between a number of the continuous variables; therefore, ‘single variable’ response curves were used. MaxEnt is less sensitive than other modeling approaches to multicollinearity between environmental covariates; thus, it is deemed unnecessary to remove correlated predictors (Elith et al., 2011).

Model validation was performed using a data partitioning method: a 10-fold cross validation, which divides the data into ten mutually exclusive subsets. Each subset was removed sequentially and the model was fitted against the retained data and tested against the removed subset (Hastie et al., 2009; Elith et al., 2011). Model fit was evaluated using area under the curve (AUC), where a value close to one indicates a very good model fit (Fielding & Bell, 1997).

Validation of model predictive accuracy using independent data sets is rarely undertaken (Araújo et al., 2005), but provides an indication of how applicable models are at predicting a species' distribution outside the input data. To evaluate the predictive accuracy of our models, we backcast and forecast each model to the alternative time period to assess whether the model could correctly predict Arkansas River shiner presence in individual segments. One assumption of this approach is that two periods are temporally independent though we recognize some level of autocorrelation is inherent with our procedure as the data sets form a continuous time period. However, examination of the trends in Arkansas River shiner captures (Fig. 2) suggests a potential breakpoint. Also we feel temporal autocorrelation is unlikely to be an overriding factor given the length of the study period and the short-lived nature of the focal species. This approach also allows an assessment of whether the universal relationships between the species locations and environmental parameters were consistent between the two time periods. MaxEnt produces probability of presence as ranked scores (cumulative probabilities) (Parolo et al., 2008). Therefore, spatial congruence between the two models in each time period was assessed using Spearman's rho correlation to test the relative rank of each segment across the two models.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Captures of Arkansas River shiner were recorded between 1923 and 2010 in 34 rivers within the study area. A total of 182 unique river segments was identified, with the greatest number of segments (number of segments indicated in parentheses) from the Canadian (59), Arkansas (including the Salt Fork) (34), Cimarron (28), and North Canadian (or Beaver River) (16) rivers. Captures of Arkansas River shiner in new areas increased at a rate of approximately 20–30 river segments per decade until the 1990s when only five new segments were recorded (Fig. 2). The number of river segments where Arkansas River shiner was recorded increased until the 1980s (Fig. 2), and then declined dramatically. The species was only recorded at three sites outside of the Canadian River catchment after 1990, and therefore this was used as the split point for the historic vs. current models. The MaxEnt models were constructed using 47 locations for the current model and 163 locations for the historic model.

Species distribution models

Discharge, stream order, land-use, and geology provided the greatest contribution to the current model (Table 2), whereas stream order, mean temperature of the coldest quarter, and precipitation in the driest month had the highest permutation importance (>15%). The model predicted a high probability of presence (>50%) for 343 individual river segments including a large section of the Canadian River, with areas of particularly high probability (>80%) near the confluence of Revuelto Creek, New Mexico (Fig. 3). There were no areas with a predicted probability >50% outside the Canadian River catchment. In the current model, Arkansas River shiner was most commonly associated with the rangeland land-use category and the coarse-grained mixed clastic and mixed clastic/carbonate geology types. Segments with a >50% predicted probability of occurrence had mean annual temperatures between approximately 14 and 15.5 °C, whereas for annual precipitation predicted probability of occurrence was >45% between 400 and 1000 mm. Probability of presence was greatest in segments with a stream order of six to eight (Fig. 4a) and mean annual discharges >10 m3 s−1 and <110 m3 s−1 (Fig. 4b). Probability of presence increased rapidly when the mean temperature of the wettest quarter rose above 10 °C to an asymptote at approximately 23 °C (Fig. 4c). In relation to the drift of ichtyoplankton, stream segments with a higher probability of Arkansas River shiner presence had a free-flowing length of 375–750 km (Fig. 4d).

Table 2. Percent contribution of the environmental variables (only those contributing ≥ 1% are displayed)
VariableContribution (%)
CurrentHistoric
Discharge (m3s−1)28.421.6
Stream order (Strahler)25.241.7
Land-use12.41.0
Geology11.23.3
Drift (km)4.6_
Temperature seasonality (°C)4.1_
Maximum elevation (m)2.8_
Precipitation of driest quarter (mm)2.3_
Precipitation seasonality (%)2.1_
Mean temperature of driest quarter (°C)1.8_
Slope (cm cm−1)1.45.1
Precipitation of warmest quarter (mm)1.01.0
Mean temperature of wettest quarter (°C)_8.6
Mean temperature of coldest quarter (°C)_4.1
Maximum temperature of warmest month (°C)_2.2
Precipitation of wettest quarter (mm)_2.0
Annual temperature range (°C)_1.3
Annual precipitation (mm)_1.2
Minimum temperature of coldest month (°C)_1.1
image

Figure 3. The probability of presence (>0.5) for Arkansas River shiner predicted from the current and historic species distribution models.

Download figure to PowerPoint

image

Figure 4. Relationship between the probability of Arkansas River shiner presence (red line current model, blue line historic model) and (a) stream order (modified Strahler); (b) discharge (m3 s−1) at the downstream end of segment; (c) mean temperature of the wettest quarter (°C) and (d) downstream drift (km).

Download figure to PowerPoint

Stream order, discharge, mean temperature of the wettest quarter, and slope contributed the most to the predictive ability of the historic model (Table 2). Discharge, stream order, and precipitation seasonality had permutation importance values >15%. A much greater area was predicted as having >50% (1312 individual river sections across 20 rivers) probability of presence for Arkansas River shiner, including large sections of the Canadian, North Canadian, Cimarron, and Arkansas rivers (Fig. 3). Areas of particularly high presence probabilities (>80%) again included the Canadian River near the confluence of Revuelto Creek, New Mexico and north of Minco, Oklahoma and near Norman, Oklahoma. Areas of high probability were also predicted for the Arkansas River upstream of Tulsa, Oklahoma, the Cimarron River near Perkins, Oklahoma and the North Canadian River near Woodward, Oklahoma. Arkansas River shiner was most likely to be present in stream segments with a stream order > five (Fig. 4a), and those with a mean annual discharge between 17 and 590 m3 s−1 (Fig. 4b). As in the current model, the probability of presence increased rapidly after 10 °C to an asymptote at approximately 23 °C (temperature of the wettest quarter, Fig. 4c). The species was more likely present in segments with underlying coarse-grained mixed clastic rock. As with the current model, segments with the highest predicted probabilities had annual temperatures between 14 and 15.5 °C and annual precipitation between 400 and 1000 mm. The probability of Arkansas River shiner presence increased substantially in segments with >600 km of river length available for downstream drifting ichthyoplankton (Fig. 4d).

Model performance and predictive accuracy

Performance of both models was ‘excellent’ (Araújo & Guisan, 2006) with a mean AUC on the test data across the ten runs of 0.98 ± 0.03 SD (current model) and 0.96 ± 0.01 SD (historic model). The Spearman's rank correlation suggested reasonable accuracy of the predictions projected from the alternate time period. The model trained on historic locations and environmental parameters projected onto the environmental conditions present in the current landscape (Spearman's rho = 0.66, P < 0.01) approximately matched the predicted probability of presence for some areas of the upper and middle Canadian River (Fig. 5, Forecast Distribution). However, of the initial 343 segments in the current model predicted as having a high (>50%) probability of presence, only 6% were again highlighted in the forecast model. The forecast distribution instead predicted segments of the Arkansas, Cimarron, and North Canadian rivers as having high (>50%) probability of Arkansas River shiner presence that were not apparent in the current model (Fig. 3, Current Distribution). The model trained on current locations and environmental parameters projected onto the historic landscape was a closer match (Spearman's rho = 0.71, P < 0.01), but failed to predict likely Arkansas River shiner presence in the upper reaches of the Arkansas, Canadian, Cimarron and North Canadian rivers (Fig. 5, Backcast Distribution) as seen in the historic model (Fig. 3, Historic Distribution). Of the initial 1312 segments in the historic model predicted as having a high (>50%) probability of presence, 40% were again highlighted in the backcast model.

image

Figure 5. The probability of Arkansas River shiner presence (>0.5) predicted from the historic model and ‘forecast’ onto current environmental conditions and from the current model and ‘backcast’ on historical environmental conditions. Congruence with prediction from current and historic models varies spatially, with areas outside the Canadian River predicted in the forecast model, and truncated distributions in upper reaches of the major rivers in the backcast model.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Distributions are assumed to represent adaptive significance to a species and as such SDMs have been used extensively to predict distributions across terrestrial, freshwater, and marine landscapes (Elith & Leathwick, 2009). Understanding a species' distributions can drive conservation planning (Liang et al., in press; Maloney et al., 2013), identify areas of conservation importance (Moilanen, 2005), highlight factors related to species decline (e.g., Anadón et al., 2007; Lötters et al., 2009), and predict future distributions due to anthropogenic disturbance (Dyer et al., 2013). We provide a basis for future conservation efforts for the Arkansas River shiner by quantitatively mapping the current and historic range of the species. The study highlights river fragments that represent potential refuges for endemic Great Plains fishes, provides a framework for selecting reintroduction sites and offers information for understanding causes of the species' decline. The SDM approach also allows an evaluation of the role of multiple environmental factors in determining species presence. Our study adds to the body of evidence linking the decline of members of the pelagic broadcast spawning cyprinids to fragmentation and changes in the natural flow regime (e.g., Bestgen et al., 1989; Platania & Altenbach, 1998; Bonner & Wilde, 2000). By forecasting and backcasting the models, we can assess model predictive power and examine how the relationship between presence and environmental covariates has changed in response to human disturbance.

Previous studies have used a combination of field surveys, literature, and museum records to plot changes in location points as a means to map the decline of the Arkansas River shiner (Cross et al., 1985; Pigg, 1987, 1991; Pigg et al., 1997, 1998, 1999). Although these efforts provide valuable information on the former presence of the species, they are unable to quantitatively predict the continuous range of the species and correlate presence with variation in multiple environmental conditions at the landscape scale. The current model conforms with our understanding of the present status of Arkansas River shiner, with the species confined to two fragments of the Canadian River (Wilde, 2002; Parham, 2009). This finding supports the suggestion by Hoagstrom et al. (2011) that these two fragments should be considered refuges for endemic plains fishes. These fragments support not only Arkansas River shiner but also populations of two other declining pelagic broadcast spawners, plains minnow Hybognathus placitus and peppered chub Macrhybopsis tetranema and other Great Plains fishes: Plains sand shiner N. stramineus missuriensis and Northern plains killifish Fundulus kansae. Persistence of Arkansas River shiner in this section of the Canadian River may be driven by the presence of flow conditions suitable to allow successful reproduction coupled with a sufficient river fragment length to allow ichthyoplankton to reach the free-swimming stage (Bonner & Wilde, 2000; Durham & Wilde, 2008). The historic model also closely matches the perceived unimpacted distribution of Arkansas River shiner (Cross et al., 1985; Larson, 1988), with extensive sections of the main rivers of the Arkansas drainage predicted as having supported the species. Should prevailing environmental conditions in these areas be remediated so that they could support Arkansas River shiner, the historical model provides a framework for selecting potential reintroduction sites (e.g., Pearce & Lindenmayer, 1998). For example, if river fragments of sufficient length to allow the ichtyoplankton to reach the free-swimming stage are present, water releases from impoundments that maintain perennial base flows (Hoagstrom et al., 2008) and more closely match the natural flow regime (Dudley & Platania, 2007) may allow the species to successfully complete its life history.

Results from SDMs can be used to infer the ecological requirements of a species (Graham & Hijmans, 2006) and thus can provide valuable information on the factors limiting a population or driving decline. Geology and climate are considered the primary factors that structure the distribution of aquatic biota (Hynes, 1975). These ‘ultimate’ indirect factors determine physicochemical processes, affect resource availability and impact biotic and abiotic factors at finer scales (e.g., salinity and temperature) (Stevenson, 1997). Geology contributed significantly to the current model, but was less important in structuring the historic distribution. Geology influences fine scale factors such as substrate and geomorphology (e.g., Polivka, 1999; Kehmeier et al., 2007) and physicochemical properties (e.g., salinity, Matthews & Hill, 1980) that determine Arkansas River shiner presence. The climatic variables contributed <15% to the predictive power of the current and ca. 25% in the historic models, with temperature variables contributing more than precipitation. Probability of presence was generally greater at higher temperatures. These findings are supported by other studies that indicate Arkansas River shiner can tolerate a wide range of temperatures (0.4–31.7 °C, Bonner et al., 1997) and has a high critical thermal maximum (35.92–38.64 °C, Matthews & Maness, 1979; Matthews, 1987). The difference in the contribution level of geology and climatic variables between the time periods is likely to be an artifact of changes in the relative importance of other predictor variables and changing relationships due to shifts in the Arkansas River shiner's distribution. Another indirect factor, stream order, was an important predictor across both models. Stream order is correlated to a number of direct environment factors (Vannote & Sweeney, 1980; Benda et al., 2004) and is therefore often a good predictor of species distribution (Smith & Kraft, 2005; Mugodo et al., 2006; although see Matthews, 1986). Our models were in line with previous studies that indicate Arkansas River shiner is generally confined to the larger tributaries of the Arkansas River catchment (Moore, 1944; Pigg, 1991). The difference in the shape of the relationship between stream order and Arkansas River shiner presence across the two time periods (Fig. 4a) likely stems from the recent extirpation of the species from the larger rivers of the study area.

Human modification has altered many physical river processes, modifying the natural flow regime (Poff et al., 1997). Aquatic organisms have evolved life-history strategies in direct response to natural flow conditions (Bunn & Arthington, 2002) and thus when anthropogenic impacts push environmental conditions past a tipping point, ecosystems can shift from one state to another (Scheffer et al., 2009). Pelagic broadcast spawning cyprinids have physiological adaptations (e.g., cutaneous sense organs and brain morphology, Moore, 1950; Davis & Miller, 1967; Huber & Rylander, 1992) especially suited to the harsh and highly variable conditions naturally prevalent in Great Plains rivers (Matthews, 1987; Taylor et al., 1993). Water demand by humans has led to construction of impoundments throughout the Great Plains (Limbird, 1993), leading to changes in the flow regime and a reduction in suspended sediment loads (and turbidity), potentially providing a competitive advantage to sight-feeding fishes (Griffith, 2003; Quist et al., 2004). Within our models, discharge ranked as one of the most important predictors of Arkansas River shiner presence, although the shape of the relationship was markedly different between time periods. The difference in response curves is probably driven by the reduction in range of the species, with Arkansas River shiner now extirpated from segments typified by higher discharges such as the Arkansas River main stem and the downstream sections of the Canadian River. Changes in the timing, magnitude, and variability of high discharge events has been proposed as a cause of decline for Arkansas River shiner (Cross et al., 1983; Platania & Altenbach, 1998) and other pelagic broadcast spawning cyprinids (e.g., Sabine shiner Notropis sabinae, Suttkus & Mettee, 2009; sharpnose shiner Notropis oxyrhynchus and smalleye shiner Notropis buccula, Durham & Wilde, 2009). Bonner & Wilde (2000) concluded the construction of Ute and Meredith reservoirs led to species replacements in the fish assemblage of the Canadian River, although differences were related to the magnitude of the change in discharge below each impoundment potentially showing a threshold effect. Below Ute Reservoir, where mean annual discharge was reduced by approximately 38%, Arkansas River shiner was still one of the dominant species; however, downstream of Lake Meredith (76% reduction in mean annual discharge) they made up only 0.2% of the community's relative abundance (Bonner & Wilde, 2000). This finding suggests that the relationship between persistence of Arkansas River shiner and discharge is complex and nonlinear (see also Fig. 4b) and that a tolerance threshold may have been exceeded. It is also likely that the reduction in discharge is interacting with other factors (e.g., temperature, salinity) to increase environmental stress.

The timing of high and low flow events in Great Plains rivers was historically subject to extensive temporal variability (Poff, 1996). Under such conditions, it has been suggested that aquatic species may undertake bet-hedging strategies during part of their life history (see Lytle & Poff, 2004). Pelagic broadcast spawning cyprinids display reproductive adaptations (fractional or extended spawning; Fausch & Bestgen, 1997) as a mechanism to cope with variability in timing of high-flow events. Elevated discharge has been proposed as a trigger for the onset of spawning in pelagic broadcast spawners (Moore, 1944; Bestgen et al., 1989; however, see Durham & Wilde, 2006). The successful development of semi-buoyant pelagic eggs requires access to free-flowing stream segments (approximately 3 days, Moore, 1944; although this is impacted by total suspended and total dissolved solids, J.S. Mueller, unpublished data). Fragmentation of Great Plains rivers led to reduced discharge potentially causing eggs to be transported near the channel floor (Worthington et al., 2013), making them vulnerable to abrasion or smothering (Bestgen et al., 1989; Osborne et al., 2006). In addition, the presence of reservoirs may also increase vulnerability of both eggs and larvae to predation (Luttrell et al., 1999; Pompeu et al., 2012) and reduce channel complexity, potentially increasing downstream drift velocity and distance required for ichtyoplankton development (Dudley & Platania, 2007; Medley et al., 2007; Widmer et al., 2012). Studies have used a number of methods to calculate the length of river required for Arkansas River shiner to complete their life history (e.g., presence/absence, minimum fragment length of 217 km Perkin & Gido, 2011; constant drift rate, 360 km in unimpeded river sections Platania & Altenbach, 1998). The drift component of our analyses contributed little to model predictive power when ichtyoplankton were not constrained by the presence of impoundments (historic model); however, the introduction of barriers resulted in an increase in relative contribution of the drift parameter (current model). Examination of the response curve revealed segments, in terms of optimal drift potential, had a free flowing length of 375–750 km.

Species distribution models (SDMs) are frequently being used for conservation planning and species risk assessments under environmental change (e.g., Loiselle et al., 2003; Esselman & Allan, 2011); therefore, robust validation of model predictive power is crucial (Araújo et al., 2005; Elith & Leathwick, 2009). We used two approaches to assess model predictive accuracy: a traditional data partitioning method and a rarely used backcast/forecast procedure (Nunes et al., 2007; Parra & Monahan, 2008). The models for Arkansas River shiner trained in one time period and projected to the alternate period performed adequately, although predictive accuracy varied spatially. The forecasting/backcasting approach also provided an opportunity to examine whether the relationship between the environmental covariates and presence of Arkansas River shiner held constant through time. Examination of the response curves (e.g., Fig. 4c) suggests the relationship between precipitation and temperature variables and Arkansas River shiner probability of presence was reasonably stable between the historic and current models. However, disturbance of the natural functioning of Great Plains rivers has rendered completely different the relationship between species presence and in-channel factors such as discharge and drift. Therefore, strategies to conserve and enhance Arkansas River shiner populations might benefit from examining the relationship between critical variables (e.g., discharge) and species presence before large scale anthropogenic disturbance. The lack of complete similarity between the outputs is not surprising and can also be attributed to several other factors. Modeling the environmental niche of mobile species or those with ontogenetic shifts in resource use is challenging (Elith & Leathwick, 2009), with modeling accuracy shown to decrease for species with greater mobility (e.g., Pöyry et al., 2008). It is likely that the SDMs are not completely accurate as relationships underlying model structure have changed due to disturbance; however, they provide a useful prediction of the current and former distribution of the Arkansas River shiner (see Box, 1979, p. 2). More importantly, the models provide some indication of environmental changes (e.g., disturbance of the natural flow regime and river fragmentation) that have occurred over time and impacted a large number of Great Plains species. The accuracy of the prediction from one time period to the other appears to vary spatially. Metapopulations may be adapted to different environmental conditions (Räsänen et al., 2003; Sinclair et al., 2010); for example, onset of spawning in common galaxias Galaxias maculatus (Barbee et al., 2011) and thermal tolerance and growth in orangethroat darter Etheostoma spectabile (Strange et al., 2002) vary across the species' distribution. Furthermore, the relationship between species and the environment is not always static through time (see Skelly et al., 2007), although adaptation may not be rapid enough to mitigate the effects of environmental change (Crisp et al., 2009).

A potential concern with the use of species locations gathered from museum and university collections and published and gray literature are spatial bias within the data due to the lack of a structured sampling approach (Graham et al., 2004; Newbold, 2010). Sample selection bias within presence-only models will result in a model that reflects both the species and the sampling distributions (Elith et al., 2011). For Arkansas River shiner, it is likely the sampling site will be biased to sections close to access points (e.g., road crossing); however, it is unlikely that this will overly impact the model predictions as the majority of the environmental covariates are unlikely to be correlated to the location of the sampling points (see Phillips et al., 2009). A limitation of presence-only SDMs is the assumption that detection probability is constant rather than having the potential to vary with the environmental covariates used to model the distribution (Yackulic et al., 2013). No information was available on how detection of Arkansas River shiner varies with the environmental variables used in this study; however, between sampling sites single pass detection probability has been estimated to range from 0.54 to 1 (Widmer et al., 2012; Archdeacon & Davenport, 2013). Although we tried to incorporate the most ecologically relevant environmental covariates, this was constrained by data availability (see Dormann, 2007). Model refinement could include hydrological parameters more pertinent to the species' life-history requirements (e.g., discharge during reproduction and the ichtyoplankton drift stages; Durham & Wilde, 2009). As with many SDMs (Cassini, 2011), our predictions are restricted by the absence of biological interactions. For example, the rapid expansion in range and abundance of the introduced, but perhaps ecologically similar Red River shiner Notropis bairdi, has been suggested as contributing to the decline of Arkansas River shiner (Cross et al., 1983). Although the data are not truly temporally independent (Araújo et al., 2005), the use of a time-step approach may provide a more meaningful approximation of species' response to environmental change, providing a basis for predictions of future changes in climate, land-use, and discharge components (Parra & Monahan, 2008).

The ichthyofauna of the Great Plains provides an opportunity to examine the impact of anthropogenic disturbance on species considered tolerant to harsh and variable conditions. These species have adapted to extremes and fluctuations in temperature, salinity, and discharge; despite this, many Great Plains fishes are declining or extinct (Hoagstrom et al., 2011). Conservation efforts are difficult or impossible if the causes of the species' initial decline are unknown (Sarrazin & Barbault, 1996; Bonebrake et al., 2010). Recovery requires a basic understanding of a species' life history; unfortunately, little is known about the ecology of many Great Plains fishes (Fausch & Bestgen, 1997). Comparatively, Arkansas River shiner has received greater attention; however, the causes of its contraction in range is still uncertain. Several factors have been suggested as possible causes for the decline of the Arkansas River shiner; however, this is the first attempt to quantitatively model the interaction between multiple impacts across the species' entire range. To further strengthen our understanding, future studies that quantify the timing and magnitude of discharge required for these species to fulfill their life history are warranted.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

This research is a contribution of the Oklahoma Cooperative Fish and Wildlife Research Unit (US Geological Survey, Oklahoma Department of Wildlife Conservation, Oklahoma State University, and Wildlife Management Institute cooperating) with collaboration from the Texas Cooperative Fish and Wildlife Research Unit. Funding was provided by the US Fish and Wildlife Service, Great Plains Landscape Conservation Cooperative (US Fish and Wildlife Service agreement F11AP00112). Any use of trade, firm, or product names is for descriptive purposes and does not imply endorsement by the US Government. We thank Dr. Bernard Kuhajda, The University of Alabama; Dr. Anthony Echelle, Oklahoma State University; Randy Parham, Oklahoma Department of Environmental Quality; Dr. Edith Marsh-Matthews, University of Oklahoma; Dr. Chris Taylor, Illinois Natural History Survey; Robert Robins, University of Florida; Melissa Mata, United States Fish and Wildlife Service; Dr. Nancy Glover McCartney, University of Arkansas Collections Facility; Dr. Darren Pollock, Eastern New Mexico University; Dr. Dean Hendrickson, University of Texas at Austin; Dr. Aaron Place, Northwestern Oklahoma State University; Brian Wagner, Arkansas Game and Fish Commission; Jason Childress, Oklahoma Water Resources Board; Karen Morton, Perot Museum of Nature and Science for help locating Arkansas River shiner location records and Mark Gregory, Oklahoma State University for GIS assistance.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Anadón JD, Giménez A, Martínez M, Palazón JA, Esteve MA (2007) Assessing changes in habitat quality due to land use changes in the spur-thighed tortoise Testudo graeca using hierarchical predictive habitat models. Diversity and Distributions, 13, 324331.
  • Anderson JR, Hardy EE, Roach JT, Witmer RE (1976) A Land Use and Land Cover Classification System for Use with Remote Sensor Data. USGS Professional Paper 964. U.S. Geological Survey, Reston, Virginia.
  • Araújo MB, Guisan A (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 16771688.
  • Araújo MB, Pearson RG, Thuiller W, Erhard M (2005) Validation of species–climate impact models under climate change. Global Change Biology, 11, 15041513.
  • Archdeacon TP, Davenport SR (2013) Comment: detection and population estimation for small-bodied fishes in a sand-bed river. North American Journal of Fisheries Management, 33, 446452.
  • Barbee NC, Hale R, Morrongiello J, Hicks A, Semmens D, Downes BJ, Swearer SE (2011) Large-scale variation in life history traits of the widespread diadromous fish, Galaxias maculatus, reflects geographic differences in local environmental conditions. Marine and Freshwater Research, 62, 790800.
  • Benda L, Poff NL, Miller D, Dunne T, Reeves G, Pess G, Pollock M (2004) The network dynamics hypothesis: how channel networks structure riverine habitats. BioScience, 54, 413427.
  • Bestgen KR, Platania SP, Brooks JE, Probst DL (1989) Dispersal and life history traits of Notropis girardi (Cypriniformes: Cyprinidae), introduced into the Pecos River, New Mexico. The American Midland Naturalist, 122, 228235.
  • Bonebrake TC, Christensen J, Boggs CL, Ehrlich PR (2010) Population decline assessment, historical baselines, and conservation. Conservation Letters, 3, 371378.
  • Bonner TH, Wilde GR (2000) Changes in the Canadian River fish assemblage associated with reservoir construction. Journal of Freshwater Ecology, 15, 189198.
  • Bonner TH, Wilde GR (2002) Effects of turbidity on prey consumption by prairie stream fishes. Transactions of the American Fisheries Society, 131, 12031208.
  • Bonner TH, Wilde GR, Jimenez R Jr, Patino R (1997) Habitat Use and Ecology of the Arkansas River Shiner and Speckled Chub in the Canadian River, New Mexico and Texas. U.S. Fish and Wildlife Service, Tulsa, OK.
  • Box GE (1979) Some problems of statistics and everyday life. Journal of the American Statistical Association, 74, 14.
  • Brewer SK, Rabeni CF, Sowa SP, Annis G (2007) Natural landscape and stream segment attributes influencing the distribution and relative abundance of riverine smallmouth bass in Missouri. North American Journal of Fisheries Management, 27, 326341.
  • Bunn SE, Arthington AH (2002) Basic principles and ecological consequences of altered flow regimes for aquatic biodiversity. Environmental Management, 30, 492507.
  • Cardillo M, Meijaard E (2012) Are comparative studies of extinction risk useful for conservation? Trends in Ecology & Evolution, 27, 167171.
  • Cassini MH (2011) Ecological principles of species distribution models: the habitat matching rule. Journal of Biogeography, 38, 20572065.
  • Crisp MD, Arroyo MT, Cook LG et al. (2009) Phylogenetic biome conservatism on a global scale. Nature, 458, 754756.
  • Cross FB, Gorman OT, Haslouer SG (1983) The Red River shiner, Notropis bairdi, in Kansas with notes on depletion of its Arkansas River cognate, Notropis girardi. Transactions of the Kansas Academy of Science, 86, 9398.
  • Cross FB, Moss RE, Collins JT (1985) Assessment of Dewatering Impact on Stream Fisheries in the Arkansas and Cimarron Rivers. Museum of Natural History, University of Kansas, Kansas, USA.
  • Cushman SA (2006) Effects of habitat loss and fragmentation on amphibians: a review and prospectus. Biological Conservation, 128, 231240.
  • Davis BJ, Miller RJ (1967) Brain patterns in minnows of the genus Hybopsis in relation to feeding habits and habitat. Copeia, 1967, 139.
  • Dodds WK, Gido K, Whiles MR, Fritz KM, Matthews WJ (2004) Life on the edge: the ecology of Great Plains prairie streams. BioScience, 54, 205216.
  • Dormann CF (2007) Promising the future? Global change projections of species distributions. Basic and Applied Ecology, 8, 387397.
  • Dudley RK, Platania SP (2007) Flow regulation and fragmentation imperil pelagic-spawning riverine fishes. Ecological Applications, 17, 20742086.
  • Durham BW, Wilde GR (2006) Influence of stream discharge on reproductive success of a prairie stream fish assemblage. Transactions of the American Fisheries Society, 135, 16441653.
  • Durham BW, Wilde GR (2008) Composition and abundance of drifting fish larvae in the Canadian River, Texas. Freshwater Ecology, 23, 273280.
  • Durham BW, Wilde GR (2009) Effects of streamflow and intermittency on the reproductive success of two broadcast-spawning cyprinid fishes. Copeia, 2009, 2128.
  • Dyer JJ, Brewer SK, Worthington TA, Bergey EA (2013) The influence of coarse-scale environmental features on current and predicted future distributions of narrow-range endemic crayfish populations. Freshwater Biology, 58, 10711088.
  • Elith J, Leathwick JR (2009) Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics, 40, 677697.
  • Elith J, Graham CH, Anderson RP et al. (2006) Novel methods improve prediction of species' distributions from occurrence data. Ecography, 29, 129151.
  • Elith J, Phillips SJ, Hastie T, Dudík M, Chee YE, Yates CJ (2011) A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17, 4357.
  • Esselman PC, Allan JD (2011) Application of species distribution models and conservation planning software to the design of a reserve network for the riverine fishes of northeastern Mesoamerica. Freshwater Biology, 56, 7188.
  • Fausch KD, Bestgen KR (1997) Ecology of fishes indigenous to the central and southwestern Great Plains. In: Ecology and Conservation of Great Plains Vertebrates (eds Knopf FL, Samson FB), pp. 131166. Springer-Verlag, New York.
  • Felley JD, Cothran EG (1981) Notropis bairdi (Cyprinidae) in the Cimarron River, Oklahoma. The Southwestern Naturalist, 25, 564.
  • Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation, 24, 3849.
  • Fry FEJ (1971) The effect of environmental factors on the physiology of fish. In: Fish Physiology, Volume 6 Environmental Relations and Behavior (eds Hoar WS, Randall DJ), pp. 198. Academic Press Inc., New York.
  • Fry J, Xian G, Jin S et al. (2011) Completion of the 2006 national land cover database for the conterminous United States. Photogrammetric Engineering and Remote Sensing, 77, 858864.
  • Gido KB, Dodds WK, Eberle ME (2010) Retrospective analysis of fish community change during a half-century of landuse and streamflow changes. Journal of the North American Benthological Society, 29, 970987.
  • Graham CH, Hijmans RJ (2006) A comparison of methods for mapping species ranges and species richness. Global Ecology and Biogeography, 15, 578587.
  • Graham CH, Ferrier S, Huettman F, Moritz C, Peterson AT (2004) New developments in museum-based informatics and applications in biodiversity analysis. Trends in Ecology & Evolution, 19, 497503.
  • Griffith JA (2003) Regions of high fish diversity for conservation concern within the US Central Plains. Journal of Freshwater Ecology, 18, 451457.
  • Hall RI, Leavitt PR, Quinlan R, Dixit AS, Smol JP (1999) Effects of agriculture, urbanization, and climate on water quality in the northern Great Plains. Limnology and Oceanography, 44, 739756.
  • Halpern BS, Diamond J, Gaines S et al. (2012) Near-term priorities for the science, policy and practice of Coastal and Marine Spatial Planning (CMSP). Marine Policy, 36, 198205.
  • Hastie T, Tibshirani R, Friedman JH (2009) The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd edn. Springer-Verlag, New York.
  • Hernandez PA, Graham CH, Master LL, Albert DL (2006) The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography, 29, 773785.
  • Hoagstrom CW, Brooks JE, Davenport SR (2008) Spatiotemporal population trends of Notropis simus pecosensis in relation to habitat conditions and the annual flow regime of the Pecos River, 1992–2005. Copeia, 2008, 515.
  • Hoagstrom CW, Brooks JE, Davenport SR (2011) A large-scale conservation perspective considering endemic fishes of the North American plains. Biological Conservation, 144, 2134.
  • Huber R, Rylander MK (1992) Brain morphology and turbidity preference in Notropis and related genera (Cyprinidae, Teleostei). Environmental Biology of Fishes, 33, 153165.
  • Hughes DA (2005) Hydrological issues associated with the determination of environmental water requirements of ephemeral rivers. River Research and Applications, 21, 899908.
  • Hynes HBN (1975) The stream and its valley. Proceedings of the International Association of Theoretical and Applied Limnology, 19, 115.
  • Jelks HL, Walsh SJ, Burkhead NM et al. (2008) Conservation status of imperiled North American freshwater and diadromous fishes. Fisheries, 33, 372407.
  • Jenkins M (2003) Prospects for biodiversity. Science, 302, 11751177.
  • Kehmeier JW, Valdez RA, Medley CN, Myers OB (2007) Relationship of fish mesohabitat to flow in a sand-bed southwestern river. North American Journal of Fisheries Management, 27, 750764.
  • Larson RD (1988) Present Status and Distribution of the Arkansas River Shiner, Notropis girardi (Pisces: Cyprinidae), and Possible Cause for its Decline. Unpublished MS thesis, Oklahoma State University, Stillwater, OK.
  • Liang L, Fei S, Ripy JB, Blandford BL, Grossardt T (in press) Stream habitat modelling for conserving a threatened headwater fish in the Upper Cumberland River Kentucky. River Research and Applications, doi: 10.1002/rra.2605.
  • Limbird RL (1993) The Arkansas River: a changing river. In: Restoration Planning for Rivers of the Mississippi River Ecosystem (eds Hesse LW, Stalnaker CB, Benson NB, Zuboy JR), pp. 8194. National Biological Survey, Washington, DC.
  • Loiselle BA, Howell CA, Graham CH, Goerck JM, Brooks T, Smith KG, Williams PH (2003) Avoiding pitfalls of using species distribution models in conservation planning. Conservation Biology, 17, 15911600.
  • Lötters S, Kielgast J, Bielby J et al. (2009) The link between rapid enigmatic amphibian decline and the globally emerging chytrid fungus. EcoHealth, 6, 358372.
  • Luttrell GR, Echelle AA, Fisher WL, Eisenhour DJ (1999) Declining status of two species of the Macrhybopsis aestivalis complex (Teleostei: Cyprinidae) in the Arkansas River basin and related effects of reservoirs as barriers to dispersal. Copeia, 1999, 981989.
  • Lytle DA, Poff NL (2004) Adaptation to natural flow regimes. Trends in Ecology & Evolution, 19, 94100.
  • Maloney KO, Weller DE, Michaelson DE, Ciccotto PJ (2013) Species distribution models of freshwater stream fishes in Maryland and their implications for management. Environmental Modeling & Assessment, 18, 112.
  • Matthews WJ (1986) Geographic variation in thermal tolerance of a widespread minnow Notropis lutrensis of the North American mid-west. Journal of Fish Biology, 28, 407417.
  • Matthews WJ (1987) Physiochemical tolerance and selectivity of stream fishes as related to their geographic ranges and local distributions. In: Community and Evolutionary Ecology of North American Stream Fishes (eds Matthews WJ, Heins DC), pp. 111120. University of Oklahoma Press, Norman, Oklahoma.
  • Matthews WJ (1988) North American prairie streams as systems for ecological study. Journal of the North American Benthological Society, 7, 387409.
  • Matthews WJ, Hill LG (1980) Habitat partitioning in the fish community of a southwestern river. The Southwestern Naturalist, 25, 5166.
  • Matthews WJ, Maness JD (1979) Critical thermal maxima, oxygen tolerances and success of cyprinid fishes in a southwestern river. The American Midland Naturalist, 102, 374377.
  • McKay L, Bondelid T, Rea A, Johnston C, Moore R, Dewald T (2012) NHDPlus Version 2: User Guide. Available at: http://www.horizon-systems.com/nhdplus/NHDPlusV2_documentation.php (accessed 27 March 2012).
  • Medley CN, Kehmeier JW, Myers OB, Valdez RA (2007) Simulated transport and retention of pelagic fish eggs during an irrigation release in the Pecos River, New Mexico. Journal of Freshwater Ecology, 22, 499513.
  • Moilanen A (2005) Reserve selection using nonlinear species distribution models. The American Naturalist, 165, 695706.
  • Moore GA (1944) Notes on the early life history of Notropis girardi. Copeia, 1944, 209214.
  • Moore GA (1950) The cutaneous sense organs of barbeled minnows adapted to life in the muddy waters of the Great Plains region. Transactions of the American Microscopical Society, 69, 6995.
  • Mora MA, Grant WE, Wilkins L, Wang HH (2013) Simulated effects of reduced spring flow from the Edwards Aquifer on population sie of the fountain darter (Etheostoma fonnticola). Ecological Modelling, 250, 235243.
  • Mugodo J, Kennard M, Liston P, Nichols S, Linke S, Norris RH, Lintermans M (2006) Local stream habitat variables predicted from catchment scale characteristics are useful for predicting fish distribution. Hydrobiologia, 572, 5970.
  • Newbold T (2010) Applications and limitations of museum data for conservation and ecology, with particular attention to species distribution models. Progress in Physical Geography, 34, 322.
  • Nix HA (1986) A biogeographic analysis of Australian elapid snakes. In: Atlas of Elapid Snakes of Australia (ed. Longmore R), pp. 415. Australian Government Publishing Service, Canberra.
  • Nunes MFC, Galetti M, Marsden S, Pereira RS, Peterson AT (2007) Are large-scale distributional shifts of the blue-winged macaw (Primolius maracana) related to climate change? Journal of Biogeography, 34, 816827.
  • O'Donnell MS, Ignizio DA (2012) Bioclimatic Predictors for Supporting Ecological Applications in the Conterminous United States. U.S. Geological Survey, Reston, Virginia.
  • Osborne MJ, Benavides MA, Alò D, Turner TF (2006) Genetic effects of hatchery propagation and rearing in the endangered Rio Grande silvery minnow, Hybognathus amarus. Reviews in Fisheries Science, 14, 127138.
  • Parham RW (2009) Structure of assemblages and recent distribution of riverine fishes in Oklahoma. The Southwestern Naturalist, 54, 382399.
  • Parolo G, Graziano R, Alessandro F (2008) Toward improved species niche modelling: Arnica montana in the Alps as a case study. Journal of Applied Ecology, 45, 14101418.
  • Parra JL, Monahan WB (2008) Variability in 20th century climate change reconstructions and its consequences for predicting geographic responses of California mammals. Global Change Biology, 14, 22152231.
  • Pearce J, Lindenmayer D (1998) Bioclimatic analysis to enhance reintroduction biology of the endangered helmeted honeyeater (Lichenostomus melanops cassidix) in southeastern Australia. Restoration Ecology, 6, 238243.
  • Pearson RG, Dawson TP (2003) Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology and Biogeography, 12, 361371.
  • Perkin JS, Gido KB (2011) Stream fragmentation thresholds for a reproductive guild of Great Plains fishes. Fisheries, 36, 371383.
  • Perkin JS, Gido KB (2012) Fragmentation alters stream fish community structure in dendritic ecological networks. Ecological Applications, 22, 21762187.
  • Peterson AT, Papeş M, Eaton M (2007) Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent. Ecography, 30, 550560.
  • Phillips S (2005) A Brief Tutorial On Maxent. AT&T Research, Florham Park, New Jersey.
  • Phillips SJ, Dudík M (2008) Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, 161175.
  • Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231259.
  • Phillips SJ, Dudík M, Elith J, Graham CH, Lehmann A, Leathwick J, Ferrier S (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications, 19, 181197.
  • Pigg J (1987) Survey of fishes in the Oklahoma panhandle and Harper County, northwestern Oklahoma. Proceedings of the Oklahoma Academy of Science, 67, 4559.
  • Pigg J (1991) Decreasing distribution and current status of the Arkansas River shiner, Notropis girardi, in the rivers of Oklahoma and Kansas. Proceedings of the Oklahoma Academy of Science, 71, 515.
  • Pigg J, Coleman M, Gibbs R (1997) Temporal and spatial distribution of cyprinid fishes between 1921 and 1955 in the North Canadian River drainage. Proceedings of the Oklahoma Academy of Science, 77, 4392.
  • Pigg J, Coleman MS, Wright J, Gibbs R, Gido K, Lemmons RR (1998) An ecological investigation of the ichthyofauna in the Deep Fork River, central Oklahoma: 1976 to 1996. Proceedings of the Oklahoma Academy of Sciences, 78, 67110.
  • Pigg J, Gibbs R, Cunningham KK (1999) Decreasing abundance of the Arkansas River shiner in the South Canadian River, Oklahoma. Proceedings of the Oklahoma Academy of Science, 79, 712.
  • Platania SP, Altenbach CS (1998) Reproductive strategies and egg types of seven Rio Grande basin cyprinids. Copeia, 1998, 559569.
  • Poff N (1996) A hydrogeography of unregulated streams in the United States and an examination of scale-dependence in some hydrological descriptors. Freshwater Biology, 36, 7179.
  • Poff NL, Allan JD, Bain MB et al. (1997) The natural flow regime. BioScience, 47, 769784.
  • Polivka KM (1999) The microhabitat distribution of the Arkansas River shiner, Notropis girardi: a habitat-mosaic approach. Environmental Biology of Fishes, 55, 265278.
  • Pompeu PS, Agostinho AA, Pelicice FM (2012) Existing and future challenges: the concept of successful fish passage in South America. River Research and Applications, 28, 504512.
  • Pöyry J, Luoto M, Heikkinen RK, Saarinen K (2008) Species traits are associated with the quality of bioclimatic models. Global Ecology and Biogeography, 17, 403414.
  • Price CP, Nakagaki N, Hitt KJ, Clawges RM (2006) Enhanced Historical Land-use and Land-cover Data Sets of the U.S. Geological Survey. U.S. Geological Survey Data Series 240, Digital Maps. U.S. Geological Survey, Reston, Virginia. Available at: http://pubs.usgs.gov/ds/2006/240/ (accessed 27 November 2012).
  • Quist MC, Guy CS, Bernot RJ, Stephen JL (2004) Factors related to growth and survival of larval walleyes: implications for recruitment in a southern Great Plains reservoir. Fisheries Research, 67, 215225.
  • Räsänen K, Laurila A, Merilä J (2003) Geographic variation in acid stress tolerance of the moor frog, Rana arvalis. I. Local adaptation. Evolution, 57, 352362.
  • Revenga C, Brunner J, Henninger N, Kassem K, Payne R (2000) Pilot Analysis of Global Ecosystems: Freshwater Systems. World Resources Institute, Washington, DC.
  • Ricciardi A, Rasmussen JB (1999) Extinction rates of North American freshwater fauna. Conservation Biology, 13, 12201222.
  • Rodríguez JP, Brotons L, Bustamante J, Seoane J (2007) The application of predictive modelling of species distribution to biodiversity conservation. Diversity and Distribution, 13, 243251.
  • Sarrazin F, Barbault R (1996) Reintroduction: challenges and lessons for basic ecology. Trends in Ecology & Evolution, 11, 474478.
  • Scheffer M, Bascompte J, Brock WA et al. (2009) Early-warning signals for critical transitions. Nature, 461, 5359.
  • Sinclair SJ, White MD, Newell GR (2010) How useful are species distribution models for managing biodiversity under future climates. Ecology and Society, 15, 8. Available at: http://www.ecologyandsociety.org/vol15/iss1/art8/ (accessed 27 January 2013).
  • Skelly D, Joseph L, Possingham H, Freidenburg L, Farrugia T, Kinnison M, Hendry A (2007) Evolutionary responses to climate change. Conservation Biology, 21, 13531355.
  • Smith TA, Kraft CE (2005) Stream fish assemblages in relation to landscape position and local habitat variables. Transactions of the American Fisheries Society, 134, 430440.
  • Stanfield LW, Kilgour BW (in press) How proximity of land use affects stream fish and habitat. River Research and Applications, doi: 10.1002/rra.2585 [Epub ahead of print].
  • Stevenson RJ (1997) Resource thresholds and stream ecosystem sustainability. Journal of the North American Benthological Society, 16, 410424.
  • Stoeser DB, Green GN, Morath LC, Heran WD, Wilson AB, Moore DW, Van Gosen BS (2005) Preliminary Integrated Geologic Map Databases for the United States Central States: Montana, Wyoming, Colorado, New Mexico, Kansas, Oklahoma, Texas, Missouri, Arkansas, and Louisiana, - The State of Oklahoma: U.S. Geological Survey Open-File Report 2005-1351, U.S. Geological Survey, Denver, CO. Available at: http://mrdata.usgs.gov/geology/state/ (accessed 3 October 2012).
  • Strahler AN (1957) Quantitative analysis of watershed geomorphology. Transactions of the American Geophysical Union, 38, 913920.
  • Strange KT, Vokoun JC, Noltie DB (2002) Thermal tolerance and growth differences in orangethroat darter (Etheostoma spectabile) from thermally contrasting adjoining streams. The American Midland Naturalist, 148, 120128.
  • Suttkus RD, Mettee MF (2009) Post-impoundment changes in the cyprinid fauna of the lower Sabine River, Louisiana and Texas. Proceedings Southeastern Fishes Council, 51, 19.
  • Taylor CM, Winston MR, Matthews WJ (1993) Fish species-environment and abundance relationships in a Great Plains river system. Ecography, 16, 1623.
  • US Army Corps of Engineers (2010) National Inventory of Dams. US Army Topographic Engineering Center, Alexandria, Virginia. Available at: http://geo.usace.army.mil/pgis/f?p=397:12 (accessed 22 October 2012).
  • USEPA, USGS (2012) National Hydrography Dataset Plus – NHDPlus, Edition 2.10. U.S. Environmental Protection Agency, Washington, DC and the U.S. Geological Survey, Reston, Virginia. Available at: http://www.horizon-systems.com/NHDPlus/NHDPlusV2_home.php.
  • Van der Putten WH, Macel M, Visser ME (2010) Predicting species distribution and abundance responses to climate change: why it is essential to include biotic interactions across trophic levels. Philosophical Transactions of the Royal Society B: Biological Sciences, 365, 20252034.
  • Vannote RL, Sweeney BW (1980) Geographic analysis of thermal equilibria: a conceptual model for evaluating the effect of natural and modified thermal regimes on aquatic insect communities. The American Naturalist, 115, 667695.
  • Warren ML, Burr BM, Walsh SJ et al. (2000) Diversity, distribution, and conservation status of the native freshwater fishes of the southern United States. Fisheries, 25, 731.
  • Westhoff JT, Rabeni CF, Sowa SP (2011) The distributions of one invasive and two native crayfishes in relation to coarse-scale natural and anthropogenic factors. Freshwater Biology, 56, 24152431.
  • Widmer AM, Burckhardt LL, Kehmeier JW, Gonzales EJ, Medley CN, Valdez RA (2010) Detection and population estimation for small-bodied fishes in a sand-bed river. North American Journal of Fisheries Management, 30, 15531570.
  • Widmer AM, Fluder JJ III, Kehmeier JW, Medley CN, Valdez RA (2012) Drift and retention of pelagic spawning minnow eggs in a regulated river. River Research and Applications, 28, 192203.
  • Wilcove DS (1987) From fragmentation to extinction. Natural Areas Journal, 7, 2329.
  • Wilde GR (2002) Threatened fishes of the world: Notropis girardi Hubbs & Ortenburger, 1929 (Cyprinidae). Environmental Biology of Fishes, 65, 98.
  • Williams CS, Bonner TH (2006) Habitat associations, life history and diet of the Sabine shiner Notropis sabinae in an east Texas drainage. The American Midland Naturalist, 155, 84102.
  • Wisz MS, Hijmans R, Li J, Peterson AT, Graham C, Guisan A (2008) Effects of sample size on the performance of species distribution models. Diversity and Distributions, 14, 763773.
  • Woods AJ, Omernik JM, Butler DR et al. (2005) Ecoregions of Oklahoma (color poster with map, descriptive text, summary tables, and photographs). U.S. Geological Survey, Reston, Virginia. Available at: http://www.epa.gov/wed/pages/ecoregions/ok_eco.htm.
  • Worthington TA, Brewer SK, Grabowski TB, Mueller J (2013) Sampling efficiency of the Moore egg collector. North American Journal of Fisheries Management, 33, 7988.
  • Yackulic CB, Chandler R, Zipkin EF, Royle JA, Nichols JD, Campbell Grant EH, Veran S (2013) Presence-only modelling using MAXENT: when can we trust the inferences? Methods in Ecology and Evolution, 4, 236243.

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
gcb12329-sup-0001-TableS1.docxWord document12KTable S1. Conversion of land-use categories between time periods.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.