Occupancy and Detection of Clinch Dace Using Two Gear Types

In 1999, the first populations of a Chrosomus dace were discovered in Tazewell County, Virginia (Skelton 2007; Figure 1). This form is most similar to the Laurel Dace Chrosomus saylori (Skelton 2001) and has been referred to as the Clinch Dace Chrosomus sp. cf. saylori (Figure S2), which remains undescribed, although life-history, morphometric, and meristic data support its classification as a distinct species (White and Orth 2013, 2014). During further sampling from 1999 to 2012, Clinch Dace C. sp. cf. saylori were captured in 17 streams of 8 larger tributaries in the upper Clinch Basin (Skelton 2007; White and Orth 2014). These assessments suggested that Clinch Dace occupy headwater streams with gravel substrate and forested catchments (White and Orth 2014). As a result of its narrow distribution and potential threats to its persistence, the Clinch Dace is considered a tier I species (very high conservation need) in the Virginia Wildlife Action Plan (Virginia Department of Game and Inland Fisheries 2015), “endangered” by Jelks et al. (2008), and “critically imperiled” by NatureServe (2016), but is not “threatened” or endangered pursuant to the U.S. Endangered Species Act (ESA 1973, as amended) nor has it been formally assessed or petitioned for inclusion.

Monitoring fish population responses to anthropogenic alterations in headwater streams is challenging. Approximately 125 km of unsampled second- and third-order streams remained within the Clinch Dace's presumed range (White 2012). Sampling protocols for rare species must address the risk of mortality or sublethal stress upon individuals. For example, the discovery of the Clinch Dace's sister species, the Laurel Dace, was first recognized from a rotenone survey collection housed in a museum (Skelton 2001). Although additional populations of Laurel Dace were subsequently found in nearby tributaries, the rotenone survey had possibly caused the elimination of the population at its type locality (George et al. 2015). Past surveys for Clinch Dace have used seining (Skelton 2007) and backpack electrofishing (White and Orth 2014). Each of these methods has drawbacks. Seining is often difficult in these streams because of rocky substrate and dense, thorny, riparian vegetation. Pulsed direct-current (DC) backpack electrofishing has generally been proven safe for small-bodied fish, but it requires operation of the sampling device within specific parameters (Holliman et al. 2003a, 2003b; Whaley et al. 1978). Safety precautions include maintaining safe distances with the electrodes and setting voltage and frequency of the unit to the minimum immobilization threshold of the target species. Minnow trapping, a passive gear for small fish, may be less intrusive, although one study documented significant self-inflicted injuries to Greenside Darters Etheostoma blenniodes captured in metal traps (Cooke et al. 1998).

The effects of specific stream habitat alterations that may be the result of anthropogenic activities have also not yet been defined for the Clinch Dace and a better understanding of these threats and its current distribution will inform management actions to conserve the species. At the local scale, beaver Castor canadensis ponds that favor piscivores and competitors (Compton et al. 2013), livestock grazing in riparian zones (Kauffman and Krueger 1984), population fragmentation from impassible road crossings (Warren and Pardew 1998), as well as discharge of chemicals and other household or sewage wastes (Cook et al. 2015) are potential threats to Clinch Dace. At larger scales, alterations to land cover primarily occur during timber harvest or surface mining for bituminous coal. The clearing of forested land cover can alter sediment distributions, flow, temperature, and water quality (Swift and Messer 1971; Arthur et al. 1998). In some surface mining operations, the topography and the deposition of mining overburden material in adjacent valleys can have lasting physical, chemical, and biotic impacts upon a watershed (Griffith et al. 2012). These impacts can include alterations to hydrologic regimes, eroded stream banks, deposition of fine sediments in stream channels, and decreased water quality, including increased conductivity (Wolman 1967; Hartman et al. 2005; Schlief and Mutz 2005; Freund and Petty 2007; Pond et al. 2008; Fox 2009; Daniels et al. 2010; Griffith et al. 2012; Hitt and Chambers 2014), which have resulted in the long-trem impairment of fish and macroinvertebrate communities in portions of Appalachia (Lemly 1997; Teh et al. 2004; Bernhardt and Palmer 2011; Cormier et al. 2013; Arnold et al. 2014; Hitt and Chambers 2014; Timpano et al. 2015).

The risk of not detecting a rare or cryptic species when it is present at a site during distribution and habitat association studies has contributed to the popularity of occupancy models, which correct presence–absence data to account for incomplete detection of the focal species (MacKenzie et al. 2002). The formula for naïve occupancy (ψ_n) without accounting for species detection probability (p) is simply the number of sites where the species was present (x) divided by the total number of sites sampled (s) (MacKenzie et al. 2002):

Naïve occupancy may underestimate the number of sites where a rare species is present, leading to false inferences about its status as well as environmental covariates associated with its presence. Traditional occupancy models produce estimates of two parameters: ψ, or the probability that a species is present at a site (occupancy probability), and p, the probability that a species is detected in a single survey given that it is present at a site (detection probability). Consequently, by accounting for site- and survey-specific covariates, many occupancy models have been developed to assist in the management of imperiled stream fishes (Albanese et al. 2007, 2010; Wenger et al. 2010; Anderson et al. 2012; Dextrase et al. 2014; Kuehne and Olden 2016;). Previous surveys for the Clinch Dace did not attempt to estimate detectability, creating ambiguity and perhaps biased estimates of occupancy (MacKenzie et al. 2002).

In the present study, we seek to provide information about the status of Clinch Dace and the habitat factors that may help explain its current distribution within an occupancy modeling framework. To accomplish this we developed the following three specific research objectives: 1) search for undocumented populations of Clinch Dace, 2) examine detection probability of Clinch Dace using backpack electrofishing and minnow trapping, and 3) investigate relationships between reach and landscape-level habitat conditions and Clinch Dace occupancy. By examining these objectives we can further efforts to monitor and conserve populations of Clinch Dace.

Previous broad-scale surveys in the Clinch Basin and adjacent watersheds found Clinch Dace at few sites, which limited the ability to infer its distribution and habitat requirements. To maximize the probability that we would detect Clinch Dace and increase our odds of discovering new populations, we narrowed our focus to include streams within 15 south-flowing tributary systems to the Clinch River in Russell and Tazewell counties, Virginia (White and Orth 2014; Figure 2). The region comprising the northern half of these two counties narrowly bounds the only records of Clinch Dace in Virginia and most streams are located in the Central Appalachian Plateau ecoregion (Woods et al. 1999; Figure 2). We selected a total of 70 stream segments using a stratified random-sampling design with higher weight for the selection of Strahler second- and third-order streams (the orders in which Clinch Dace are most likely to occur). We set the stream segment boundaries as the nodes of stream intersections in the National Hydrography Dataset stream layer. Placement of the sampling reach within each selected segment was contingent on site access. We selected 46 second-order, 15 third-order, and 9 fourth-order streams and sampled 48 sites in 2014 and 22 sites in 2015.

Land cover in the catchments above sample sites was on average 75% forest, 13% agriculture, 8% developed, 3% grassland, and 1% barren. Pasture and crop cultivation are most common in the Thompson Creek and North Fork Clinch watersheds, which were the only sites we sampled in the Ridge and Valley ecoregion. Narrow strips of residential and small-scale agriculture exist in some valleys. Some of the grassland is reclaimed mined land that has not supported forest regeneration. Active or abandoned mines are also sometimes classified as barren. Only two of our field sites, Middle Fork Clinch River and Coal Creek (both heavily mined watersheds), had barren land cover > 5% in their catchments. A few sampled streams have undergone channel reconstruction following mining (Northington et al. 2011). Sites permitted for coal mining by the Virginia Department of Mines, Minerals, and Energy currently exist in 11 of 15 watersheds with historical records of Clinch Dace.

We tailored our sampling design to an occupancy modeling approach (MacKenzie et al. 2002; Nichols et al. 2008). The standardized sampling protocol that we used at all 70 sites consisted of four 50-m subreaches, for a total of 200 m of sampled length. We marked and georeferenced the boundaries of each subreach. We randomly assigned minnow trapping at two of the four subreaches and used backpack electrofishing at the other two. During the primary sampling period, we placed six Promar^® cloth-mesh minnow traps in pool or run habitat deep enough to submerge the downstream opening in two of the 50-m subreaches. We baited each trap with half of a slice of bread and a handful of dry dog food, as baited traps had higher catch rates for a similar species, Blackside Dace (Chrosomus cumberlandensis; Detar and Mattingly 2013). We oriented traps with the long axis parallel to the direction of flow. Traps were usually retrieved after 3 to 4 hours and were never set overnight. At the other two subreaches, a two-person crew conducted single-pass backpack electrofishing with pulsed DC current using a Smith-Root LR-24 backpack electrofishing unit. Both crew members netted fish. We recorded the abundance of each species in the sample and measured total length (mm) for all Clinch Dace that we captured. We repeated the sampling procedure with the same gears allocated to the same subreaches on the following day.

To characterize in-stream habitat conditions and catchment land use, we assembled a database of 11 coarse- and fine-scale habitat variables for all 70 sites sampled. We assessed stream substrate using a modified Wolman pebble count method classifying the size of 100 pebbles at a site (Wolman 1954). We measured particle diameter along the β axis on a Wentworth scale using a gravelometer (U.S. Forest Service Stream Systems Technology Center 1996). We combined all particles < 2 mm into a single category representing fine sediments, namely sands, silts, and clays. Additionally, the presence of fine sediments at each site was captured using measures of embeddedness. While conducting pebble counts, we measured the embeddedness of each substrate particle as the percentage of the vertical extent of a particle that was embedded in fine sediment (Bunte and Abt 2001) within one of five categories: < 5%, 5–25%, 25–50%, 50–75%, and > 75%.

All field-collected habitat data were measured at the site scale. Ten equally spaced habitat transects were completed at each 200-m sampling site. First, we measured percent canopy cover using a spherical densitometer (Forestry Suppliers; convex model A) while standing in the center of the stream channel. We also measured wetted stream width and maximum depth along each transect. We recorded water-quality parameters at the lowermost and uppermost points of each site, including pH using a Hanna Phep 5 pH meter and specific conductance (conductivity) using a Hanna HI98311 DiST^® 5 EC/TDS or YSI EcoSense EC300A meter.

We used geographic information system software to calculate landscape-level habitat variables. We delineated catchment areas draining to each survey site with the 1 arc-second (30-m) National Elevation Dataset (U.S. Geological Survey 2002) using the watershed tool in the spatial analyst package of ArcMap version 10.1. We calculated land cover of each watershed by overlaying the catchment polygons with the 2011 National Land Cover Database (Homer et al. 2015). We estimated stream gradient using the change in elevation of the digital elevation model divided by the length of the stream segment in which the survey site occurred. Elevation was calculated using the digital elevation model at the downstream end of the sampling reach. We calculated the average proportion of each catchment with layers of three periods of historical mining: legacy mines, prelaw (The Surface Mining Control and Reclamation Act of 1977), and active. We scaled all habitat data by subtracting the mean value and dividing by the standard error to make the data symmetrical around zero.

Four assumptions must be met by single-season occupancy models: 1) occupancy status at each site does not change over the sampling period (i.e., the population is closed), 2) the probability of occupancy is constant across sites, or modeled using covariates, 3) probability of detection is constant across all surveys or is a function of covariates, and 4) detection of species and detection histories are independent among locations (Mackenzie et al. 2006). To satisfy the closure assumption, we minimized the time between events by sampling on consecutive days. We allowed occupancy probability to vary among sites by incorporating site covariates in our models. We addressed assumption 3 by using a sampling covariate.

We used multimodel inference to analyze our records of Clinch Dace presence at the 70 sites with repeat sampling data. We selected multiscale occupancy models in program PRESENCE version 7.5 (U.S. Geological Survey, Patuxent, Maryland) to test the impact of gear type on detection probability (Nichols et al. 2008). These models contained parameters for site-scale occupancy (ψ), detection probability (p), and subreach-scale occupancy (θ; Table S1). Each 50-m subreach was considered a unique detection/nondetection event, such that over 2 days each site had eight possible detection events.

Site covariates were used to model ψ (habitat variables, x). A sampling covariate (gear type, y) was coded for each of the eight sampling events to indicate which subreaches had been sampled using minnow trapping and which had been sampled with electrofishing to allow p to vary between our two sampling gears. In all models, θ was coded to be equal on both days and conditional on large-scale occupancy. All 50-m subreaches within sites had similar habitat and were expected to have the same probability of occupancy. To validate this assumption, we tested models where θ was allowed to vary among subreaches against models where θ was constant across subreaches. The Akaike information criterion corrected for a small sample size (AICc) and log likelihood values indicated that the constant θ parameterization better fit the data. We first determined the top model for p and then applied this in the model-fitting exercises for ψ.

We developed 17 specific a priori hypotheses about the habitat factors that drive Clinch Dace occupancy containing from one to three additive site covariates and organized under broader model categories (see covariate justification and descriptions in Table 1). The model categories and nested hypotheses were as follows: landscape (elevation , gradient, catchment area, gradient + elevation, gradient + catchment area, elevation + catchment area, and gradient + elevation + catchment area), water quality (conductivity, pH, conductivity + pH), channel morphology (width : depth, embeddedness, width : depth + embeddedness), catchment land cover (all forest, evergreen forest), resource extraction (historical mining in catchment), and riparian (canopy cover). Multicollinearity among covariates included in the same model can lead to false inferences in their effects on occupancy. Correlation between site covariates included in the same models was not a problem in our data set (Pearson correlation coefficients all < 0.7). First we determined that our best detection model gear type as a covariate for detection as ΔAIC was 3.47 relative to the model that did not include this covariate. Thereafter, all occupancy models included gear type as a covariate for detection. Our final model set included a total of 18 models.

We ranked the 18 models (or hypotheses including null model) using an information theoretical approach based on minimum AICc. We tested for goodness of fit with our null model using a Pearson chi-square test with 1,000 bootstraps. Ĉ is a measure of overdispersion and should be ≤ 1, as larger values indicate a lack of fit. We determined that model fit was adequate (Ĉ = 0.90). Models with ΔAICc ≤ 2 are considered to have relatively similar support, whereas models with ΔAICc ≥ 9–11 have very little support relative to the top model (Burnham et al. 2011). AIC weights are the relative likelihood that a given model is the best among the models in the candidate set and is calculated for each model with the following equation: ⁠. We present model-averaged parameter estimates based on the upper 95% confidence sets for the standard coefficients of ψ and p using all models with weights > 0.10 (Anderson 2008).

Overall predicted occupancy at the 70 sites corrected for incomplete detection was calculated by summing the occupancy probabilities conditional on detection history for each site and then dividing by the number of sites (70). Using the gear-specific estimates of detection probability, we developed a cumulative detection curve to determine how many 50-m spatial replicates would be needed for each gear to reach a desired detection probability. The equation for this was

where p is the detection probability per sampling event, and x is the number of sampling events.

Models with a gear-specific sampling covariate always had a higher AICc weight relative to the corresponding model that lacked the gear-specific sampling covariate and confidence intervals on regression coefficients did not overlap 0 (Table 2). Thus all future models contained the gear covariate for p. Backpack electrofishing was 55% more effective at detecting Clinch Dace given that the species is present at a site. Detection probability was 0.65 (95% CI = 0.49–0.78) for electrofishing and 0.42 (95% CI = 0.28–0.56) for minnow trapping. Cumulative detection probability curves indicate that an estimated 95% probability of detection given presence was reached within just three 50-m replicates using backpack electrofishing. Minnow trapping took twice as many, six 50-m replicates, to reach the 95% detection probability threshold (Figure 3).

We detected Clinch Dace at 13 of 70 sites (naïve ψ = 0.19). The conditional occupancy estimate corrected for incomplete detection using the top-ranked model did not change (13.19 sites) because of the high detection probability for electrofishing and moderate detection probability for minnow trapping and the survey effort. Subreach occupancy probability given presence at the site, θ, was estimated at 0.91. We discovered new records for Clinch Dace in three streams: Lewis Creek, Middle Creek, and Zeke Creek. Yet we failed to detect Clinch Dace in two streams, Town Hill Creek and Jackson Fork, where populations had previously been detected.

Contrary to expectations, the channel morphology hypothesis, driven by a strong positive relationship with substrate embeddedness, best explained Clinch Dace occupancy (Table 2). The standardized regression coefficient for substrate embeddedness in the top-ranked model was 0.85 (95% CI = 0.24–1.47), which yields strong support that Clinch Dace occupancy probability was higher at sites with more fine sediment (Figure 4). The AIC weight for this model was 0.42, meaning that it had a conditional probability of 0.42 of being the best model. This is a high probability relative to other models in the candidate set and is not higher because of the large number of models in the candidate set whose weights must sum to 1. The model with the proportion of the catchment that was forested as a covariate of occupancy under the landscape hypothesis also received support and Clinch Dace were more likely to be found in highly forested catchments (β = 0.95, 95% CI = 0.051–1.86; Figure 4). All other hypotheses were poorly supported by the data, with model weights less than 0.06.

Results from analysis of presence–absence data provide three major insights into the management of the imperiled Clinch Dace: 1) gear choice is a crucial component of monitoring plans for Clinch Dace, 2) Clinch Dace populations may persist in somewhat silty or sandy habitats, and 3) Clinch Dace populations may be sensitive to conversion of forested catchments from logging or surface mining.

Both backpack electrofishing and minnow trapping may be valuable sampling techniques in future monitoring protocols. The higher detection probabilities associated with backpack electrofishing indicate that this sampling method is superior in streams with low densities of Clinch Dace. Minnow traps could be used in multisession demographic assessments to estimate survival, population size, or movement (Detar and Mattingly 2013), where individual Clinch Dace would receive unique marks with elastomer tags. Differences in detection probability between sampling gears has also been reported in other studies. Albanese et al. (2010) found that detection probability of Wounded Darter Etheostoma vulneratum was higher with electrofishing compared with snorkel surveys, whereas the opposite was true for Tangerine Darter Percina aurantiaca. Biologists may also find that minnow traps are the only option to capture Clinch Dace in nonwadeable reaches that have been impounded by beavers. We recommend that single-pass electrofishing monitoring of Clinch Dace populations cover at least 150 m of stream habitat. Minnow trapping of at least 300 m of habitat or an increase in the density of minnow traps will improve detection probability.

The Clinch Dace's vulnerability to minnow traps underscores the threat that local fishermen who harvest bait pose to populations. Virginia outlawed bait harvest in six streams containing Tennessee Dace, and special regulations on harvest also may be warranted for Clinch Dace (Virginia Department of Game and Inland Fisheries 2016). Bait regulations often differ on a state-by-state basis, and can be confusing to anglers (Meronek et al. 1995). As such, place-based trapping bans may be ineffective and poorly enforced.

Correcting for incomplete detection did not make a difference in our estimates of occupancy. Models suggested that survey effort was most likely sufficient to detect Clinch Dace at sites if they were present. This provides strong support that Clinch Dace are extirpated from portions of Town Hill Creek and Jackson Fork where they were previously found but were not detected in this survey. It also confirms that populations of Clinch Dace are rare and highly fragmented in the region, occurring at < 20% of tributary stream segments within the study area.

The positive relationship between substrate embeddedness and Clinch Dace occupancy in the models under the channel morphology hypothesis was unexpected, but may be attributable to multiple factors: First, substrate embeddedness may be related to stream gradient. Although not supported by the current analysis, lower stream gradient has been correlated to Blackside Dace presence (Black et al. 2013). These lower-gradient streams contain more pool habitat, which is preferred by Chrosomus daces. The slow flow in streams with more pool habitat allows fine sediments to accumulate, especially in catchments that have been logged and roaded over many decades. This may also be a reason forest cover and substrate embeddedness were not more strongly negatively correlated. Some higher-gradient mountain streams may have sufficient power to flush fine sediments that erode from the landscape, thereby preventing aggradation of fine sediments (Hartman et al. 2005; Pond et al. 2008). Second, nest building by Stonerollers Campostoma anomalum and Creek Chubs Semotilus atromaculatus may provide sufficient quantities of clean substrate for Chrosomus dace to successfully reproduce even in degraded habitats (Johnston 1999; Peoples et al. 2011; Mattingly and Black 2013). The two instances when we observed spawning aggregations of Clinch Dace occurred on gravel nests in pool tails with abundant fine sand in the surrounding pools (Video S4); however, nests still may be smothered with fine sediments if a high flow event occurs during or after deposition of gametes. Third, siltation may be an artifact of the spatial position of these sampling points. Three of the sites with the highest substrate embeddedness, Laurel Fork and two sites on Greasy Creek, were located within network distances of 11 km of one another and are all within the Indian Creek drainage. Beaver activity is high in Laurel Fork, which has also led to the trapping of substantial amounts of fine sediments. Densities of Clinch Dace were low at all three of these sites (Moore et al. 2017). Overall, Clinch Dace may be able to tolerate some silt or sand; however, assuming that increased deposition of fine sands and silts does not negatively affect their populations may lead to management decisions that harm Clinch Dace populations if high levels of fine sediments are actually a stressor.

Strong model performance in the catchment land-cover hypothesis set corroborates the positive association between Clinch Dace and forested land cover from White and Orth (2014). Most of the area that is not forested within the study area has been cleared for surface mining, timber harvest, or pasture. Large forested tracts in catchments reduce sediment inputs, as well as fluctuations in water clarity, temperature, and hydrologic regimes. Forest could also indicate the absence of recent mining or residential development that will directly lead to degradation of water quality.

It is unclear why riparian forest cover (canopy cover) received less support as a driver of Clinch Dace distribution. Landscape ecologists have wrestled with the influence of scale and land cover in riverine management (Jones et al. 1999; Rabeni and Sowa 2002; Allan 2004; Frimpong et al. 2005). Previous studies suggest that altered riparian corridors are more detrimental to benthic-dwelling or -feeding species than water-column drift- or surface-feeding species such as Clinch Dace (Jones et al. 1999). Few benthic fishes were syntopic with Clinch Dace. Although Fantail Darters Etheostoma flabellare commonly co-occur with Clinch Dace (12/13 sites), we rarely found sculpin Cottus spp. alongside Clinch Dace (1/13 sites). Regardless of scale, intact forests appear important for maintaining Clinch Dace populations. Future research is needed to describe riparian vegetation condition in more detail such as noting the dominant riparian tree communities at sites as well as forest stand age in the riparian buffer.

Conductivity was elevated at 64 of our 70 study sites over the reference conditions for Appalachian streams of < 133 μS/cm (Pond et al. 2008). However, only nine sites (one of which contained Clinch Dace) had conductivity > 500 μS/cm. The lack of heterogeneity in conductivity across sites may explain the absence of a stronger relationship between conductivity and Clinch Dace occupancy in this analysis. As freshwaters in Central Appalachia become saltier, scientists are recommending location-specific total dissolved solids criteria to protect sensitive species (Cañedo-Argüelles et al. 2016). The threshold of conductivity where biological impairment occurs may be 300–500 μS/cm (Pond et al. 2008; Bernhardt et al. 2012; EPA 2011). Elevated conductivity may indirectly affect fish by causing declines in populations of their macroinvertebrate prey. Sensitive taxa may or may not be important components of Clinch Dace diets. In 63 Clinch Dace stomachs, White (2012) was able to identify macroinvertebrates from only five orders: Megaloptera, Coleoptera, Diptera, Ixodida, and Hymenoptera. Ephemeroptera, Plecoptera, and Trichoptera, often considered the macroinvertebrate orders most indicative of water quality, were not identified. Others have hypothesized that Blackside Dace and the Kentucky Arrow Darter Etheostoma spilotum exhibit a nonlinear threshold response to elevated conductivity as the energetic demands for foraging increase as macroinvertebrate populations decline (Hitt et al. 2016). To determine the specific biological impacts of ionic concentrations requires experimental manipulations, whereas most studies have been observational in nature (Freund and Petty 2007). Such in situ manipulations would be inappropriate for Clinch Dace, but propagation may support toxicity tests in the future.

Other unsupported covariates of Clinch Dace occupancy included pH, watershed area, historical mining, water temperature, and transparency. The pH values of the streams in our study area did not vary much, and were all basic (> 7 and < 9.5). The geology of this portion of the Appalachian coal seam in southwest Virginia does not produce acid mine drainage like deposits farther north do. Chrosomus dace are headwater specialists (Skelton 2007; White and Orth 2014) and although we detected a few individuals in larger streams (Mudlick and Pine creeks), they were likely strays from populations upstream. Negative impacts of historical mining on the current occupancy of Clinch Dace were not strongly supported by our analysis. Water-quality recovery follows a negative quadratic pattern after valley fill creation, with conductivity reaching a peak before declining to < 500 μS/cm an average of 19.6 years later (Evans et al. 2014). Water quality in some streams may have recovered after past mining disturbances to the point where it would now be possible to translocate individuals to establish new populations where natural recolonization cannot occur. Other variables such as water temperature and transparency depended on the time of collection. In future studies the deployment of temperature loggers will help to better document temperature regimes in these streams.

There may be other worthwhile analytical approaches to deal with complex habitat associations of Clinch Dace. Occupancy models are essentially a form of logistic regression analysis (when log-link function is used) with added parameters to account for detection probability. The assumptions of logistic regression may constrain our ability to detect complex marginal, nonlinear, and threshold relationships of individual habitat covariates with the presence of the Clinch Dace. Another worthwhile approach may be hierarchical Bayesian threshold models (e.g., Wagner and Midway 2014).

Trade-offs often exist between detection probability and immediate stress to individuals. Although backpack electrofishing produced the highest detection probabilities and rarely produced noticeable injury to fish in our study, it may be a more intrusive sampling gear than minnow trapping. Further development of technologies such as environmental DNA (eDNA) sampling (Goldberg et al. 2011; Jerde et al. 2011; Minamoto et al. 2012; Olson et al. 2012; Mahon et al. 2013; Jane et al. 2014) may maximize detection probability and eliminate handling of fish. Forest management and surface mine reclamation to facilitate forest recovery after disturbance at the catchment scale may be important for rare fishes in the region. Conductivity may become an important water-quality parameter to monitor and regulate to promote the persistence of insectivorous fishes as surface mining in the Central Appalachians threatens populations of their macroinvertebrate prey. Best management practices for timber harvest such as maintaining vegetated buffer strips around streams, minimizing road construction, and controlling runoff from logging roads may mitigate the effects of commercial logging operations that occur in watersheds containing rare fishes. More work is needed to test the ionic tolerances of rare fishes and their prey resources (Hitt et al. 2016). Where the threat of large-scale land-cover alteration exists, managers can compile lists of other sites with conditions that are suitable to support Clinch Dace to potentially establish new experimental populations.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Detection history, sampling gear covariate, and habitat covariates used in occupancy modeling for Clinch Dace (Chrosomus sp. cf. saylori) in tributaries to the Clinch River, Russell and Tazewell counties, Virginia. Data were collected at 70 sites during June through August of 2014 and 2015.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S1 (26 KB XLSX).

Video S1. Underwater video footage of Clinch Dace Chrosomus sp. cf. saylori prespawn behavior on a Creek Chub Semotilus atromaculatus nest in Greasy Creek, Tazewell County, Virginia during early June 2015.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S2 (6193 KB MP4).

Reference S1. Bunte K, Abt SR. 2001. Sampling surface and subsurface particle-size distributions in wadable gravel-and cobble-bed streams for analyses in sediment transport, hydraulics, and streambed monitoring. Fort Collins, Colorado: U.S. Forest Service.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S3; also available at http://www.krisweb.com/biblio/gen_usfs_bunteetal_2001_gtr74.pdf (9940 KB PDF).

Reference S2. Daniels LW, Stewart B, Zipper CE. 2010. Reclamation of coal refuse disposal areas. Virginia Cooperative Extension Publication 460-131.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S4; also available at https://www.pubs.ext.vt.edu/content/dam/pubs_ext_vt_edu/460/460-131/460-131_pdf.pdf (1344 KB PDF).

Reference S3. George AL, Kuhajda BR, Neely DA, Call G. 2015. Recovery plan for the Laurel Dace (Chrosomus saylori). Atlanta, Georgia: Southeast Region, U.S. Fish and Wildlife Service.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S5; also available at https://ecos.fws.gov/docs/recovery_plan/20161012_Laurel%20dace%20RP%20final%20.pdf (2124 KB PDF).

Reference S4. Skelton CE. 2007. Distribution and status of Blackside Dace (Phoxinus cumberlandensis) and Clinch Dace (Phoxinus sp. cf. saylori) in the Upper Clinch River system, Virginia. Blacksburg, Virginia: Virginia Department of Game and Inland Fisheries Final Report.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S6 (1266 KB PDF).

Reference S5. [EPA] U.S. Environmental Protection Agency. A field-based aquatic life benchmark for conductivity in Central Appalachian streams (Final Report). 2011. Washington, D.C.: U.S. Environmental Protection Agency, EPA/600/R-10/023F, 2011.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S7; also available at https://cfpub.epa.gov/ncea/risk/recordisplay.cfm?deid=233809 (5818 KB PDF).

Reference S6. U.S. Forest Service Stream Systems Technology Center. 1996. Gravelometers: gravel templates for pebble counting in gravel-bed streams. Stream notes: to aid in securing favorable conditions of water flows.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S8; also available at http:// http://studylib.net/doc/10550807/april-1996–286-kb—8-pages (135 KB PDF).

Reference S7. Virginia Department of Game and Inland Fisheries 2015. Virginia's 2015 wildlife action plan.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S9; also available at http://bewildvirginia.org/wildlife-action-plan/pdf/2015-Virginia-Wildlife-Action-Plan.pdf (36257 KB PDF)..

Reference S8. White SL. 2012. Distribution and life history of Chrosomus sp. cf saylori in the upper Clinch River Watershed, VA. Master's thesis. Blacksburg: Virginia Polytechnic and State University.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S10; also available at https://vtechworks.lib.vt.edu/bitstream/handle/10919/19200/White_SL_T_2012.pdf?sequence=1 (3454 KB PDF)..

Reference S9. Woods AJ, Omernik JM, Brown DD. 1999. Level III and IV ecoregions of Delaware, Maryland, Pennsylvania, Virginia, and West Virginia. Corvallis, Oregon: U.S. Environmental Protection Agency.

Found at DOI: http://dx.doi.org/10.3996/022017-JFWM-017.S11; also available at https://extension.umd.edu/sites/extension.umd.edu/files/_docs/programs/master-gardeners/Natives/1999_Woods_Omernik_reg3_ecoregion_descriptions.pdf (656 KB PDF).

This research was funded through a U.S. Fish and Wildlife Service State Wildlife Grant managed through the Virginia Department of Game and Inland Fisheries. All work was in accordance with collection permits issued by Virginia Department of Game and Inland Fisheries and protocol 13-184 FiW, approved by the Virginia Tech Institutional Animal Care and Use Committee. We particularly thank Hunter Hatcher, Skylar Wolf, and Derek Wheaton for field assistance; Gregory Anderson and Brandon Peoples for assistance with data analysis; and Michael Pinder for advice and direction. We also thank two anonymous reviewers and the editorial staff of this journal for their helpful suggestions in improving this manuscript.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.