Distribution, Occupancy, and Habitat of the Endangered Carolina Madtom: Implications for Recovery of an Endemic Stream Fish

Abstract

Endemic fish are important components of freshwater ecosystems because they contribute to biodiversity and provide vital ecological functions. The Carolina Madtom, Noturus furiosus, is a small catfish endemic to the Neuse and Tar river basins of North Carolina, USA. Three previous surveys over the past 60 years have shown declining occurrence and abundance in the basins, and as such, the species was listed as federally endangered in 2021. To provide critical information to guide Carolina Madtom conservation and recovery strategies, we surveyed 36 sites (75 locations) in both basins to (1) determine the current distribution of the Carolina Madtom, (2) develop occupancy models to estimate probability of detection and occurrence of the species throughout its range, and (3) determine its instream habitat use and suitability. We collected 59 Carolina Madtom during snorkel surveys in the Tar River basin and none from the Neuse River basin, indicating that Carolina Madtom populations are still declining in both occurrence and abundance throughout their historical and recent range, especially in the Neuse River basin. Occupancy modeling estimated low occupancy probability (0.35), while detection probability was high (0.81). Carolina Madtom occupied slow-to-moderate velocity water over sand and gravel substrate, using cobble and woody debris as cover. Habitat suitability distributions quantified the most suitable ranges of microhabitat parameters for Carolina Madtom occupancy. A comparison of available suitable habitat in the two river basins revealed that adequate suitable habitat was available in both basins, suggesting that other factors such as pollution or predation from the non-native Flathead Catfish Pylodictis olivaris, may be contributing to population declines. The application of our results will aid in management and recovery actions for the species.

1. Introduction

North America supports the greatest temperate freshwater biodiversity globally, and the southeastern United States is especially rich with freshwater fauna [1,2]. Sixty-two percent of United States fishes (493 species) occur in the southeast, and 91% (269 species) of the Nation’s freshwater mussels are found in the southeastern United States [3,4]. The southeastern United States also has the greatest number of imperiled freshwater species in North America [1]. Through anthropogenic factors such as habitat degradation, fragmentation and loss, flow modification, and pollution, many southeastern fishes are experiencing population declines [5,6,7]. Many of these imperiled fishes are endemic species and critical to freshwater ecosystems because they contribute to biodiversity and provide important ecological functions; however, many are in need of conservation and are understudied and data deficient [7,8]. Moreover, endemic species are limited in range or suitable habitat, typically exhibit small population sizes, and are at increased risk of extinction due to human-induced habitat modifications [5,9]. Thus, many endemic species remain poorly understood and minimally managed globally [10,11].

The Carolina Madtom Noturus furiosus, is a small catfish (less than 200 mm total length) endemic to the Neuse and Tar river basins of North Carolina [12]. The species occurs in free-flowing streams in riffles, runs, or pools in shallow areas [12,13]. Carolina Madtom are benthic-associated organisms, and as such, most commonly occupy areas with substrates consisting of a mixture of sand or gravel, with leaf litter and small cobble included for cover [13]. The Carolina Madtom spawning season begins in May, and reproduction may last through July [12]. The species seeks out spawning areas with low water velocity and adequate cover to nest. Males guard nests consisting of natural cavities, such as leaf litter packs, large woody debris, under small rocks, inside empty native unionid mussel shells, or in discarded beverage containers [12,13]. Artificially constructed and deployed cover units may enhance Carolina Madtom detections [14], and they were found to select such artificial cover units over natural cover items in laboratory choice tests [15].

The Carolina Madtom has experienced substantial declines in distribution since the 1960s [16,17], and are currently protected with federal endangered species status [18]. Carolina Madtom populations have been relatively stable in the Tar River basin but have declined precipitously in the Neuse River basin [19,20]. The Neuse River is a degraded river basin, and the North Carolina Department of Environmental Quality has determined that 14% of the total stream distance in the basin is impaired, according to their water quality thresholds. Impacts from urban wastewater, fertilizer, residential and industrial development, and commercial animal operations contribute to pollution and eutrophication of the system [21].

The Carolina Madtom has not been widely studied and was until recently, considered data deficient [22]. In the past 60 years, only three intensive sampling efforts have been conducted for the species. In the 1960s, Smith and Bayless performed basin-wide rotenone sampling and found that the Carolina Madtom was common in both basins [23,24]. In the 1980s, Burr et al. [12] re-sampled the Neuse and Tar river basins and found slight declines in the species’ distributions in both basins. Wood and Nichols [19,20] then sampled Carolina Madtom in 2007 and found substantial decreases in their distribution in the Neuse River, while Tar River distributions were stable, relative to those sampled in the 1980s. Population distribution research had not occurred since 2007 on the Carolina Madtom until this study in 2016–2017.

Our study was designed to provide an evaluation of the status and trends of Carolina Madtom populations and their habitat, prior to an impending endangered species status assessment and to build upon the aforementioned studies. Our specific objectives were to (1) determine the current range and distribution of the Carolina Madtom in the Neuse and Tar river basins, (2) develop occupancy models to estimate probability of detection and occurrence of the species throughout its range, and (3) determine its current instream habitat use and suitability and compare findings to previous studies.

2. Materials and Methods

2.1. Study Area

This research was conducted in the Neuse and Tar river basins of the Piedmont and Coastal Plain physiographic provinces of North Carolina, USA. The Neuse River flows approximately 325 km through North Carolina from its headwaters originating in the Piedmont at the confluence of the Eno and Flat rivers to its mouth at Pamlico Sound near the city of New Bern [25]. The basin covers an area of 10,034 km² and spans 18 counties. Approximately 2.5 million people currently live in the Neuse River basin, and human-associated activities in the watershed adversely impact the habitat and water quality of the Neuse River [25]. Approximately 13% of the basin is considered urban, 45% forested, and 29% cropland and pastureland [26]. Non-point source pollution from a variety of intensive land uses has degraded water quality and habitats throughout the basin (e.g., urbanization, agriculture, and forestry without adequate riparian protections or proper erosion and sediment management). Commercial farming inputs, such as animal waste and fertilizers, contribute 60% of the nitrates and phosphates in the system [26]. Due to the dense human population in the basin, many municipalities have constructed dams and withdraw water from the created impoundments for human use, affecting river flow. Aquatic habitat loss, especially in the Neuse River basin, is an issue of importance, as an increasing human population results in the loss of natural areas and increases sedimentation and polluted runoff from impervious surfaces [25].

The Tar River runs through North Carolina from its origin in Person County to the town of Washington, where it becomes the Pamlico River and flows 65 more kilometers to its mouth at Pamlico Sound [27]. The basin covers 8755 km² and spans 16 counties. The Tar-Pamlico basin is more rural and less impacted by human activities than the Neuse River basin, with a human population of only 415,000. Approximately 55% percent of the basin is classified as forest and wetland, 28% cropland and pastureland, and 7% urban [28]. The primary habitat problems affecting the basin are erosion and sedimentation [26]. Future urbanization in this area of the southeastern Piedmont of North Carolina [29] is a concern for water quality and instream habitat degradation.

2.2. Study Design and Conditions

In 2016, we studied 11 sites in the Tar River basin and 9 sites in the Neuse River basin for a total of 20 sites to encompass previously extant Carolina Madtom population distributions. These sites included the main waterways of each basin, including the mainstem Tar, Fishing Creek, and Swift Creek in the Tar River basin and Contentnea Creek and Little River in the Neuse River basin. The surveyed sites included areas of historical occurrence, recent occurrence (post-2006), and exploratory areas where Carolina Madtom had not been detected. Each site was 150-m in length and followed GPS coordinates corresponding to the same 150-m site sampled by Wood and Nichols [19]. In 2017, we expanded our study site length from 150-m to 4-km to improve detection of this rare species within waterways containing a patchwork of suitable habitat. This modification was made due to low catch-rates in 2016 and the need to establish a flexible sampling protocol for long-term monitoring efforts (i.e., ability to adjust sampling sites within dynamic river systems). Surveys in 2017 included three to four 30-m locations within the expanded site length, which focused on areas with, or adjacent to, recent madtom occurrence (post-2006). Survey sites were established at areas deemed to have suitable Carolina Madtom habitat, with a random survey (30-m) added to reduce positive sampling bias. In total, we sampled twelve 4-km sites in the Tar River basin and four 4-km sites in the Neuse River basin for a total of sixteen sites consisting of fifty-five 30-m locations in 2017.

We surveyed stream sites for Carolina Madtom from May through October in 2016 and 2017 [30]. Each survey was conducted following a snorkeling protocol similar to that of Wood and Nichols [19]. Based on concerns over impaired visibility and sampling efficiency [31,32,33], our snorkel sampling was not performed when Secchi depth was ≤1 m. In 2016, surveys were conducted at the 150-m sites, with two to five snorkelers participating, and in 2017, surveys were conducted at the 30-m locations within the 4-km site, with two to four snorkelers. Each snorkeler covered no more than a 5-m wetted width to standardize effort and ensure thorough searching. If the river width (in meters) was five times greater than the total number of snorkelers, then each snorkeler sampled more than one 5-m section of the stream width. Surveys were conducted during daylight hours between 0800 and 1800. Each snorkeler started at the downstream portion of the site or location and slowly surveyed upstream, upturning any leaf litter, woody debris, cobble, or litter present in the river to maximize detection potential. Detected Carolina Madtom were promptly collected in a dip net and placed into buckets along the shore for further processing upon conclusion of the snorkel survey. We noted only three Carolina Madtom that were detected but escaped capture during surveys; all occurred in the Tar River basin and these observations were not included in our analyses. We weighed the captured fish to the nearest 0.1 g (Ohaus Scout Model SPX8200, Ohaus Corporation, Parsippany, NJ, USA) and measured total length in millimeters. A right-side pelvic fin clip was collected for genetic analysis [34,35]. The point-of-capture was designated by a weighted marker for the data collection of microhabitat use. After data collection was complete, each fish was released alive at their point-of-capture. We also recorded the presence of co-occurring native Margined Madtom Noturus insignis, two co-occurring non-native ictalurid fish species (Channel Catfish Ictalurus punctatus, and Flathead Catfish Pylodictis olivaris), and the endemic Neuse River Waterdog Necturus lewisi, a federally threatened aquatic salamander [17,36,37] at the surveyed sites. These co-occurring species may serve as bioindicators of similar habitat quality needs or offer insight into possible interactions (positive or negative) with the Carolina Madtom.

Microhabitat use data were collected during both the 2016 and 2017 seasons at all sites with Carolina Madtom occurrence. Depth (m), bottom velocity (m/s), mean-column velocity (m/s), dominant substrate composition, subdominant substrate composition, distance to nearest cover, and cover type were measured at the point-of-capture of each Carolina Madtom. Depth, bottom velocity, and mean-column velocity were measured using a top-set wading rod and a Marsh-McBirney Model 2000 flow meter. Bottom velocity was measured directly on the substrate surface and mean-column velocity was measured at 60% of the total depth from the water surface. Dominant and subdominant substrate compositions were determined by greatest and second greatest percentage of substrate type at the location, according to a modified Wentworth particle scale [38]. Distance to nearest cover was measured as the distance from point-of-capture to the nearest cover object (zero if the fish was directly associated with cover). Cover type was classified as small woody debris, large woody debris, cobble, leaf litter, artificial, and none.

Available instream microhabitat surveys were performed at all sites during 2016 and 2017 under base-flow conditions. Instream microhabitat data were collected by sampling cross-sectional transects. For the 150-m sites in 2016, 10 cross-sectional transects were spaced 15 m apart, starting from a random point within the first 15 m and moving upstream through the reach. The same transect spacing was used in 2017; however, each location encompassed two cross-sectional transects for a total of six to eight transects per location. Each transect was sampled at a minimum of 10 evenly spaced points, usually 1–3 m apart, depending on stream width. At each transect point, we measured depth, bottom velocity, mean column velocity, dominant substrate, subdominant substrate, distance to nearest cover, and cover type using the methods previously described.

2.3. Occupancy Modeling Data Analysis

Snorkel survey results from 2017 were modeled to estimate Carolina Madtom occupancy and detection probabilities using a single-species, single-season occupancy model [39] in the software program PRESENCE version 2.13.47 ([40], USGS Patuxent Wildlife Research Center, Laurel, MD, USA, https://www.mbr-pwrc.usgs.gov/software/presence.html; accessed on 5 November 2024). Given the large potential habitat range of the Carolina Madtom and the large time allotment for each sampling event, repeat visits to each individual surveying location were not logistically feasible during the field season. Instead, we implemented a spatial replication approach with each 4-km site serving as the single sampling unit relevant to occupancy and each 30-m location within the site serving as a repeat visit [41]. In total, 16 sites were snorkel surveyed, each with three to four 30-m locations (spatial replicates) within each site. This resulted in 16 occupancy sites with three to four repeat visits serving as the detection histories for each site. Prior to developing occupancy models, we assessed our 10 microhabitat covariates and placed them into the following two groups: variables expected to influence occupancy and those expected to influence detection. Using information from the most commonly used microhabitats by madtom species [42,43], we selected the following seven microhabitat covariates as most likely to influence Carolina Madtom occupancy at a site: depth, mean-column velocity, dominant substrate, subdominant substrate, distance to nearest cover; and cover type (large woody debris; cobble). We selected the following three variables likely to influence detection of Carolina Madtom: depth, mean-column velocity, and distance to nearest cover. Due to the relatively low number of sites surveyed and Carolina Madtom detected, we restricted the complexity of the fitted models by investigating models with only one covariate influencing occupancy and one influencing detection. We compiled all models and ranked them following an information-theoretic approach (Akaike information criterion corrected for small sample size, AIC_c) [44]. Individual estimates of occupancy (Ψ) and detection (p) probabilities were provided for all fitted models. Models within 10% of the AIC weight (W_i) of the top-performing model were included as confidence models, as described by Ruiz and Peterson [45].

2.4. Habitat Data Analysis

Using the cross-sectional transect and point-of-capture microhabitat data collected from sites surveyed in 2016 and 2017, we analyzed the available instream microhabitat and microhabitat use data to determine current habitat use, selectivity, and suitability distributions for Carolina Madtom. To determine Carolina Madtom microhabitat use, we calculated the mean depth, bottom velocity, and mean-column velocity measurements of occupied microhabitats and the mode of substrate and cover type. Habitat selectivity was assessed using the percentage of habitat used by Carolina Madtom compared to the available instream habitat from our surveyed sites. To determine these percentages, we developed separate frequency distributions for Carolina Madtom microhabitat use and available instream microhabitat for each variable. The two frequency distributions were then compared using a Kolmogorov–Smirnov (K-S) two-sample test for the continuous variables (depth, bottom velocity, mean column velocity, dominant substrate, subdominant substrate) and a log-likelihood ratio G-test for independence for the categorical variable (cover type), with p-values < 0.05 (α) indicating significant differences between available instream habitat and Carolina Madtom habitat use (i.e., non-random habitat use). Habitat suitability was calculated as the proportion of habitat use divided by available habitat for each habitat variable range in corresponding frequency distributions. The resulting values were standardized to a maximum of 1.0, and the range of the habitat variable with a suitability value of 1.0 was deemed the most suitable. If multiple ranges had values approximate to 1.0, the entire combined range was deemed most suitable. Habitat suitability analysis was based on fish locations from the Tar River, as no Carolina Madtom were detected in the Neuse River basin during our snorkeling surveys.

We also compared the available habitat between the Neuse and Tar river basins to determine if the drastic decrease in Neuse River basin populations may be attributed to a lack of suitable instream physical habitat. All available instream habitat measurements were pooled over both seasons separately for the Neuse and Tar river basins. Available instream habitat frequency distributions were then developed for each microhabitat variable, and the two distributions were compared using the K-S two-sample test and G-test as described previously. Using the ranges of the most suitable Carolina Madtom habitat from our calculations, we determined if there was available habitat in the suitable ranges in the Neuse River basin and compared the proportion of each basin’s available instream habitat to determine if there were differences between the amount of suitable habitat in the Neuse and Tar river basins.

3. Results

3.1. Snorkel Surveys

We captured a total of 59 Carolina Madtom in standardized snorkeling surveys during 2016–2017. During the 2016 season, we captured only 15 fish, which all occurred at only one of the 20 surveyed 150-m sites (5%). The one site with observed specimens was Swift Creek in the Tar River basin (site 9; Figure 1). No individuals were found in the Neuse River basin in 2016.

During the 2017 season, we captured 44 Carolina Madtom. They were found at five of the 16 surveyed 4-km sites (31%), or 14 of the 55 total 30-m survey locations (25%; Figure 1). Carolina Madtom were found at three sites on Fishing Creek and two sites on Swift Creek in the Tar River basin. Among the five sites where Carolina Madtom were captured, their overall mean total length was 79 mm (range 64–98 mm) and their overall mean total weight was 7.3 g (range 4.2–12.6 g; Figure 1). Although not collected during our standardized snorkeling surveys in the Neuse River basin in 2017, we captured two Carolina Madtom in the Little River while assisting state agency biologists performing freshwater mussel surveys. These two fish were not included in our analyses or capture total but reported herein as incidental collections (Figure 1) because of their importance to future conservation and management. Likewise, we report the incidental collection of six Carolina Madtom in 2017 from two sites in the Tar River basin (Figure 1) while assisting with mussel or salamander surveys. These six fish were not included in our capture total, but their data were used in the habitat suitability analysis.

While conducting snorkel surveys over the two sampling seasons, we also found 206 Margined Madtom, a native congeneric and potential competitor to the Carolina Madtom (

Table 1. Number (#) of Carolina Madtom, co-occurring native Margined Madtom, two co-occurring non-native ictalurid fish species (Channel Catfish and Flathead Catfish), and the endemic salamander, Neuse River Waterdog, observed during Carolina Madtom surveys in 2016 and 2017.

	2016		2017
Species	# Observed	# Sites	# Observed	# Sites (Locations)
Carolina Madtom	15	1	44	5 (14)
Margined Madtom	60	9	146	12 (31)
Channel Catfish	49	11	94	7 (16)
Flathead Catfish	12	1	4	1 (3)
Neuse River Waterdog	29	8	58	11 (23)

). We also observed 143 Channel Catfish, a non-native competitor and possible predator, and 16 Flathead Catfish, a voracious non-native predator [47,48,49]. Finally, we counted 87 occurrences of the Neuse River Waterdog salamander (Table 1).

3.2. Occupancy Modeling

Our naïve occupancy value was Ψ = 0.31, which is based on our survey results of five of 16 sites with Carolina Madtom occurrences. We ran a suite of 31 models and ranked them by lowest AIC_c value and highest W_i (

Table 2. The occupancy models estimating occupancy probability ($\widehat{\mathsf{\Psi }}$) and detection probability ($\widehat{p}$) for Carolina Madtom in the Tar River basin, including microhabitat covariates (in parentheses). The models presented are within 10% AIC weight (W_i) of the top performing model. Dominant = dominant substrate; mean velocity = mean-column water velocity; depth = water depth; subdominant = subdominant substrate; distance to cover = distance to nearest cover object; large woody = large woody debris as cover; cobble = cobble as cover; (.) = constant (null).

Model	AICc	ΔAICc	Wi
Ψ(dominant), p(mean velocity)	37.64	0.00	0.20
Ψ(.), p(mean velocity)	39.03	1.39	0.10
Ψ(dominant), p(.)	39.36	1.72	0.08
Ψ(dominant), p(depth)	39.98	2.34	0.06
Ψ(subdominant), p(mean velocity)	40.06	2.42	0.06
Ψ(dominant), p(distance to cover)	40.39	2.75	0.05
Ψ(.), p(.)	40.60	2.96	0.04
Ψ(.), p(distance to cover)	40.89	3.25	0.04
Ψ(depth), p(mean velocity)	40.89	3.25	0.04
Ψ(.), p(depth)	40.91	3.27	0.04
Ψ(large woody), p(mean velocity)	41.03	3.39	0.04
Ψ(cobble), p(mean velocity)	41.19	3.55	0.03
Ψ(subdominant), p(.)	41.61	3.97	0.03
Ψ(large woody), p(.)	42.14	4.50	0.02

). Because of constraints in model fitting due to small sample size, we were unable to run a fully parameterized (global) model, but considered models with one covariate for Ψ and one covariate for p. Our most plausible occupancy model was Ψ (dominant substrate) and p (mean-column velocity), with dominant substrate having a negative effect on occupancy (β = −0.09, 95% CI = −0.23–0.04) and mean-column velocity having a negative effect on detection probability (β = −33.00, 95% CI = −63.00–−3.01) (

Table 3. Top performing occupancy model and models within two ΔAIC_c of top performing model for Carolina Madtom across its extant range. Beta values and 95% confidence intervals for Ψ and p covariates are reported; dominant = dominant substrate; mean velocity = mean-column velocity. (.) indicates no covariate used for given parameter, so intercept beta and 95% CI are reported.

Model	β(Ψ)	95% CI(Ψ)	β(p)	95% CI(p)
Ψ(dominant), p(mean velocity)	−0.09	−0.23–0.04	−33.00	−63.00–−3.01
Ψ(.), p(mean velocity)	0.67	−0.43–1.76	−31.69	−63.73–0.34
Ψ(dominant), p(.)	−0.08	−0.20–0.04	1.53	0.26–2.79

). Goodness of fit (χ²) testing determined that the model adequately fit the observed data (p = 0.47) and there was no evidence of overdispersion ($\widehat{c}=0.99)$. Site-specific occupancy ranged from <0.01 (95% CI = <0.01–0.98, site 26) to 0.68 (95% CI = 0.25–0.93, site 7), and site-specific detection probability ranged from 0.07 (95% CI = <0.01–0.66, site 5) to 0.99 (95% CI = 0.72–>0.99, site 9). Because of limitations in the occupancy modeling dataset and inherent uncertainty in fish modeling [45], eight other occupancy models were selected as confidence models to help determine which microhabitat covariates influence Carolina Madtom occupancy and detection (Table 2). All confidence models had W_i greater than 10% of our selected model (i.e., ≥0.02), and covariates included in the confidence models for occupancy were subdominant substrate, depth, large woody debris, and cobble. Covariates included in confidence models for detection probability were depth and distance to nearest cover. Confidence models within two ΔAIC_c yielded similar beta values and confidence intervals for covariates (Table 3), which resulted in similar occupancy and detection probability estimates.

3.3. Microhabitat Use, Availability, and Suitability

From our collections of Carolina Madtom (59 from the standardized snorkeling surveys plus the six incidental from the Tar basin in 2017), we found that Carolina Madtom occupied areas with a mean water depth of 0.47 m, a mean bottom velocity of 0.10 m/s, and a mean-column velocity of 0.22 m/s (

Table 4. Comparison of Carolina Madtom microhabitat use between Midway et al. [13] and the current study.

	Midway et al. [13]			Current Study
Variable	Mean/Mode	95% CI	Range	Mean/Mode	95% CI	Range
Depth (m)	0.42	0.39–0.45	0.01–0.92	0.47	0.42–0.52	0.18–1.00
Bottom velocity (m/s)	0.14	0.12–0.16	0.00–0.43	0.1	0.08–0.12	−0.03–0.39
Mean-column velocity (m/s)	0.22	0.20–0.24	0.00–0.58	0.22	0.18–0.26	−0.02–0.77
Substrate	Sand	–	–	Sand	–	–
Cover type	Cobble	–	–	Cobble	–	–

). The most commonly occupied substrate was sand, and the predominant cover was cobble. All microhabitat use and availability data came from Carolina Madtom collections in the Tar River basin in our study because there were zero standardized collections in the Neuse River basin and only two incidental collections, which were not appropriate to use for this analysis.

By comparing the available instream habitat with the habitat use of Carolina Madtom, we found that the species non-randomly selected instream habitat. For all six habitat parameters that we measured, Carolina Madtom inhabited a specific subset of the total available microhabitat in the basin (K-S two-sample test, p < 0.05). A wide range of depths were available in the Tar River basin from extremely shallow to over 1 m deep. However, Carolina Madtom microhabitat use followed a distribution with peak occurrences between 0.30 and 0.49 m deep (Figure 2). The distribution for bottom velocity availability was positively skewed, with a majority of the velocity values negative (i.e., upstream flow) or near 0.00 m/s. Carolina Madtom occupied some of the same slow-moving areas; however, bottom velocity use was recorded up to 0.39 m/s, indicating their selection for slightly faster water than the slow, stagnant velocities that dominated the available microhabitat. The distribution of available mean-column velocities followed the same pattern as bottom velocities with a positively skewed distribution and low velocities prevalent. The mean-column velocity use was approximately normal in distribution, and again, Carolina Madtom were found occupying areas with slightly faster moving water than the slow, stagnant velocities that were common throughout the surveyed reaches. The dominant substrate composition was predominantly sand and silt, with lower occurrences of gravel and cobble. Carolina Madtom most frequently occupied habitats over sand with gravel used almost as frequently, and cobble occupied less frequently (Figure 2). Carolina Madtom were never observed over silt substrate, even though silt was prevalent in over 16% of the total stream area surveyed. The subdominant substrate habitat associations were similar to the dominant substrate, with sand and gravel over the majority of the available habitat, and Carolina Madtom occupying substrates of sand, gravel, and cobble, with a strong selection for gravel subdominant substrates. The available cover type in the basin was equally distributed among cobble, small woody debris, and large woody debris. Artificial habitat (i.e., beverage containers or other litter) and leaf packs were much less commonly available cover. Nearly 40% of the surveyed available habitat in the Tar River basin had little to no adequate cover for Carolina Madtom to inhabit. Carolina Madtom were found most frequently under cobble cover, although both large and small woody debris were used as well, and 20% of our captures came from Carolina Madtom occupying artificial cover (Figure 2).

We developed Carolina Madtom habitat suitability distributions, based on microhabitat use and availability data, to determine the most suitable ranges of microhabitat for each of the six microhabitat parameters. The most suitable range of water depth was between 0.30 and 0.49 m (Figure 3). The most suitable bottom velocity was 0.30–0.39 m/s. Interestingly, six Carolina Madtom were collected at a site with particularly fast velocity, as compared to relatively slow water elsewhere in the basin. Mean column velocity had multiple suitable ranges with modal peaks at 0.20–0.29, 0.40–0.49, and 0.70–0.79 m/s. These multiple peaks were likely influenced by the six Carolina Madtom captures at the site with particularly high water velocity. For both the dominant and subdominant substrate composition, gravel and cobble were the most suitable substrate types (Figure 3).

The most suitable cover type was artificial. This is because we placed constructed artificial cover units in the stream for another component of the study [14,30] and treated them as existing instream cover if they were in place when we conducted our snorkel surveys. Carolina Madtom occurrence was high inside the cover units; however, artificial habitat availability, either placed purposefully by us or occurring as instream litter, was extremely low throughout the basin, leading to high suitability values. Availability of the suitable habitat ranges was generally low (

Table 5. Comparison of the percentage by area of available microhabitat at surveyed sites, suitable for Carolina Madtom occupancy in the Neuse and Tar river basins.

		Neuse Basin	Tar Basin
Variable	Suitable Range	% Available	% Available
Depth	0.30–0.49 m	23.80	22.22
Bottom velocity	0.30–0.39 m/s	2.71	1.97
Mean-column velocity	0.20–0.29 m/s	10.96	13.25
	0.40–0.49 m/s	2.91	3.50
	0.70–0.79 m/s	0.42	0.70
Dominant substrate	Cobble	7.91	3.24
Subdominant substrate	Gravel	33.66	35.43
Cover type	Artificial	0.97	0.39

). Most suitable ranges accounted for less than 15% of the available instream habitat for each parameter measured.

We also compared the instream habitat between the Tar River basin and Neuse River basin to assess whether the decline in Carolina Madtom populations in the Neuse River basin might be attributed to a loss of suitable instream habitat. Comparisons of the two basins showed significant differences in the availability of habitat for each parameter (Figure 4). The Tar River basin was shallower than the Neuse, as the depth availability in the Neuse peaked at 0.20–0.29 m and had higher proportion percentages of 0.70 m and above. The Tar River basin reaches were also faster moving in general. Although both basins had right-skewed distributions, the Neuse River had slower velocities and exhibited lower proportions of available habitat at velocities of 0.20 m/s and above. Substrate composition was similar between the basins, with the Neuse River basin having more cobble and boulders as the dominant substrate, whereas the Tar River basin was composed of a sand and gravel mixture.

The Tar River basin also had a higher amount of total available cover for all cover types except for artificial cover, which could be explained by the more degraded, urban nature of the Neuse River basin that may be more likely to contain artificial litter. The suitability analysis between the two basins revealed variability in the amount of available suitable habitat. Using the ranges calculated above, the Neuse River basin generally had more habitat available in the suitable ranges for Carolina Madtom (Table 5). The Neuse River basin had higher percentages of available habitat in the suitable ranges for depth, bottom velocity, dominant substrate, and cover type, while the Tar River basin had higher percentages of suitable habitat for mean column velocity and subdominant substrate. As such, the lack of suitable instream physical habitat for Carolina Madtom cannot wholly explain the lack of Carolina Madtom occurrences in the Neuse River basin.

4. Discussion

Carolina Madtom populations have decreased steadily and substantially over time in both occurrence and abundance throughout their historical and current range, and our results revealed these continued downward trends between the study of Wood and Nichols [19] and ours in 2016–2017. The Neuse River population may be approaching extirpation, as we only captured two individuals, which were not part of our standardized sampling, from a single site in the Neuse River basin. The Tar River population is experiencing less drastic population losses; however, we documented a decrease in the current range of populations in the Tar River basin. Along with a decline in occurrence, we also observed a large decrease in captured individuals from the previous studies, suggesting that Carolina Madtom populations have also decreased in abundance. We captured a total of 67 fish (59 from standardized surveys, six from incidental collections in the Tar, and two from incidental collections in the Neuse) from only five sites during the 2016 and 2017 field seasons, even with substantial sampling efforts among 36 sites (75 sampling locations). Previous surveys have also shown these steady declines in Carolina Madtom populations over the previous 30 years of sampling; our research confirmed the continued trend of precipitous declines. For example, Wood and Nichols [19] found only two sites in the Neuse River basin that supported Carolina Madtom, accounting for a 92% decrease in population occurrence since the 1960s. Their corresponding capture abundance was 208 individuals during the 2007 field season [19], and a temporally overlapping study by [13] observed 274 Carolina Madtom over their 2007 and 2008 field seasons. Even with different sampling efforts and coverage between these two studies, their relatively consistent capture numbers suggest that Carolina Madtom population abundances have decreased when compared to our recent findings. Although occurrences and capture abundances were low in our surveys, our model results suggest that the visual snorkel survey technique, which was also employed by Wood and Nichols [19] and Midway et al. [13], remains an efficient means to study Carolina Madtom populations.

The detection probability of Carolina Madtom during our snorkel surveys was quite high but varied among sites ($\widehat{p}$ = 0.07–0.99), validating the snorkel technique as an effective method to capture Carolina Madtom. Even though Carolina Madtom occur in and under cover, employing a thorough search in snorkel surveys allowed us to effectively find and capture fish. Wood and Nichols [19] also estimated high detection probabilities ($\widehat{p}$ = 0.94) in their snorkel surveys for Carolina Madtom, and although occupancy models were not developed by Midway et al. [13], they reported Carolina Madtom densities similar to Wood and Nichols [19], further validating the snorkel sampling method. In wadable streams, snorkeling is a common and effective visual technique to sample fish with minimal mortality. Although snorkel surveying sometimes raises concerns over sampling bias and impaired visibility [31,32,33], studies on related madtom species and other benthic fishes have shown that visual detection through snorkeling is as effective as traditional methods such as electrofishing or seining [50,51,52]. Snorkel surveying has also been recommended as a non-lethal method to sample rare species, as in our study of the Carolina Madtom [53,54].

Our modeled estimate of Carolina Madtom occupancy probability was low, but variable among sites ($\widehat{\mathsf{\Psi }}$ = <0.01–0.68). Low occupancy with high detection demonstrates the patchy spatial distribution of the remaining Carolina Madtom populations. If an area encompassing a small patch population is sampled, it is likely that Carolina Madtom will be detected in nearly every survey of the sampling area. Low occupancy is expected among rare, endemic species sampling, and occupancy estimates below 0.60 are common [55,56], primarily because of both their rare or imperilment status and their fragmented distributions. Our modeling analysis also found that microhabitat covariates influenced occupancy and detection probability estimates. The top performing model included the influence of dominant substrate on occupancy and mean-column velocity on detection. Substrate composition is an important microhabitat variable for all madtom species, as most are found in riffles with sand and gravel substrates [42,43], including the Carolina Madtom. Our site-specific occupancy estimates showed that occupancy was greatest at sites with a predominantly sand substrate composition. The mean-column velocity detection results were similar, confirming that madtom species are found in free-flowing streams with adequate velocity to flush fine sediments and provide oxygenated water [42,43].

The microhabitat use observed during our study was in accord with all known documented sources of Carolina Madtom habitat occupancy [12,13,19]. Substrate, flow, and instream cover are important microhabitat features for madtom occupancy; however, they are generally among the first variables affected in impacted streams [9]. Interestingly, we found that almost 40% of the surveyed available habitat in the Tar River basin had no adequate cover for Carolina Madtom to inhabit. Carolina Madtom were found most frequently under cobble cover, although both large and small woody debris were used as well. In the more impacted Neuse River basin, we are uncertain whether the available cover percentage is similar to the Tar because of the lack of fish detections. Nonetheless, this is a potentially important parameter to evaluate in the future.

Among the threats to the continued existence of Carolina Madtom, increased urbanization and related land development are among them [18,29]. These activities are a major cause of sedimentation, and increased fine sediment and silt load in a river can be detrimental to aquatic life, including Carolina Madtom [18,21,57]. Suspended sediment decreases the ability of light to penetrate the water, leading to decreases in growth and biomass of primary producers and macrophytes, and adversely affects the visibility, respiratory, and feeding function of fishes and invertebrates [58,59]. Deposited sediment associated with land development also has adverse effects on freshwater organisms. Fine sediment deposition fills in interstitial spaces between larger substrate particles and increases their embeddedness [57]. This not only reduces abundance and biomass of primary fish food sources, such as invertebrate larvae, but also decreases suitable habitat and areas of cover for benthic fishes such as the Carolina Madtom, which loses spawning habitat as available cavities become filled with sediment. Sedimentation problems from bridge construction, mining operations, and clearcutting riparian vegetation have been well-documented for the Ouachita Madtom N. lachneri [60]; Frecklebelly Madtom N. munitus [61]; and Neosho Madtom N. placidus [62,63]. Sedimentation is likewise among the threats to the Carolina Madtom [13,18,64].

Dams are another threat to many freshwater species. Reduced river continuity and altered flows from dams have been shown to negatively affect the intensively studied Neosho Madtom [62,65]. There are multiple dams causing habitat and population fragmentation along both the Neuse and Tar river basins [26] that are likely affecting the Carolina Madtom. Although habitat degradation is common throughout most areas inhabited by madtom species, they continue to inhabit specific microhabitats. Whether due to reproductive constraints or obligate cavity nesting common to all madtom species [43], we found that Carolina Madtom non-randomly select microhabitat. These findings support those of Midway et al. [13] and identify important specific stream microhabitats and reaches, especially relating to substrate, flow, and cover, that support madtom occurrence. These findings may assist with guidance for maintaining or restoring these habitats that are necessary to conserve Carolina Madtom populations in the future.

We developed suitability functions to determine current suitable habitat parameter ranges for the Carolina Madtom. Monitoring instream flow and habitat to develop models or management scenarios to maximize suitable habitat generally rely upon either single- or multi-species suitability functions, in which suitable or preferred microhabitat ranges are determined for the target species [66,67]. Determination of suitable habitat is especially important for madtom species because their obligate cavity nesting and benthic association does not allow them to adapt to rapidly changing instream conditions compared to other more generalist species. Habitat use has been widely studied for many madtom species [42,43], and habitat suitability criteria have been developed for the Freckled Madtom N. nocturnus [68], and Orangefin Madtom N. gilberti [69]. Habitat suitability criteria are especially useful in impacted or flow-regulated basins, such as the Neuse River basin, where flow and substrate composition can change rapidly and drastically. Midway et al. [13] originally determined some limited suitable habitat ranges for Carolina Madtom, and the suitable habitat ranges that we developed were similar to theirs, indicating that microhabitat use and availability distributions had changed little over the 8–10 years between the studies, even though the Neuse River basin has likely continued to experience habitat change and degradation [21]. These findings suggest that the species’ habitat affinities have remained relatively stable, but the area and number of sites where the species occur have decreased over time.

Given that we and Midway et al. [13] have both shown that suitable instream physical habitat seems to exist in both the Neuse and Tar river basins, another threat that may be contributing to the decline in Carolina Madtom populations is possible predation by the non-native Flathead Catfish. Between the two river basins, Flathead Catfish were first detected in the Neuse basin and more recently, the Tar basin [48]. Flathead Catfish are common in the mainstem and larger tributaries, where historical Carolina Madtom populations no longer exist. This study is the first to document the co-occurrence of Carolina Madtom and Flathead Catfish at the same site. We found both species occupying a site on the Little River in the Neuse River basin (site Neuse 25; incidental collection) and a site on Fishing Creek in the Tar River basin (Tar 7). Both of these instances are highly concerning findings because Flathead Catfish are voracious predators that have been shown to suppress native fish populations by up to 50% and positively select for ictalurid prey, including native bullhead catfish and madtom species in North Carolina rivers. [47,49,70]. Future studies focusing on Flathead Catfish range overlap and diet analysis may be warranted if they continue expansion into smaller tributaries that have previously been refugia for Carolina Madtom.

Another potential threat influencing Carolina Madtom decline is pollution. Although physical habitat degradation has been widely studied in madtom species, water quality and toxicology studies are less common. However, metals and other aquatic contaminants have been identified as potential factors leading to population decrease in the Neosho Madtom [62,63,71]. Such pollutants may also affect the Neuse and Tar river basins and their biota. Human population growth, as well as increases in intensive animal agriculture in the basins, may be contributing to greater point and non-point source pollutants entering the ecosystem (e.g., via sewage, industrial effluent, road and field runoff). Future research may include water quality analysis throughout both basins in relation to Carolina Madtom occurrence. Moreover, little is known of Carolina Madtom chemical sensitivity; however, given their restricted range and low fecundity (hundreds of eggs) [43], water quality impacts are of concern and toxicological studies of pollutants on reproduction and survival are warranted.

The Carolina Madtom is listed as Endangered in the United States by the U.S. Fish and Wildlife Service, and our findings herein on occurrence, abundance, and habitat should help inform future recovery actions. Reintroduction efforts are common among federally listed species as viable population recovery goals [72]. Concerns have been raised over reintroduction efforts, as species are often reintroduced into habitat that retains problematic factors that caused the original population collapse [73]. However, multiple madtom species have been successfully reintroduced in the southeastern United States. The Smoky Madtom N. baileyi, and Yellowfin Madtom N. flavipinnis, have both been successfully propagated in captivity, returned to stream habitat in their native ranges in Tennessee, and successful post-reintroduction reproduction and population increases in the wild have been documented [74,75,76]. Successful captive breeding, augmentation of extant wild populations, and reintroduction of Carolina Madtom into historically occupied habitats are integral to preventing the species’ extinction and promoting its resiliency and recovery [18,64]. In 2018, the U.S. Fish and Wildlife Service and North Carolina Wildlife Resources Commission began partnering with Conservation Fisheries, Inc., in Knoxville, TN, USA to develop and refine successful captive spawning and husbandry protocols for the species. Since then, these partners have collected adults, eggs, and larvae from the wild, and Conservation Fisheries, Inc., has been successful in developing controlled propagation protocols. Partners released the first cohort of about 260 juvenile Carolina Madtom in 2021, augmenting the Fishing and Swift Creek populations in the Tar River basin. Progeny have been released annually at these locations since then, thus far contributing about 700 fish to improve the resilience of Carolina Madtom in those watersheds. Captive propagation will also support reintroduction into the species’ historical range under a new Safe Harbor Agreement, carried out through an Endangered Species Act Section 10 Enhancement of Survival Permit (ESPER0041144) issued to the North Carolina Wildlife Resources Commission by the U.S. Fish and Wildlife Service. However, identifying and ameliorating the drivers of Carolina Madtom decline will be essential before expanding species restoration efforts beyond populations in suitable habitats and protected watersheds (e.g., State Game Lands and conservation holdings).

5. Conclusions

The Carolina Madtom is small but mighty and iconic among ictalurid species; along with the Tar River Spinymussel (Parvaspina steinstansana) and Neuse River Waterdog, this triplet of endemic species highlights the special nature of the Neuse and Tar Rivers—a biodiversity trait broadly characteristic of many southeastern United States freshwater ecosystems. Swift and Fishing Creeks in the Tar River basin have been called the most biologically diverse waterways in the state, and Swift Creek may be the most important lotic system remaining on the Atlantic Seaboard [77,78]. Conservation of these fauna and ecosystems is vital [79], not only to maintain a unique, endemic species, but to also maintain diversity and overall health of these waterways. Although we found a relatively high detection probability of Carolina Madtom at our surveyed locations, their patchy distribution along with the fact that not all possible locations in the Neuse and Tar river basins were sampled, indicate that our findings may underestimate the actual number of fish still residing in the systems. However, this research adds to and reinforces the body of knowledge about the Carolina Madtom’s populations and habitat use. Our results will help guide natural resource managers toward informed recovery and protective management decisions to improve the viability of this important endemic species.