nena masthead
SENA Home Staff & Editors For Readers For Authors

Threshold Responses of Blackside Dace (Chrosomus cumberlandensis) and Kentucky Arrow Darter (Etheostoma spilotum) to Stream Conductivity
Nathaniel P. Hitt, Michael Floyd, Michael Compton, and Kenneth McDonald

Southeastern Naturalist, Volume 15, Issue 1 (2016): 41–60

Full-text pdf (Accessible only to subscribers.To subscribe click here.)


Access Journal Content

Open access browsing of table of contents and abstract pages. Full text pdfs available for download for subscribers.

Issue-in-Progress: Vol. 22 (2) ... early view

Current Issue: Vol. 21 (4)
SENA 21(3)

All Regular Issues


Special Issues






JSTOR logoClarivate logoWeb of science logoBioOne logo EbscoHOST logoProQuest logo

Southeastern Naturalist 41 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 22001166 SOUTHEASTERN NATURALIST Vo1l5.( 115):,4 N1–o6. 01 Threshold Responses of Blackside Dace (Chrosomus cumberlandensis) and Kentucky Arrow Darter (Etheostoma spilotum) to Stream Conductivity Nathaniel P. Hitt1,*, Michael Floyd2, Michael Compton3, and Kenneth McDonald4 Abstract - Chrosomus cumberlandensis (Blackside Dace [BSD]) and Etheostoma spilotum (Kentucky Arrow Darter [KAD]) are fish species of conservation concern due to their fragmented distributions, their low population sizes, and threats from anthropogenic stressors in the southeastern United States. We evaluated the relationship between fish abundance and stream conductivity, an index of environmental quality and potential physiological stressor. We modeled occurrence and abundance of KAD in the upper Kentucky River basin (208 samples) and BSD in the upper Cumberland River basin (294 samples) for sites sampled between 2003 and 2013. Segmented regression indicated a conductivity change-point for BSD abundance at 343 μS/cm (95% CI: 123–563 μS/cm) and for KAD abundance at 261 μS/cm (95% CI: 151–370 μS/cm). In both cases, abundances were negligible above estimated conductivity change-points. Post-hoc randomizations accounted for variance in estimated change points due to unequal sample sizes across the conductivity gradients. Boosted regression-tree analysis indicated stronger effects of conductivity than other natural and anthropogenic factors known to influence stream fishes. Boosted regression trees further indicated threshold responses of BSD and KAD occurrence to conductivity gradients in support of segmented regression results. We suggest that the observed conductivity relationship may indicate energetic limitations for insectivorous fishes due to changes in benthic macroinvertebrate community composition. Introduction The southeastern United States is a global hotspot for freshwater biodiversity (Hitt et al. 2015) including more than 660 native fish taxa (Warren et al. 2000). This region also supports the greatest number of imperiled fish taxa in North America (Jelks et al. 2008) due to small population sizes, population fragmentation, and anthropogenic stressors (Etnier 1997, Warren et al. 2000). Chrosomus cumberlandensis (Blackside Dace [BSD]) and Etheostoma spilotum (Kentucky Arrow Darter [KAD]) are species of conservation concern in this regard. Here we evaluate BSD and KAD occurrence and abundance as related to stream conductivity, an index of environmental quality and potential physiological stressor. We assess potential thresholds in the conductivity–abundance relationship and explore the relative importance of stream conductivity against other geophysical and hydrological parameters known to influence stream fishes. 1US Geological Survey, Leetown Science Center, Kearneysville, WV 25430. 2US Fish and Wildlife Service, Kentucky Ecological Services Field Office, Frankfort, KY 40601. 3Kentucky State Nature Preserves Commission, Frankfort KY 40601. 4US Fish and Wildlife Service, Tennessee Ecological Services Field Office, Cookeville, TN 38501. *Corresponding author - Manuscript Editor: Nathan Franssen Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 42 BSD is a schooling cyprinid fish endemic to the upper Cumberland River basin in Tennessee and Kentucky (Fig.1; Etnier and Starnes 1993, Starnes and Starnes 1978a) with recent expansions into Virginia. It typically inhabits forested headwater streams characterized by low turbidity, high dissolved oxygen, and low conductivity (Black et al. 2013a, Starnes and Starnes 1981). The diet of BSD switches from periphyton and detritus in summer months to benthic invertebrates in winter months (Starnes and Starnes 1981). BSD is an obligate nest-associate spawner with mound-building cyprinid fishes such as Semotilus atromaculatus (Creek Chub) and Campostoma anomalum (Central Stoneroller) (Mattingly and Black 2013, Rakes et al. 1999, Starnes and Starnes 1981). O’Bara (1990) reported apparent extirpations of BSD in 9 watersheds with historical data and observed occupancy in only 30 of 151 (20%) watersheds sampled across the upper Cumberland River basin. Black et al. (2013a) reported BSD absence in 78 of the 119 sample reaches (66%) sampled within the upper Cumberland River basin. BSD declines have been attributed to effects of surface mining and agriculture (Eisenhour and Strange 1998, O’Bara 1990, Starnes and Starnes 1978b). In 1987, BSD was listed as a threatened species under the US Endangered Species Act (ESA; US Fish and Wildlife Service [USFWS] 1987), and recovery planning efforts are on-going. KAD is a small benthic fish endemic to the upper Kentucky River drainage in eastern Kentucky (Etnier and Starnes 1993; Fig. 1). This species is largely restricted to headwater streams (1st–3rd order) where it inhabits pool or transitional areas between riffles and pools in moderate- to high-gradient streams (Etnier and Starnes 1993, Kuehne and Barbour 1983). KAD feeds on a variety of aquatic invertebrates, but adults feed predominantly on larval mayflies (Ephemeroptera), specifically Heptageniidae and Baetidae (Lotrich 1973, USFWS 2013). Its endemism, fragmented distribution, and low densities within populations combine to make this taxon inherently susceptible to anthropogenic effects and a high priority for research and conservation (Thomas 2008). KAD has undergone significant range reductions over the last several decades. Recent surveys by the Kentucky Department of Fish and Wildlife Resources and the USFWS detected KAD in only 45 of 100 historical stream sites (USFWS 2013). KAD was designated as a “species of conservation need” in Kentucky in 2005 and a candidate for listing under the ESA in 2010 (USFWS 2010). In 2015, the USFWS published proposals to list KAD as a threatened species under the ESA and to designate critical habitat (USFWS 2015a, b). The distribution and abundance of this species is thought to be limited by effects of surface mining, oil and gas development, urbanization, logging, and agriculture (Thomas 2008, USFWS 2013). Figure 1. Left panel: Chrosomus cumberlandensis (Blackside Dace). Right panel: Etheostoma spilotum (Kentucky Arrow Darter). Photographs © Dr. Matthew Thomas, Kentucky Department of Fish and Wildlife Resources. Southeastern Naturalist 43 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 In this paper, we investigate BSD and KAD relationships to stream conductivity, the ionic composition of water as measured by its ability to carry an electrical charge (Webster and Valett 2006). As an index of the ionic composition of streams, conductivity is associated with intercellular ion-transport mechanisms regulating osmoregulation and acid/base balance in freshwater organisms (Griffith 2014), but specific dose-response pathways will typically depend on the suite of anions and cations yielding a given conductivity level (Griffith et al. 2012, USEPA 2011a). High levels of stream conductivity in the eastern US are largely attributable to chlorides from road salts (Corsi et al. 2010) or sulfates from surface-mining effluent (USEPA 2011a, b). Background conductivity levels in southern Appalachian streams typically range approximately 100–150 μS/cm (i.e., 25th–75th quantiles of minimally disturbed sites; USEPA 2011a), but can exceed 1000 μS/cm in miningimpaired sites (Lindberg et al. 2011, Pond et al. 2008, USEPA 2011a). Benthic macroinvertebrate community responses to high conductivity are characterized by a loss of sensitive taxa, particularly Ephemeroptera (mayflies; Hartman et al. 2005; Kennedy et al. 2003; Lindberg et al. 2011; Merriam et al. 2011; USEPA 2011a, b), but effects of conductivity on stream fish populations are less understood. Some investigators have interpreted conductivity as an indicator of ecosystem productivity, but inconsistent relationships to fish growth and condition have been reported (Copp 2003, Dennis et al. 1995, DiCenzo et al. 1995). Dennis et al. (1995) hypothesized that energetic costs of ion balance regulation in low-conductivity waters reduced growth of Rhinichthys atratulus (Hermann) (Blacknose Dace) in Virginia, but recognized that covarying physical habitat conditions will also influence growth. Prior research has detected mining-influenced changes in fish community composition above conductivity values of 3000–3500 μS/cm (Kimmel and Argent 2010) and 600–1000 μS/cm (Hitt and Chambers 2014). However, short-term laboratory toxicity experiments with juvenile BSD did not observe mortality in reconstituted waters representative of mining-impacted and reference sites (although some sublethal histological effects were detected; J. Kunz, USGS Columbia Environmental Research Center, Columbia, MO, pers. comm.). Hopkins and Roush (2013) reported an insignificant relationship between KAD occurrence probability and upstream mining operations but lacked empirical measures of water quality in their analysis. An improved understanding of the statistical and mechanistic relationships between conductivity and fish abundance may facilitate recovery planning for BSD and KAD. In this study, we applied machine-learning and segmented regression techniques to (1) evaluate the relative importance of conductivity and other environmental predictors for BSD and KAD occurrence and (2) assess potential thresholds for conductivity effects. Given its physiological effects and habitat associations, we interpret stream conductivity as an index of environmental quality and potential physiological stressor for fishes. Methods We used data on fish abundance and water-quality parameters compiled by USFWS across the ranges of BSD and KAD (Fig. 2). The BSD dataset included Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 44 294 samples collected between 2003 and 2012 by researchers funded partially by the USFWS (Black et al. 2013a, b) and by consultants for surface-mining companies as part of ESA permitting actions by the USFWS and Office of Surface Mining. For the latter source, data were collected by “qualified biologists” as defined by USFWS (2009) using backpack electrofishing techniques during baseflow conditions. The KAD dataset included 208 samples collected between 2007 and 2013 primarily by state and federal agencies in Kentucky (Kentucky Department of Fisheries and Wildlife Resources, Kentucky State Nature Preserves Commission, USFWS) and compiled by the USFWS. For both species, fish data were collected from single-pass backpack electrofishing within 100– 300-m sample reaches including available pool, riffle, and run habitats in wadeable streams (i.e., excluding boat- and barge-electrofishing). Figure 2. Map of sample site locations (filled circles) in the upper Kentucky River basin (dark grey polygon) and upper Cumberland River basin (light grey polygon) within Kentucky and Tennessee. Southeastern Naturalist 45 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 Water-quality parameters were collected during fish sampling and included conductivity, dissolved oxygen, stream temperature, and pH (Table 1). We compiled additional covariates for analysis from the National Hydrography Dataset (NHDPlus version 2; We used ArcGIS tools to link sampling points to medium-resolution (1:100,000 scale) flow lines and to compile NHDPlus attributes representing geophysical, hydrological, and land-use data (Table 1). Geophysical attributes included elevation, slope (calculated as difference in maximum and minimum elevation divided by length for each reach), and Strahler (1957) stream order. Hydrological attributes included mean annual flow and mean annual velocity estimated for the downstream pour-point using the extended unit runoff method (McKay et al. 2012). We calculated these values from annual estimates derived for 1971–2000. We applied the gage-adjusted values for each parameter (Q0001E and V0001E, respectively). Land-use data included 2 spatial scales: local catchment (i.e., “allocated” data in NHDPlus) and watershed scale (i.e., “accumulated” data in NHDPlus). We combined classes of National Land Cover Database (NLCD) codes to represent generalized categories of forest cover, agriculture, and developed land at both spatial scales (Table 1). We defined “developed land” from NLCD classifications of urbanization and “barren land” from unreclaimed surface mines (Table 1). Table 1. Environmental variables used for boosted regression-tree analysis of Blackside Dace (BSD) and Kentucky Arrow Darter (KAD) occurrence. NHDPlus v2 = National Hydrography Dataset (http://; NCLD =National Land Cover Database 2011 ( BSD KAD Category/variable (units) Range Mean (sd) Range Mean (sd) Data source Geophysical Elevation (m) 236–641 393 (62.4) 194–543 289 (56.9) NHDPlus v2 Slope (m/m) 0–0.14 0.02 (0.02) 0–0.08 0.02 (0.01) NHDPlus v2 Strahler stream order 1–3 2 (1) 1–4 2 (1) NHDPlus v2 Watershed area (km2) 0.6–128.7 20.0 (26.3) 0.6–203.9 17.6 (26.1) NHDPlus v2 Hydrological Mean annual flow (ft3/s) 0.5–105.0 17.0 (22.9) 0.3–127.3 10.8 (16.2) NHDPlus v2 Mean annual velocity (f/s) 0.49–1.41 0.96 (0.16) 0.61–1.29 0.92 (0.10) NHDPlus v2 Land use Agriculture - local (%) 0–38.1 9.2 (8.2) 0–53.6 11.5 (10.8) NLCD (71, 81, 82) Agriculture - watershed (%) 0–38.0 8.0 (6.2) 0–53.6 11.6 (9.2) NLCD (71, 81, 82) Developed - local (%) 0–26.2 4.8 (3.4) 0–57.1 6.8 (5.8) NLCD (21, 22, 23, 24, 31) Developed - watershed (%) 0–16.7 3.7 (2.5) 0–16.5 5.4 (3.0) NLCD (21, 22, 23, 24, 31) Forest - local (%) 49.4–100 81.8 (13.6) 9.5–100 79.5 (15.1) NLCD (41, 42, 43) Forest - watershed (%) 50.9–100 85.4 (8.7) 34.0–99.5 80.7 (12.2) NLCD (41, 42, 43) Water quality Conductivity (μS/cm) 13–1100 271 (236) 28–3677 443 (570) This survey Dissolved oxygen (mg/L) 0.3–20.8 8.5 (2.3) 2.1–10.1 7.1 (1.7) This survey pH 5.5–9.1 7.7 (0.7) 6.5–8.5 7.4 (0.5) This survey Stream temperature (°C) 14.7–30.3 19.9 (2.4) 9.9–26.2 16.6 (4.0) This survey Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 46 We used boosted regression trees (BRT) to evaluate the relative importance of environmental predictors for BSD and KAD occurrence. BRT analysis is a machine-learning technique such that it extends simple classification and regression- tree analysis by “boosting” results to many models and solving for optimal values across models (De’ath 2007, Elith et al. 2008). This analytical method has several advantages over general linear modeling: it can accommodate continuous and categorical data types, controls for intercorrelation among predictors, and can reveal non-linear relationships with response variables (Elith et al. 2008). BRT applications are therefore becoming more common in ecological studies (e.g., Hopkins and Roush 2013, Leathwick et al. 2006, Merovich et al. 2013). For all BRT models, we specified a Gaussian-response tree-complexity level of 3 (i.e., enabling interactive effects), learning rate of 0.001 (i.e., specifying the minimum increment of model improvement with additional trees), and bag fraction of 0.5 (i.e., specifying an equal proportion of data held out for subsampling procedure) as recommended by Elith et al. (2008) for ecological analysis. Following Merovich et al. (2013), we used 10-fold cross-validation to assess BRT model performance and calculated “deviance explained” as the difference between the total deviance (null deviance) and residual deviance divided by the total deviance. We interpreted “variable importance” as the number of times a variable was selected for splitting the dataset, weighted by the improvement to the model from each split, and averaged across all trees in the final model (Friedman and Meulman 2003). We implemented boosted regression in R using package ‘gbm’ with R scripts provided by Elith et al. (2008). We evaluated the relationship between conductivity and abundance of BSD and KAD using segmented linear regression techniques. This method estimated the change-point in conductivity minimizing residual variation in 2 non-overlapping linear regression fits (Bacon and Watts 1971, Muggeo and Adelfino 2011, Toms and Lesperance 2003). We calculated 95% confidence intervals for estimated change-points and piecewise model slopes from 1000 bootstrap samples using the R package ‘segmented’ (Muggeo 2008). We then implemented a randomization procedure to assess the potential effect of the underlying sample-environment distribution on the estimated change-points. Prior research has shown that the number of samples collected along an environmental gradient (i.e., sample–environment distribution) may influence the performance of threshold-detection techniques (Daily et al. 2012). Randomizations accounted for this potential source of error by splitting the dataset at conductivity levels ranging 250–500 μS/cm (by increments of 50 μS/cm), randomizing draws from the high-sample size subset to equal the low-sample size subset, estimating the change-point, and repeating the process for 1000 random draws. R code for the randomization procedure is provided in Appendix A. We evaluated variance in the randomized change-point estimates as an index of the potential effect of the underlying sample-environment distribution on threshold detection. Results BSD abundance ranged from 0 to 153 individuals with a mean of 11.7 and standard deviation of 26.0 individuals (294 samples). KAD abundance ranged from 0 Southeastern Naturalist 47 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 to 11 individuals with a mean of 0.6 and standard deviation of 1.6 individuals (208 samples). Sample sites focused on wadeable streams with mean watershed areas of 20 km2 (BSD region) and 17.6 km2 (KAD region) (Table 1). Land use was dominated by forest cover in both study areas and spatial scales (>79.5%; Table 1) but also included 8–12% agricultural development and 4–7% developed land (Table 1). Mean water-quality parameters were similar between BSD and KAD regions, but maximum conductivity values were greater in the KAD region than the BSD region (3677 μS/cm and 1100 μS/cm, respectively; Table 1). In contrast, observed maximum stream temperatures were somewhat greater in the BSD region than the KAD region (30.3 °C and 26.2 °C, respectively; Table 1). The final BRT model of BSD occurrence explained 75.7% of the total deviance and 51% of the mean cross-validated deviance (Table 2). Conductivity, stream temperature, slope, and watershed area were the 4 most important variables in explaining BSD occurrence (Table 3, Fig. 3). BSD occurrence probability decreased as conductivity values increased from the minimum value (13 μS/cm) to approximately 350 μS/cm, above which BSD occurrence probability remained diminished. Low stream temperatures (i.e., below approximately 18 ºC) were associated with high BSD occurrence probabilities (Fig. 3). Slope exhibited a unimodal relationship to BSD occurrence such that the highest occurrence probabilities were associated with mid-range slope values (Fig. 3). BSD occurrence probabilities were greatest in the smallest watersheds (i.e., less than approximately 20 km2; Fig. 3). Land-use variables had less of an influence on BSD occurrence than empirical measures of water quality (conductivity and temperature; Table 3). The final BRT model of KAD occurrence explained 53.1% of the total deviance and 26.7% of the mean cross-validated deviance (Table 2). Conductivity was the strongest predictor of KAD occurrence, with a variable-importance score (31%) more than 2.5-times that of the next-most important variable (watershed forest cover) in the final model (Table 3). Conductivity exhibited a non-linear relation to KAD occurrence probability such that occurrence probabilities decreased from the minimum conductivity value (28 μS/cm) to approximately 300 μS/cm, above which low occurrence probabilities were consistently observed. Watershed-level Table 2. Model performance statistics for boosted regression-tree models of Blackside Dace and Kentucky Arrow Darter occurrence from environmental variables (see Table 1). Cross-validation (X-val) statistics indicate the mean and standard error (SE) from 10 randomized partitions of the data. The percent deviance explained was calculated as ([total deviance - residual deviance]/total deviance)*100. Statistic Blackside Dace Kentucky Arrow Darter Number of trees 6550 2550 Mean residual deviance 0.060 0.082 X-val residual deviance 0.121 0.128 X-val residual deviance SE 0.008 0.013 Total mean deviance 0.246 0.175 % deviance explained 75.7 53.1 % X-val deviance explained 51.0 26.7 Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 48 Figure 3. Boosted regression partial dependence plots for the 8 most influential variables for presence/absence of Blackside Dace. Variableimportance scores are provided in parentheses, and variables are presented in decreasing rank order from left to right by row. Importance scores for all predictor variables are provided in Table 3. Positive values on the y-axis are associated with species presence and negative values are associated with species absence. Southeastern Naturalist 49 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 land-use patterns were relatively important in the model, with KAD occurrence probabilities increasing where forest land cover exceeded approximately 90% and developed land cover was less than approximately 4% (Fig. 4). Stream temperature showed a non-linear relation to KAD occurrence probability with a rapid transition to low occurrence probabilities as temperature increased from approximately 17 ºC to 19 ºC (Fig. 4). Water-quality variables were more important than geophysical or hydrological variables for BRT models in both species. Conductivity ranked first in variable importance in both cases, and both species showed nonlinear responses in this regard. However, stream temperature was relatively more important for BSD occurrence than for KAD occurrence (Table 3). Both species exhibited nonlinear responses to stream temperature and dissolved oxygen concentrations, but the location of effect thresholds was similar for stream temperature (i.e., less than 20 ºC), whereas KAD occurrence responded to dissolved oxygen change over a narrower range than for BSD (i.e., 5–10 mg/L for BSD and 7–8 mg/L for KAD) (Figs. 3, 4). Land-use variables generally were more important for KAD occurrence than for BSD, whereas hydrological variables and stream order were relatively unimportant in both cases (Table 3). Plots of BSD and KAD abundance along the observed conductivity gradients indicated nonlinear relationships whereby either low or high abundances were observed in low-conductivity streams, but only low abundances were observed in high-conductivity streams (Table 4, Fig. 5). Segmented linear regression analyses indicated a change-point for BSD abundance at 343 μS/cm (95% CI: 123–563 μS/ cm; model R2 = 0.08) and for KAD abundance at 261 μS/cm (95% CI: 151–370 Table 3. Boosted regression tree variable importance scores for models of Blackside Dace and Kentucky Arrow Darter occurrence. Variable summary statistics are presented in Table 1. Model performance statistics are presented in Table 2. Blackside Dace Kentucky Arrow Darter Variable Relative importance Variable Relative importance Conductivity 15.86 Conductivity 30.96 Stream temperature 15.59 Forest-watershed 11.53 Slope 10.35 Developed - watershed 9.08 Watershed area 10.07 Stream temperature 7.31 Dissolved oxygen 6.11 Dissolved oxygen 4.70 Developed - local 5.61 Developed - local 4.53 Agriculture - watershed 5.03 Watershed area 4.47 Developed - watershed 4.88 Elevation 4.33 Mean annual velocity 4.70 Mean annual flow 3.98 Agriculture - local 4.51 Slope 3.74 pH 4.44 Agriculture - local 3.60 Elevation 3.91 Agriculture - watershed 3.35 Forest - local 3.36 pH 3.12 Forest - watershed 2.66 Mean annual velocity 3.11 Mean annual flow 2.44 Forest - local 2.01 Strahler stream order 0.48 Strahler stream order 0.19 Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 50 Figure 4. Boosted regression partial dependence plots for the 8 most influential variables for presence/absence of Kentucky Arrow Darter. Variable-importance scores are provided in parentheses, and variables are presented in decreasing rank order from left to right by row. Importance scores for all predictor variables are provided in Table 3. Positive values on the y-axis are associated with species presence and negative values are associated with species absence. Southeastern Naturalist 51 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 Table 4. Segmented regression results for models relating fish abundance to stream conductivity. Confidence intervals (CI) were calculated from 1000 bootstrap sa mples. Model parameter Statistic Blackside Dace Kentucky Arrow Darter β0: y-intercept Estimate 23.9 2.0 β1: slope segment 1 Estimate -0.058 -0.007 Lower 95% CI -0.094 -0.011 Upper 95% CI -0.022 -0.002 α: change-point Estimate 343.5 260.8 Lower 95% CI 123.4 151.3 Upper 95% CI 563.6 370.3 β2: slope segment 2 Estimate -0.005 -0.0001 Lower 95% CI -0.033 -0.0005 Upper 95% CI 0.023 0.0003 Figure 5. Segmented linear regression analysis for abundance of Blackside Dace (A) and Kentucky Arrow Darter (B) in relation to stream conductivity. Solid vertical lines indicate modeled change-points (343 μS/cm for Blackside Dace; 261 μS/cm for Kentucky Arrow Darter). Dashed vertical lines show 95% confidence intervals for modeled change-points. Plotted conductivity range for Kentucky Arrow Darter is truncated to 1000 μS/cm to facilitate comparison with Blackside Dace. Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 52 μS/cm; model R2 = 0.12) (Fig. 5). For both species, bootstrapped-model slopes below the estimated change-points were less than 0 (i.e., decreasing abundance with increasing conductivity), but bootstrapped-model slopes encompassed 0 at conductivity levels above the estimated change-points (Table 4) where fish abundances were negligible. We found similar estimated change-points and confidence intervals from log-transformations of fish abundances (results n ot shown). The KAD dataset supported nearly equal numbers of samples above and below the estimated change-point (101 and 107, respectively), but the BSD dataset supported more samples below the estimated change-point than above it (204 and 90, respectively). Post-hoc randomizations based on rarified datasets (i.e., standardizing sample sizes above and below candidate thresholds) indicated that estimated change-points from the full datasets for both species were within the 95% confidence intervals of estimated change-points from randomized draws of equivalent sample size (Fig. 6). Variation in randomized change-point estimates for KAD was less than for BSD due to the lower intrinsic variation in KAD abundances (Fig. 5). Discussion Our results indicate that stream conductivity is an important predictor of imperiled fish occurrence and abundance. Conductivity was more important for BSD and KAD occurrence than other natural and anthropogenic factors known to influence stream fish distributions in Appalachia (e.g., watershed area; Hitt and Roberts 2012). The consistency of the estimated conductivity change-points between species is noteworthy given autecological differences in local abundance (high for BSD and low for KAD), schooling behavior (high for BSD and low for KAD), habitat-use preferences (pelagic for BSD and benthic for KAD), Figure 6. Randomization results evaluating the sample-environment distribution effects on modeled change-points (segmented linear regression) for Blackside Dace (A) and Kentucky Arrow Darter (B). Analysis of Kentucky Arrow Darter at a threshold of 250 μS/cm was not conducted because there were equal sample sizes above and below this value. Southeastern Naturalist 53 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 and feeding behaviors (algivore/detritivore/invertivore for BSD and obligate invertivore for KAD). The estimated conductivity change-points for BSD and KAD were within 50 μS/cm of the conductivity benchmark identified by USEPA (2011a) to be protective of 95% of the benthic macroinvertebrate taxa typical of Appalachian headwater streams (300 μS/cm). The estimated conductivity change-point for BSD (343 μS/cm) was higher than a previously reported conductivity threshold for BSD occurrence of 240 μS/cm (Black et al. 2013b). This difference may be due to several factors: (1) we evaluated changes in fish abundance, whereas Black et al. (2013b) evaluated changes in fish occurrence; (2) the prior analysis assumed a logistic response in conductivity at 240 μS/cm and therefore was not designed to assess continuous variation in response to conductivity gradients; (3) their analysis incorporated detection probabilities and ours did not; (4) our analysis included more recent population samples which may have been influenced by adaptation to local stressors, as reported for other fishes (Fisher and Oleksiak 2007, Wirgin et al. 2011). Nonetheless, our results are consistent with Black et al.’s (2013b) main conclusion that conductivity and stream temperature are primary drivers of BSD occurrence. In addition, our results support O’Bara’s (1990) observation that BSD tend to occur where stream temperatures do not exceed 23 ºC, as well as Hopkins and Roush’s (2013) conclusion that forest cover and developed land are important predictors of KAD occurrence (positive and negative effects, respectively; Figs. 3, 4). We suggest that conductivity may influence BSD and KAD by regulating the quality and quantity of benthic macroinvertebrate prey available as energy sources. Although the total benthic macroinvertebrate biomass typically does not diminish in response to high conductivity in Appalachian streams (Hartman et al. 2005, Johnson et al. 2013), macroinvertebrate assemblages consistently lose mayfly taxa (Ephemeroptera; Chambers and Messinger 2001, Hartman et al. 2005, Kennedy et al. 2003, Merriam et al. 2011, Pond 2010, Pond et al. 2008), probably due to physiological limitations for exporting metabolic wastes across gill membranes (McCulloch et al. 1993). Stream fishes are known to exhibit selectivity among benthic macroinvertebrate prey (Worischka et al. 2015), so changes in benthic macroinvertebrate community structure could affect the quality and quantity of prey for fishes. BSD would not be subject to this energetic limitation during summer months (detritus and periphyton diet) but could be during winter months when their diet shifts to benthic macroinvertebrates (Starnes and Starnes 1978a). Moreover, KAD are obligate invertivores that feed primarily on mayflies (Lotrich 1973) and thus could be strongly influenced by changes in benthic macroinvertebrate community composition. In support of this hypothesis, Hitt and Chambers (2014) found that obligate invertivorous fishes declined more than species with other feeding strategies in high-conductivity streams. We recommend new research to test the energetic-limitation hypothesis through experimental manipulations. Our BRT analysis provided reasonable models for interpretation of the relative importance of environmental predictors for fish occurrence. Although the BSD Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 54 model provided a better fit to the data than the KAD model (explaining 75.7% and 53.1% of total deviance, respectively; Table 2), prior studies have interpreted BRT models with similar performance. Merovich et al. (2013) interpreted BRT models that explained 70–74% of the total deviance in water quality as a function of land-use variables in Appalachian streams. Hopkins and Roush (2013) interpreted BRT models that explained 52.5% of the total deviance in KAD occurrence in the upper Kentucky River basin, similar to our results. Differences in model performance between BSD and KAD may be explained by the fact that BSD was more common among sites than KAD (44% and 23% site occupancy, respectively), and therefore the BSD model had more opportunities for environmental variables to explain variation in occurrence. Although Merovich et al. (2013) did not consider interactive effects among predictor variables (i.e., tree complexity = 1), Hopkins and Roush (2013) accounted for interactive effects as in our approach (i.e., tree complexity = 3). We accounted for a possible statistical bias in threshold detection due to the underlying distribution of samples across the conductivity gradient (Daily et al. 2012). As is common in ecological datasets, the BSD and KAD datasets contained an unequal number of observations across the environmental gradients of interest. Daily et al. (2012) showed that such unequal distribution might affect various statistical methods for threshold detection. Our randomization procedure accounted for the potential influence of the underlying sample-environment distribution through bootstrapped resampling of rarefied subsets of the data with equal sample sizes. We detected a greater potential influence of this effect on the estimated change-point for BSD than for KAD (Fig. 6). For example, at a candidate threshold of 350 μS/ cm, the median for randomized change-points (controlling for sample size) for BSD occurrence was ~500 μS/cm (~150 μS/cm above the empirical change-point) whereas the median for randomized change points for KAD was nearly equivalent to the empirical estimate (Fig. 6). This result was probably because greater variation in BSD abundance among sites yielded greater variation in change-point estimates from random resampling of the data. Natural-resource managers may wish to account for this source of uncertainty in their establishing management thresholds for BSD based on their relative tolerance for type I and type II errors. We agree with Buhl-Mortenesen (1996) that the consequences of failing to detect an effect (type II error) should usually outweigh the consequences of falsely detecting an effect (type I error) for rare species conservation. Changes in stream conductivity may covary with changes in physical habitat and other attributes of water quality that can influence fish abundance and occurrence. For example, Hartman et al. (2005) reported increases in fine sediments downstream from headwater mining operations in association with elevated conductivity. Partitioning effects of physical habitat from water quality was not possible in this case because physical habitat data were not collected for comparison across sites. However, our analysis included measures of basin size and stream gradient, which are important surrogates for local physical-habitat conditions (Frissell et al. 1986), and these variables were less important than Southeastern Naturalist 55 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 conductivity for both species in our analysis (Table 3). Hitt and Chambers (2014) also detected changes in stream fish community composition across a conductivity gradient in southern West Virginia that could not be attributed to substrate composition, mesohabitat structure, or woody cover. However, water-quality attributes such as dissolved and particulate selenium concentrations may also co-vary with conductivity (Hitt and Chambers 2014, USEPA 2011b) and could affect stream fish community composition (Young et al. 2010), but such data were unavailable for the present study. Unmeasured physical-habitat and water-quality parameters are probably important for limiting BSD and KAD abundances below the conductivity thresholds we observed. Sampling limitations associated with high-conductivity water may have influenced our results to some degree. Electrofishing efficiency typically decreases as stream conductivity increases (Bohlin et al. 1989, Hill and Willis 1994, but see Scholten 2003), and the fish data analyzed here were collected with electrofishing techniques. However, it is unlikely that this effect could explain our main inferences because decreased sampling efficiency would be more likely to affect abundance than occurrence, yet we see similar patterns. Moreover, the inherently low densities of KAD (Thomas 2008) and high potential densities of BSD (Black et al. 2013a) would tend to cause decreased sampling efficiency to bias inferences on KAD more than BSD, but we detected consistent effects of conductivity for both species. We recommend a sampling experiment using rotenone to quantify sampling efficiency for fishes in high-conductivity streams. Ideally, such an experiment would also include multiple electrofishing wave forms and voltage levels to help optimize electrofishing applications in high-conductivity streams (e.g., Bohlin et al. 1989). Our analysis indicates that stream conductivity can provide important information for conservation and management of imperiled fishes in Appalachia. The relationship is characterized by threshold responses of fish abundance and occurrence to increasing conductivity gradients. Applications of the change-points identified here could assist recovery planning for BSD and KAD in conjunction with spatial stream-network models to route potential downstream exposure levels (Johnson et al. 2010). We recommend new research to test the hypothesis that conductivity-induced changes in benthic macroinvertebrate community composition affect stream fish abundance and community composition through energetic pathways. Acknowledgments We thank H. Mattingly (TTU), T. Black (TTU), J. Detar (PFBC), M. Thomas (KDFWR), and S. Brandt (KDFWR) for providing critical data for this study. We also thank R. Ford (USFWS), D. Smith (USGS), E. Snook (USGS), J. Young (USGS), S. Faulkner (USGS), C. Ingersoll (USGS), A. Velasco (USFWS), and E. Brett (KSNPC) for administrative and analytical support. Funding for US Geological Survey research (N. Hitt) was provided by the USFWS Science Support Partnership. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the US Government. The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the US Fish and Wildlife Service. Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 56 Literature Cited Bacon, D.W., and D.G. Watts. 1971. Estimating the transition between two intersecting straight lines. Biometrika 58:525–534. Black, T.R., J. Detar, and H.T. Mattingly. 2013a. Population densities of the threatened Blackside Dace, Chrosomus cumberlandensis, in Kentucky and Tennessee. Southeastern Naturalist 12(Special Issue 4):6–26. Black, T.R., B.K. Jones, and H.T. Mattingly. 2013b. Development and validation of habitat models for the threatened Blackside Dace, Chrosomus cumberlandensis, at two spatial scales. Southeastern Naturalist 12(Special Issue 4):27–48. Bohlin, T., S. Hamrin, T.G. Heggberget, G. Rasmussen, and S.J. Saltveit. 1989. Electrofishing- theory and practice with special emphasis on salmonids. Hydrobiologia 173:9–43. Buhl-Mortensen, L. 1996. Type-II statistical errors in environmental science and the precautionary principle. Marine Pollution Bulletin 32:528–531. Chambers, D.B., and T. Messinger. 2001. Benthic invertebrate communities and their responses to selected environmental factors in the Kanawha River basin, West Virginia, Virginia, and North Carolina. Water-Resources Investigations Report 01-4021. US Geological Survey, Charleston, WV. 61 pp. Copp, G.H. 2003. Is fish condition correlated with water conductivity? Journal of Fish Biology 63:263–266. Corsi, S.R., D.J. Graczyk, S.W. Geis, N.L. Booth, and K.D. Richards. 2010. A fresh look at road salt: Aquatic toxicity and water-quality impacts on local, regional, and national scales. Environmental Science and Technology 44:7376–7382. Daily, J.P., N.P. Hitt, D.R. Smith, and C.D. Snyder. 2012. Experimental and environmental factors affect spurious detection of ecological thresholds. Ecology 93:17–23. De’ath, G. 2007. Boosted trees for ecological modeling and prediction. Ecology 88:243–251 Dennis, T.E., S.E. MacAvoy, M.G. Steg, and A.J. Bulger. 1995. The association of waterchemistry variables and fish condition in streams of Shenandoah National Park (USA). Water, Air and Soil Pollution 85:365–370. DiCenzo, V.J., M.J. Maceina, and W.C. Reeves. 1995. Factors related to growth and condition of the Alabama subspecies of Spotted Bass in reservoirs. North American Journal of Fisheries Management 15:794–798. Eisenhour, D.J., and R.M. Strange. 1998. Threatened fishes of the world: Phoxinus cumberlandensis Starnes & Starnes. Environmental Biology of Fishes 51:140. Elith, J., J.R. Leathwick, and T. Hastie. 2008. A working guide to boosted regression trees. Journal of Animal Ecology 77: 802–813. Etnier, D.A, 1997. Jeopardized Southeastern freshwater fishes: A search for causes. Pp. 87– 104, In G.W. Benz, and D.E. Collins (Eds.). Aquatic Fauna in Peril: The Southeastern Perspective. Southeast Aquatic Research Institute Special Publication 1. Lenz Design and Communications, Decatur, GA. 553 pp. Etnier, D.A., and W.C. Starnes. 1993. The Fishes of Tennessee. University of Tennessee Press, Knoxville, TN. 681 pp. Fisher, M.A., and M.F. Oleksiak. 2007. Convergence and divergence in gene expression among natural populations exposed to pollution. BMC Genomics 8:108. Friedman, J.H., and J.J. Meulman. 2003. Multiple additive regression trees with application in epidemiology. Statistics in Medicine 22:1365–1381. Frissell, C.A., W.J. Liss, C.E. Warren, and M.D. Hurley. 1986. A hierarchical framework for stream habitat classification: Viewing streams in a watershed context. Environmental Management 10:199–214. Southeastern Naturalist 57 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 Griffith, M.B. 2014. Natural variation and current reference for specific conductivity and major ions in wadeable streams of the coterminous USA. Freshwater Science 33:1–17. Griffith, M.B., S.B. Norton, L.C. Alexander, A.I. Pollard, and S.D. LeDuc. 2012. The effects of mountaintop mines and valley fills on the physiochemical quality of stream ecosystems in the central Appalachians: A review. Science of the Total Environment 417/418:1–12. Hartman, K.J., M.D. Kaller, J.W Howell, and J.A. Sweka. 2005. How much do valley fills influence headwater streams? Hydrobiologia 532:91–102. Hill, T.D., and D.W. Willis. 1994. Influence of water conductivity on pulsed AC and pulsed DC electrofishing catch rates for Largemouth Bass. North American Journal of Fisheries Management 14:202−207. Hitt, N.P., and D.B. Chambers. 2014. Temporal changes in taxonomic and functional diversity of fish assemblages downstream from mountaintop mining. Freshwater Science 33:915–926. Hitt, N.P., and J.R. Roberts. 2012. Hierarchical spatial structure of stream-fish colonization and extinction. Oikos 121:127–137. Hitt, N.P., L.K. Bonneau, K.V. Jayachandran, and M.P. Marchetti. 2015. Freshwater ecosystems and biodiversity. Lessons in Conservation 5:5–16. Hopkins, R.L., and J.C. Roush. 2013. Effects of mountaintop mining on fish distributions in central Appalachia. Ecology of Freshwater Fish 22:578−586. Jelks, H.L., S.J. Walsh, N.M. Burkhead, S. Contreras-Balderas, E. Díaz-Pardo, D.A. Hendrickson, J. Lyons, N.E. Mandrak, F. McCormick, J.S. Nelson, S.P. Platania, B.A. Porter, C.B. Renaud, J.J. Schmitter-Soto, E.B. Taylor, and M.L. Warren Jr. 2008. Conservation status of imperiled North American freshwater and diadromous fishes. Fisheries 33:372–407. Johnson, B.R., A. Haas, and K.M. Fritz. 2010. Use of spatially explicit physicochemical data to measure downstream impacts of headwater stream disturbance. Water Resources Research 46:W09526. Johnson, B., K. Fritz, and R. Price. 2013. Estimating benthic secondary production from aquatic insect emergence in streams affected by mountaintop removal coal mining, West Virginia USA. Fundamental and Applied Limnology 182:191–204. Kennedy, A.J., D.S. Cherry, and R.J. Currie. 2003. Field and laboratory assessment of a coal-processing effluent in the Leading Creek Watershed, Meigs County, Ohio. Archives of Environmental Contamination and Toxicology 44:324–331. Kimmel, W.G., and D.G. Argent. 2010. Stream fish community responses to a gradient of specific conductance. Water, Air, and Soil Pollution 206:49–56. Kuehne, R.A., and R.W. Barbour. 1983. The American Darters. The University of Kentucky Press, Lexington, KY. 177 pp. Leathwick, J.R., J. Elith, M.P. Francis, T. Hastie, and P. Taylor. 2006. Variation in demersal fish species richness in the oceans surrounding New Zealand: An analysis using boosted regression trees. Marine Ecology Progress Series 321:267–281. Lindberg, T.T., E.S. Bernhardt, R. Bier, A.M. Helton, R.B. Merola, A. Vengosh, and R.T. Di Giulio. 2011. Cumulative impacts of mountaintop mining on an Appalachian watershed. Proceedings of the National Academy of Sciences 108:20929–20934. Lotrich, V.A. 1973. Growth, production, and community composition of fishes inhabiting a first- second-, and third-order stream of eastern Kentucky. Ecological Monographs 43:377–397. Mattingly, H.T., and T.R. Black. 2013. Nest association and reproductive microhabitat of the threatened Blackside Dace (Chrosomus cumberlandensis). Southeastern Naturalist 12(Special Issue 4):49−63. Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 58 McCulloch, W.L., W.L. Goodfellow, and J.A. Black. 1993. Characterization, identification, and confirmation of total dissolved solids as effluent toxicants. Environmental Toxicology and Risk Assessment 2:213−227. McKay, L., T. Bondelid, T. Dewald, J. Johnston, R. Moore, and A. Rea. 2012. NHDPlus Version 2: User Guide. US Environmental Protection Agency. Washington, DC. 182 pp. Merovich, G.T., J.T. Petty, M.P. Stanger, and J.B. Fulton. 2013. Hierarchical classification of stream condition: A house–neighborhood framework for establishing conservation priorities in complex riverscapes. Freshwater Science 32:874−89 1. Merriam, E.R., J.T. Petty, G.T. Merovich, J.B. Fulton, and M.P. Strager. 2011. Additive effects of mining and residential development on stream conditions in a central Appalachian watershed. Journal of the North American Benthological Society 30:399−418. Muggeo, V.M.R. 2008. Segmented: An R package to fit regression models with broken-line relationships. R News 8:20−26. Muggeo, V.M.R., and G. Adelfino. 2011. Efficient change-point detection in genomic sequences of continuous measurements. Bioinformatics 27:161–166. O’Bara, C.J. 1990. Distribution and ecology of the Blackside Dace, Phoxinus cumberlandensis (Osteichthyes: Cyprinidae). Brimleyana 16:9–15. Pond, G. 2010. Patterns of Ephemeroptera taxa loss in Appalachian headwater streams (Kentucky, USA). Hydrobiologia 641:185–120. Pond, G.J., M.E. Passmore, F.A. Borsuk, L. Reynolds, and C.J. Rose. 2008. Downstream effects of mountaintop coal mining: Comparing biological conditions using family- and genus-level macroinvertebrate bioassessment tools. Journal of the North American Benthological Society 27:717−737. Rakes, P.L., J.R. Rakes, and P.W. Shute. 1999. Reproductive behavior, captive breeding, and restoration ecology of endangered fishes. Environmental Biology of Fishes 55:31–42. Scholten, M. 2003. Efficiency of point-abundance sampling by electro-fishing modified for short fishes. Journal of Applied Ichthyology 19:265–277. Starnes, L.B., and W.C. Starnes. 1981. Biology of the Blackside Dace (Phoxinus cumberlandensis). American Midland Naturalist 106:360–371. Starnes, W.C., and L.B. Starnes. 1978a. A new cyprinid of the genus Phoxinus endemic to the upper Cumberland River drainage. Copeia 1978:508–516. Starnes, W.C., and L.B. Starnes. 1978b. Status report on a new and threatened species of Phoxinus from the upper Cumberland drainage. Proceedings of the Southeastern Fishes Council 2:1–3. Strahler, A.N. 1957. Quantitative analysis of watershed geomorphology. Transactions of the American Geophysical Union 38:913–920. Thomas, M. 2008. Distribution and Status of the Kentucky Arrow Darter. Section I of the Interim Annual Performance Report for State and Tribal Wildlife Grant: T-9-RSI-1. Ichthyofauna Resources of Kentucky. Fisheries Division, Kentucky Department of Fish and Wildlife Resources. Frankfort, KY. Toms, J.D., and M.L. Lesperance. 2003. Piecewise regression: A tool for identifying ecological thresholds. Ecology 84:2034–2041. US Environmental Protection Agency (USEPA). 2011a. A field-based aquatic life benchmark for conductivity in central Appalachian streams (Final Report). EPA/600/R- 10/023F. Washington, DC. USEPA. 2011b. The effects of mountaintop mines and valley fills on aquatic ecosystems of the central Appalachian coalfields. EPA/600/R-09/138F. Washington, DC. US Fish and Wildlife Service (USFWS). 1987. Endangered and threatened wildlife and plants; determination of threatened species status for the Blackside Dace. Federal Register 52:22580–22585. Southeastern Naturalist 59 N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 USFWS. 2009. Coal mining in Tennessee: Minimum guidelines for the development of protection and enhancement plans for Blackside Dace (Phoxinus cumberlandensis). USFWS Tennessee Ecological Services Field Office, Cookville, TN. 20 March 2009. 34 pp. USFWS. 2010. Endangered and threatened wildlife and plants; review of native species that are candidates for listing as endangered or threatened. Federal Register 75(217): 69222–69294. USFWS. 2013. Endangered and threatened wildlife and plants; review of native species that are candidates for listing as endangered or threatened; annual notice of findings on resubmitted petitions; annual description of progress on listing actions. Federal Register 78:70103–70162. USFWS. 2015a. Endangered and threatened wildlife and plants; threatened species status for Kentucky Arrow Darter; proposed rule. Federal Register 80:60962–60988. USFWS. 2015b. Endangered and threatened wildlife and plants; designation of critical habitat for Kentucky Arrow Darter; proposed rule. Federal Register 80:61030–61081. Warren, M.L., Jr., B.M. Burr, S.J. Walsh, H.L. Bart, Jr., R.C. Cashner, D.A. Etnier, B.J. Freeman, B.R. Kuhajda, R.L. Mayden, H.W. Robison, S.T. Ross, and W.C. Starnes. 2000. Diversity, distribution, and conservation status of the native freshwater fishes of the southern United States. Fisheries 25:7–29. Webster, J.R., and H.M. Valett. 2006. Solute dynamics. Pp. 169–186, In F.R. Hauer, and G.A. Lamberti (Eds.). Methods in Stream Ecology. Elsevier London, UK. 877 pp. Wirgin, I., N.K. Roy, M. Loftus, R.C. Chambers, D.G. Franks, and M.E. Hahn. 2011. Mechanistic basis of resistance to PCBs in Atlantic Tomcod from the Hudson River. Science 331:1322–1325. Worischka, S., S.I. Schmidt, C. Hellmann, and C. Winkelmann. 2015. Selective predation by benthivorous fish on stream macroinvertebrates: The role of prey traits and prey abundance. Limnologica 52:41–50. Young, T.F., F. Keith, W. Adams, J. Besser, W.A. Hopkins, D. Jolley, E. McNaugton, T.S. Presser, D.P. Shaw, and J. Unrine. 2010. What you need to know about selenium. Pp. 7–46, In Chapman, P.M., W.J. Adams, M.L. Brooks, C.G. Delos, S.N. Luoma, W.A. Maher, H.M. Ohlendorf, T.S. Presser, and D.P. Shaw (Eds.). Ecological Assessment of Selenium in the Aquatic Environment. Society of Environmental Toxicology and Chemistry, CRC Press Boca Raton, FL. 339 pp. Southeastern Naturalist N.P. Hitt, M. Floyd, M. Compton, and K. McDonald 2016 Vol. 15, No. 1 60 Appendix A. R code for randomization procedure to evaluate potential effects of the sample-environment distribution on estimated change-points. This code requires an input file “BSD.csv” and produces an output matrix R object “outbsd”. # Sample-environment distribution randomization example for Blackside Dace # by N.P. Hitt <> condthresh < - seq(250,500,50) # Set candidate thresholds for analysis nreps < - 1000 # Set number of replicates for analysis outbsd < - matrix(ncol=length(condthresh),nrow=nreps) # Initialize output matrix colnames(outbsd) < - condthresh # Label columns for output matrix bsd < - read.csv(BSD.csv, header=T, sep=“,”) # Read dataset in .csv format with rows as # sites, one column for conductivity values # (labeled COND), and one column for # abundance values (labeled ABUND) for(i in 1:length(condthresh)){ # Start loop over thresholds bsd_hvals < - subset(bsd, COND>condthresh[i]) # Subset above-threshold data bsd_lvals < - subset(bsd, COND< condthresh[i]+1) # Subset below-threshold data for(j in 1:nreps){ # Start loop over replicates bsd_lvalsamp < - bsd_lvals[sample(nrow(bsd_lvals), nrow(bsd_hvals)), ] # Take random selection of below-threshold # data to equal sample size for above- # threshold data bsd2 < - rbind(bsd_lvalsamp, bsd_hvals) # Compile rarefied dataset bsd3 < - lm(ABUND ~ COND, data=bsd2) bsd4 < - segmented(bsd3, seg.Z=~COND) # Estimate change-point using segmented # linear regression outbsd[j,i] < - bsd4$psi[2] # Write change-point to output matrix } } boxplot(outbsd)