Temporal distribution modelling reveals upstream habitat drying and downstream non‐native introgression are squeezing out an imperiled headwater fish

To review the conservation status of Headwater catfish Ictalurus lupus (Girard,1859) in the United States, including quantifying environmental correlates with range contraction and hybridization and introgression with Channel catfish Ictalurus punctatus (Rafinesque, 1818) to inform conservation prioritization.


| INTRODUC TI ON
Growing human population densities and human domination of ecosystems are key factors contributing to a global loss of biodiversity McKee et al., 2004;Turak et al., 2017;Tylianakis et al., 2008). The introduction of non-native species, municipal and industrial development, increased agricultural land use, and anthropogenic greenhouse gas emissions are all contributing factors (Sala et al., 2000). Across systems, two threats to biodiversity that commonly occur together are habitat degradation (i.e., destruction of habitat) and non-native species invasions (Arlinghaus et al., 2016;Dudgeon et al., 2006;Vörösmarty et al., 2010). The loss of biodiversity in aquatic ecosystems (i.e., marine and freshwater) is widely documented (Moyle & Leidy, 1992), yet historically received less attention compared with terrestrial systems (Aerts & Honnay, 2011;Sala et al., 2000). Though freshwater covers only 0.8% of the Earth's surface, freshwater fauna represent 9.5% of known species (Dudgeon et al., 2006;Strayer & Dudgeon, 2010).
Consequently, there is a growing urgency to address threats to biodiversity in freshwater ecosystems through application of systematic conservation planning (Reid et al., 2019).
The aim of systematic conservation planning is to provide conservation managers with decision-support tools that allow for efficient and effective allocation of limited resources (Hermoso et al., 2015;Poiani et al., 2000;Possingham et al., 2015;Williams et al., 2011).
Systematic conservation assessments of freshwater systems and their biota are relatively new because of the spatial and temporal challenges posed by rivers and their drainage networks (Barmuta et al., 2011;Erős et al., 2018). Natural riverine landscapes (riverscapes hereafter) maintain connectivity across four dimensions, including longitudinal (upstream to downstream), lateral (main channels to floodplains), vertical (groundwater to surface water), and temporal (flow variability through time; Ward, 1998). Human alterations to riverscapes affect each of these dimensions and result in changes to natural structuring mechanisms for biotic assemblages (Cooper et al., 2017;Dudgeon et al., 2006;Perkin et al., 2015). Consequently, design of freshwater protected areas must integrate information across multiple riverscape dimensions to identify areas where anthropogenic alterations can be mitigated to benefit the greatest number of species (Hermoso et al., 2012). Freshwater stream fishes serve as biological indicators of multi-dimensional riverscape connectivity (e.g., Perkin et al., 2017;Schmutz & Jungwirth, 1999) and are commonly used to identify freshwater protected areas (Araújo & Williams, 2000;Williams et al., 2011). This means assessments of fish distributions can inform systematic conservation planning by identifying priority conservation areas where preservation or restoration actions might be targeted (D'amen et al., 2017;Hermoso et al., 2016).
Fishes in arid riverscapes are especially in need of conservation.
Drylands, which consist of arid, semi-arid, and desert regions, make up approximately 40% of Earth's terrestrial surface and contain nearly 33% of the human population (James et al., 2013). Streams in drylands rely on vertical connectivity with groundwater to maintain base flow, which usually has limited spatial extent and can vary seasonally (Murray et al., 2003). Human water demand and the frequency and intensity of extreme hydrologic events, such as drought, also affect the availability of water for dryland fishes (Heino et al., 2009;Perkin, Starks, et al., 2019). Thus, the preservation of perennial water sources in dryland regions is a major challenge for conserving aquatic biodiversity (Davis et al., 2017). Furthermore, introductions of non-native fish species are particularly harmful in highly isolated dryland waterbodies where there is increased competition and limited dispersal capacity (Cambray, 2003;Orians, 1995).
Non-native introductions can lead to hybridization where individuals with distinguishable, heritable characters from two distinct populations or groups of populations interbreed. Introgression describes the incorporation of alleles from one species into the gene pool of another, typically through hybridization and backcrossing (Harrison & Larson, 2014). Hybridization and introgression are recognized as both potentially destructive as well as important in the evolutionary process, but present many controversial issues in conservation policy (Allendorf et al., 2001), primarily due to conflict with species-centric management practices (Chafin et al., 2019;Cooke et al., 2005;Fitzpatrick et al., 2015). This means successful conservation planning for dryland fishes must address habitat integrity, natural genetic structure, and potential introgression (Echelle, 1991;Hermoso et al., 2015;Meffe & Vrijenhoek, 1988). Simultaneously measuring each of these features across a riverscape requires multidisciplinary frameworks focused on applying modelling tools, genetic techniques, and spatial conservation planning to benefit species persistence in occupied habitats or reintroduction into restored habitats (Malone et al., 2018).
The Headwater catfish Ictalurus lupus Girard, 1859 is a member of the family Ictaluridae in the order Suliformes and occurs in riffles, runs, and pools of spring-fed streams and small-to moderate-sized rivers in the American Southwest ( Figure 1). The native range of Headwater catfish includes the Rio Grande and Pecos River basins in the United States and Mexico (Hubbs et al., 2008).
Although Headwater catfish was among the least studied North American freshwater fishes just a few decades ago (Gilbert & Burgess, 1980), subsequent studies showed its native distribution in the United States was declining as a result of habitat degradation and competition and introgression with Channel catfish Ictalurus punctatus Rafinesque, 1818 (Ictaluridae, Suliformes) beginning in the 1980s (Bean et al., 2011;Kelsch & Hendricks, 1990;McClure-Baker et al., 2010). Channel catfish are native to streams east of the Rocky  (Hubbs et al., 2008). As a result of these pressures, Headwater catfish was listed as a Species of Special Concern by Williams et al. (1989) andHubbs et al. (2008), Threatened by Jelks et al. (2008), and State Threatened by the Texas Parks and Wildlife Department (M. Bean, professional communication (Kelsch & Hendricks, 1990). However, Headwater catfish is extirpated from a large portion of this range, with only limited portions of the Pecos, Rio Grande, and Frio River basins currently inhabited (Bean et al., 2011;Kelsch & Hendricks, 1990;McClure-Baker et al., 2010). The systems where Headwater catfish persists in the United States are threatened primarily by land use changes related to irrigated agriculture, declining groundwater tables and spring discharges, and continued introduction of non-native Channel catfish via reservoir stockings (Contreras-Balderas & Escalante, 1984;Souza et al., 2006). Jelks et al. (2008) suggested the minimum time period that a species goes undocumented before it is considered extirpated is 20 years; thus, there is a need to understand where Headwater catfish has been documented since 2000 to inform conservation planning. Furthermore, systematic conservation planning is needed to determine the watershed conditions associated with historical and contemporary occurrences, where these conditions currently exist, the genetic integrity of existing populations, and how protective measures might be spatially allocated to ensure persistence of the species.
The goal of this study is to provide a comprehensive review of the conservation status of Headwater catfish in the United States.
Our first objective was to evaluate change in the geographic distribution of Headwater catfish using machine learning methods to construct species distribution models (SDMs) based on historical (1980-1999) and contemporary (2000-2019) presence-absence data and remotely sensed stream network data. Identifying areas suitable for Headwater catfish will inform decision-making for conservation managers, including locations of suitable habitat across the riverscape and the occurrence of Headwater catfish at these habitats both historically and recently. Our second objective was to measure hybridization and introgression with widely introduced Channel catfish at locations where Headwater catfish persist in Texas, the core range of the species in the United States. The process of hybridization is not a single process leading to a uniform outcome, but rather a set of processes and outcomes shaped by ecological conditions and variations in life history (Epifanio & Nielsen, 2000). The interbreeding of these two closely related species occurs by external fertilization naturally and as a result of human-related activities such as habitat degradation and introductions of Channel catfish.
Previous studies by Kelsch and Hendericks (1990) reported introgression between these two species with backcrossing of hybrids F I G U R E 1 Study area illustrating (a) the border of the United States and Mexico, (b) range of reported occurrences of Headwater Catfish (Ictalurus lupus) from Global Biodiversity Information Facility database, and (c) the range of Headwater catfish in the United States. (c) Study region (light gray), major streams (blue), focal streams where Headwater catfish have been reported (dark gray), and locations of Headwater catfish tissue collections for genetic analysis (red squares) towards Headwater catfish (Argue & Dunham, 1999). Characterizing the genetic integrity of extant populations will help to identify potentially non-introgressed refuge populations that can be included in restoration planning.

| Study area
We defined the spatial extent of our study based on level IV ecoregions of the conterminous United States (Omernik & Griffith, 2014) and the locations of Headwater catfish occurrences ( Figure 1). The Rio Grande, or Rio Bravo del Norte, is located in the south-western United States and northern Mexico (Figure 1a). It is a water supply source for agriculture, industry, municipalities and wildlife (Ward et al., 2006). The Rio Grande flows through multiple biomes, including deserts, wetlands, mountains and subtropical coastal regions. The river forms approximately 2,008 km of international border between Mexico and the United States from El Paso, TX to the Gulf of Mexico (Benke & Cushing, 2011). Other tributaries such as the Pecos River and the smaller Devils River join the Rio Grande at Amistad Reservoir in Val Verde County, Texas. The historical range of the Headwater catfish spreads east to the Edwards Plateau region in Texas, where it was once found in the upper Nueces, Frio, Guadalupe and San Saba rivers (Edwards et al., 2004;Kelsch & Hendricks, 1990) (Kelsch & Hendricks, 1986;Miller et al., 2005;Sublette et al., 1990).

| Spatial data collection
We downloaded occurrence data from the Global Biodiversity Information Facility (GBIF) to model the distribution of Headwater catfish. The GBIF is a portal that organizes digitized collection and survey data and is the largest online distributional database (Beck et al., 2014). Within the United States, records for Headwater catfish became increasingly prevalent beginning in the 1980s and 1990s because of work conducted by Kelsch andHendricks (1986, 1990) and more recent collections by McClure-Baker et al. (2010). Given the temporal nature of occurrences, we used GBIF records (GBIF, 2019a) from across the study area split into historical (1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999) and contemporary (2000-2019) time periods. Because these data represent occurrence-only data, we used the target-group absence (TGA) approach described by Mateo et al. (2010) to assign Headwater catfish absences at locations where Channel catfish (GBIF, 2019b), but not Headwater catfish, was collected. For each time period, georeferenced data were linked to medium-resolution (1:100,000-scale) National Hydrography Dataset (NHD) version 2 Plus inter-confluence stream segments (EPA, 2012) using ArcGIS 10.5.1. Occurrence data were assigned to the nearest segment within 50 m (Frimpong et al., 2005), such that any segment that was linked to a Headwater catfish record was denoted as a presence location (class = 1) and any segment that was linked to only a Channel catfish record was denoted as an absence location (class = 0). Occurrences georeferenced to locations > 50 m away from study area streams were excluded from analysis, and we only retained specimens with physical vouchers or categorized as "research-grade" observations on GBIF. Segments with both Headwater catfish and Channel catfish were denoted as presence locations (class = 1). The TGA method is commonly used in stream fish species distribution modelling implemented with presence-absence model algorithms (Huang & Frimpong, 2015;Malone et al., 2018;. We used 23 environmental predictor variables as covariates for Headwater catfish occurrence (Table 1). Covariate choice was based on previous studies linking catchment attributes to the occurrence of stream fishes (Malone et al., 2018). Covariates described anthropogenic land use, as well as hydrologic, climatic and physiographic conditions for each stream segment across the study area. Hydrologic variables included upstream watershed area (km 2 ), stream order (Strahler, 1957), stream channel maximum and minimum elevation (cm), stream channel slope (m/m), discharge (m 3 /s), water velocity (m/s) and distance to the nearest known spring outflow (km). Climatic variables included mean annual air temperature (°C) and mean annual precipitation (mm/y) for the period  obtained from EPA (2010). Physiographic variables included the major underlying geologic features and the identity of the 8-digit hydrologic unit code (HUC) in which the stream segment occurred.
Land use data were from Falcone et al. (2015) and described the area of upstream watershed covered by 11 land use classes (30-m resolution, Table 1)   thus, we used random forest (RF; Breiman, 2001) models fit to historical and contemporary periods separately. Datasets assigned to each time period suffered from class imbalance such that presence records (class = 1) were fewer compared with absence records (class = 0) in the historical (1 = 31, 0 = 154), contemporary (1 = 19, 0 = 114), and combined (1 = 45, 0 = 234) periods. We addressed class imbalance using the synthetic minority over-sampling technique (SMOTE) described by Chawla et al. (2002) and applied the "SMOTE" function from the "DMwR" package in R version 3.6.0 (R Core Team, 2019; Torgo, 2011). This process resulted in balanced numbers of occurrences for historical (1 = 93, 0 = 93), contemporary (1 = 76, 0 = 76), and combined (1 = 180, 0 = 175) periods. Finally, we checked for multicollinearity among predictor variables using the package "corrplot" and removed redundant variables with Pearson correlation absolute values > 0.70. The process resulted in the removal of stream order, maximum elevation, discharge, and air temperature from the model fitting datasets. We then fit RF models to the SMOTE-adjusted datasets using the "randomForest" function from the "randomForest" package (Liaw & Wiener, 2002) in R. The "tuneRF" function from the "rfUtilities" package (Evans & Murphy, 2018) was used to determine the number of variables to try at each split. We used model-specific tree numbers and variables tried at each split, including historical (1,200 trees, 4 variables), contemporary (250 trees, 4 variables), and combined (250 trees, 2 variables) time periods. We assessed model performance using k-fold cross validation (k = 5) in which 75% of the data were used to train the model and 25% of the data were used to test the model through the "rf.crossValidation" function and the "confusion.matrix" function from the "dismo" package. The area under the curve (AUC) of the receiver operating characteristic (ROC) is a common measure of predictive accuracy and is threshold independent . The AUC can range from 0 to 1 where values above 0.5 indicate a better than random performance. Cohens Kappa is another metric used to correct the overall accuracy of the model predictions by the accuracy expected to occur by chance. Kappa values range from −1 to +1, where +1 indicates perfect agreement and values of zero or less indicate a performance no better than random (Cohen, 1960). We also report alternative performance metrics such as sensitivity, specificity, Jaccard's similarity index, and Sørensen's similarity index (reviewed in Leory et al., 2018). Sensitivity is the proportion of observed presences correctly predicted, and specificity is the proportion of observed absences correctly predicted. Jaccard's similarity index and Sørensen's similarity index are calculated based on the confusion matrix of correctly classified occurrences (i.e., true positives, TP), correctly classified absences (i.e., true negatives, TN), incorrectly classified occurrences (i.e., false positive, FP), and incorrectly classified absences (i.e., false negatives, FN). The formula we used for Jaccard's similarity index was TP/(FN + TP+FP), the formula we used for Sørensen's similarity index was 2TP/(FN + 2TP+FP), and for both metrics values range 0-1 with values nearer to 1 representing stronger agreement between observed versus predicted occurrences. We explored relationships between Headwater catfish suitability and environmental covariates using partial dependence plots generated with the "rf.partial.prob" function from the "rfUtilities" package in R (Evans & Murphy, 2018).

| Species distribution models
Finally, we assessed differences in sampling locations and environmental conditions between the two periods to determine whether any changes in occurrence detected between periods were related to Headwater catfish distribution or simply the habitats sampled. For this assessment, we used the Kolmogorov-Smirnov test to identify differences in continuous variables between periods and plotted empirical cumulative distribution functions to illustrate patterns. For categorical variables, we used Kolmogorov-Smirnov tests and plotted proportions of categories between the two periods. We conducted these tests for habitat covariates identified by RF models as most important for predicting occurrence and report Cohen's D and P-values. Cohen's D is a common effect size measure for comparing two or more group means (Nakagawa et al., 2015). A Cohen D value of 0.2, 0.5, and 0.8 correspond to small, medium, and large effects, respectively (Cohen, 1988). The effect size indicates the magnitude of the observed effect or relationship between variables, whereas the significance test indicates the likelihood that the effect or relationship is due to chance (Maher et al., 2013). A p-value will indicate statistical significance (difference from the null distribution) but cannot reveal the size of the effect; therefore, reporting Cohen's d is complementary to the reporting of results from a test of statistical significance.

| Specimen collections
In 2018, we sampled 36 sites within the historical range of Headwater catfish to collect tissue samples for molecular analysis (Figure 1c).
These sites included the Rio Grande and tributaries that directly empty into the Rio Grande, including Cibolo Creek, Alamito Creek, Terlingua Creek, Tornillo Creek, Devils River, Dolan Creek, San Felipe Creek, Pinto Creek, Las Moras Creek, and Elm Creek. We also sampled tributaries to the Pecos River, including the Delaware River, Salt Creek, Independence Creek, and San Solomon Springs in Balmorhea State Park. Specimens were collected using seines, mini gillnets, and backpack electrofishing equipment. Catfishes were euthanized in a lethal solution of Tricaine methanesulfonate (MS-222), and tissue was obtained from the adipose fin and the right maxillary barbel and preserved in 95% non-denatured ethanol. Specimens were tagged using Floy T-bar tags with unique ID numbers and specimens from which tissues were removed were fixed in 10% formaldehyde solution and later transferred to 70% non-denatured ethanol. All Similarities among Headwater catfish and Channel catfish necessitate that detailed morphological analyses be used to identify each species. For each catfish specimen collected in 2018, we recorded anal fin ray count, standard length, pectoral spine length, caudal peduncle depth, and mouth width. Morphometrics were measured with TA B L E 1 Environmental predictor variables used for species distribution models, including parameter descriptions, sources, and variable importance measured as mean decrease in Gini for historical (1980-1999), contemporary (2000-2019), and combined (1980-2019) time periods. Land use classes of the NAWQA Wall-to-Wall Anthropogenic Land Use Trends dataset (NAWQA, U.S. Geological Survey's National Water-Quality Assessment Program, Falcone, 2015). Variables with no value (-) for mean decrease in Gini were removed prior to analysis because of multicollinearity A linear canonical discriminant function score (Kelsch, 1995) was calculated from these data with the goal of separating Headwater catfish and Channel catfish based on their morphology. Individual morphology scores (S) were calculated using the equations: where anl is anal fin ray count, psl is pectoral spine length, stdl is standard length, cpd is caudal peduncle depth, and mw is mouth width (Kelsch, 1995). The resulting score for each specimen was later paired with molecular data.
For each specimen-paired tissue sample, DNA was isolated,

| Conservation prioritization
We combined information from the contemporary SDM and genetic analyses to prioritize locations where conservation resources might

| Species distribution models
Headwater catfish SDMs differed among historical, contemporary, and combined periods. The contemporary and combined models generally had higher performance statistic values than the historical model (Table 2)  suitability among headwater streams and a general downstream increase in suitability during the contemporary period ( Figure 4d).

| Molecular markers
Mitochondrial gene sequencing and nuclear SNP analyses facilitated the identification of hybrids, misclassified individuals, and the resolution of genetic structure among populations (Appendix E; Figure 5).

| Morphology versus molecular status
Sampling during 2018 yielded 145 catfishes. The canonical discriminant function analysis of phenotype morphology classified 131 F I G U R E 2 (a) Random forest model results for historical (1980-1999; blue (Figure 6a). The majority of specimens from Balmorhea State Park had Headwater catfish phenotypes and genotypes; however, seven hybrids (RAG2 heterozygotes) were collected from the canal system during 2017 (Figure 6b). Two Channel catfish and one hybrid were collected from Cienegas Creek (Figure 6c), and a mixture of phenotypes and genotypes were collected from the Delaware River Creek (Figure 6h), three Headwater catfish and one hybrid were collected from Pinto Creek (Figure 6i) and four Headwater catfish were collected from San Felipe Creek (Figure 6j).

| Conservation prioritization
The conservation prioritization scheme identified priorities based on both habitat suitability and molecular information. The priority ranking for conservation based on combined criteria was (a) Dolan Creek

| D ISCUSS I ON
Our results identified habitats most suitable for Headwater catfish as headwater streams with high elevations, steep slopes, fast current velocities, small watershed areas, low discharges, and with little land development. Our temporal assessment of distribution showed that suitability in these habitats was generally lower in the last 20 years compared with 40 years ago. However, suitability for Headwater catfish occurrence increased in intermediate sized streams that are the preferred habitat for Channel catfish (Goldstein & Meador, 2004). When and where these species coexist, hybridization and introgression is a consequence. Texas have become isolated from larger rivers, and headwater springs that once connected pools and perennial streams have shrunk to become isolated pools or ciénegas (Hoyt, 2002). Habitat loss in the form of declining spring outflows is one of the biggest threats to the species persistence (Bean et al., 2011;Kelsch & Hendricks, 1990). With groundwater declines, flows from headwater springs decrease and, in turn, cause reductions in flow to tributaries that serve as essential habitat for Headwater catfish.
For example, Toyah Creek and the surrounding spring-fed system in the vicinity of Balmorhea was once an area of high Headwater catfish occurrence (Kelsch & Hendricks, 1990), but currently the creek is void of surface water due to the overdraft of groundwater and lowering of the water table (Sharp, 2001). Although Toyah Creek  (Saunders et al., 2002).
As water flows away from spring outflows, streams become larger, the effects of land use alterations increase, and our work suggests suitability for Headwater catfish declines.
Morphological and molecular information can be used to direct conservation actions targeting preservation of Headwater catfish. There is a high degree of morphological similarity between Headwater catfish, Channel catfish, and their hybrids. Although Kelsch (1995)  Creek, then this location could represent a candidate repatriation location (e.g., Malone et al., 2018). However, this approach can only proceed after a larger number of diagnostic loci are inspected to assure that only parental individuals are used for recovery (Allendorf et al., 2001). Populations where introgression has taken place should remain areas of conservation priority because they contain unique Headwater catfish phenotypic and even genetic variation within the hybrid genome that should not be disregarded (Demarais et al., 1992;McClure-Baker et al., 2010).
Our use of occurrence records obtained from GBIF paired with remotely sensed environmental predictor variables provided a basis for modelling temporal changes in suitable habitat for Headwater catfish. Although the resulting models provided excellent predictive power, we recognize our projections are subject to some limitations.
First, bias in the collection of species occurrence records could result in model predictions that are biased towards environments that have received more intense sampling (Araújo & Guisan, 2006). This issue confounds temporal shifts in species distribution because differences between historical and contemporary time periods could be due to either shifts in species distributions or shifts in the areas sampled. We addressed this issue by considering differences in environmental gradients sampled during each period. Our results suggested similar geographic regions and environmental gradients were sampled during each period, and thus, changes in suitability through time were likely reflective of decline in occurrence of Headwater catfish as opposed to sampling bias. Second, GBIF data represent occurrence-only records and therefore limit the use of presenceabsence modelling, a method demonstrated to be more accurate compared with presence-only modelling (Elith et al., 2006). We addressed the issue of no absence data by assigning TGA identified as sites where Channel catfish but not Headwater catfish were collected (Mateo et al., 2010). However, there is potential risk when assigning this method because of the possibility to falsely identify Headwater catfish. Species misidentification may act to contract or expand the predicted distribution of the target species and should not be neglected (Costa et al., 2015). We also point out that the degree to which competition between Channel catfish and Headwater catfish influences the range of Headwater catfish may be worth exploring for future SDMs (Araújo & Guisan, 2006). Third, SDMs should be interpreted as predictions of potential habitat that are useful as guiding information for conservation planning, but occurrences outside of the predicted areas are still possible. Recent documented occurrence of Headwater catfish in the Frio River where the species was previously believed to be extirpated (Bean et al., 2011) illustrates that the species may persist (though at low densities) in other areas of the riverscape not identified in our SDMs. In addition, low detection among commonly used gears (e.g., seines) could result in potential false negatives among sites with high probability of occurrence (Budy et al., 2015). Finally, our predictions rely on the assumption that current modelled environmental conditions are the primary drivers of Headwater catfish distributions and that these relationships will persist in the future Guisan & Thuiller, 2005). Other environmental variables might be identified and could be included in future research of Headwater catfish distribution. Significant habitat alterations in portions of the study area make Headwater catfish persistence unlikely in some locations (e.g., the now desiccated Toyah Creek), and further repeated surveys are necessary before local extinctions can be confirmed in these locations.
Systematic conservation planning informs decision-makers of the most effective and efficient ways to achieve conservation goals (Hermoso et al., 2015). Our study approach demonstrates the potential for using systematic methods for conservation planning in freshwater ecosystems to identify appropriate habitats and the most likely current distribution of Headwater catfish. The SDMs and molecular assessment provided here will aid in focusing rehabilitation and conservation efforts in priority areas (e.g., Figure 7), as well as other portions of the lower Pecos River and isolated spring habitats. Koppelman Taylor et al., 2018), and Cyprinidae (Pyke, 2008;Wilde & Echelle, 1992). For these and other species, maintaining balance between watershed management for natural resources and human usages remains a significant challenge, especially considering the future projections for water availability (Rodell et al., 2018). Even among larger municipalities in the American Southwest where there is a reasonably strong message of water conservation, the surrounding rural areas show signs of increasing agricultural production and water consumption (Edwards et al., 2002(Edwards et al., , 2004.

DATA AVA I L A B I L I T Y S TAT E M E N T
Datasets generated during the production of this study are available from the corresponding author upon reasonable request.