Maxent-directed field surveys identify new populations of narrowly endemic habitat specialists

Background Rare or narrowly endemic organisms are difficult to monitor and conserve when their total distribution and habitat preferences are incompletely known. One method employed in determining distributions of these organisms is species distribution modeling (SDM). Methods Using two species of narrowly endemic burrowing crayfish species as our study organisms, we sought to ground validate Maxent, a commonly used program to conduct SDMs. We used fine scale (30 m) resolution rasters of pertinent habitat variables collected from historical museum records in 2014. We then ground validated the Maxent model in 2015 by randomly and equally sampling the output from the model. Results The Maxent models for both species of crayfish showed positive relationships between predicted relative occurrence rate and crayfish burrow abundance in both a Receiver Operating Characteristic and generalized linear model approach. The ground validation of Maxent led us to new populations and range extensions of both species of crayfish. Discussion We conclude that Maxent is a suitable tool for the discovery of new populations of narrowly endemic, rare habitat specialists and our technique may be used for other rare, endemic organisms.

However, as the importance and number of roles crayfishes inhabit become clearer, so do the nature and breadth of the threats they face. Approximately one-third of the world's crayfish species are threatened with extinction (Richman et al. 2015). In the United States and Canada, 48% of all crayfishes are considered imperiled ). In fact, crayfishes in the United States trail only freshwater mussels and snails in their level of imperilment (Wilcove and Master 2005). The specific threats facing native species of crayfishes include modification of species' habitats or ranges, over-utilization, disease, and limited distributions ). Animals that are rare or possess a restricted range are more sensitive to these cumulative stresses than more common, widespread fauna. Intensifying these threats is the lack of natural history data for many crayfish species, particularly North American primary burrowing crayfishes (Moore et al. 2013).
It is hypothesized all crayfishes have the ability to construct refugia and access ground or atmospheric water for oxygen extraction by burrowing into the soil or substrate (Berrill and Chenoworth 1982;Hobbs 1981). Based on differences in natural history, Hobbs (1981) described three classes of burrowing crayfishes: tertiary, secondary, and primary burrowers. Tertiary burrowers dig shallow burrows only to escape frost, lay and brood eggs, or avoid desiccation.
Secondary burrowers spend much of their lives in their burrows, which normally have a connection to an open, permanent water body; however they do move out into open water occasionally. In contrast, primary burrowing crayfishes spend their complete life cycle underground. As primary burrowers only leave their burrows to forage and locate a burrow of the opposite sex for mating (Hobbs 1981), their burrows are rarely tied to permanent open water.
Rather, primary burrowers use subsurface groundwater for moisture and oxygenation. Primary burrowing crayfishes reach a high level of diversity in the eastern United States.
The Ouachita Highlands Freshwater Ecoregion, which covers southeastern Oklahoma, northeastern Texas, southern Arkansas, and northwestern Louisiana, has the sixth highest diversity of native crayfishes in the United States and Canada (Moore et al. 2013). Arkansas has 13 endemic crayfish species , of which eight are primary burrowing crayfishes. Within the Ouachita Highlands Freshwater Ecoregion, the Ouachita Mountains Ecoregion (OME) of southwestern Arkansas (Woods et al. 2004) has six species (Fallicambarus harpi, F. jeanae, F. strawni, Procambarus liberorum, P. parasimulans, and P. reimeri), which represents the highest diversity of primary burrowing crayfishes in the state. Because these crayfish occur at such a constrained geographic scale, it is important to accurately describe their habitat preferences to ensure management of habitats appropriate for those species.
Two narrowly endemic, primary burrowing crayfishes found in the OME are Fallicambarus harpi and Procambarus reimeri. Fallicambarus harpi was discovered in 1985 by H.H. Hobbs, Jr. and H.W. Robison and was known from Montgomery, Hot Spring, Garland, and Pike counties in Arkansas before 2015. Fallicambarus harpi is categorized as a primary burrower and is known from wet seepage areas with abundant sedges such as roadside ditches and other right of ways (Robison and Crump 2004). However, since 2004, no new information has been published relating to the range or habitat requirements of F. harpi. Procambarus reimeri was first discovered in 1979 by H.H. Hobbs, Jr. and was known from only one county (Polk) in western Arkansas before 2015. This primary burrowing crayfish reportedly constructs simple burrows in sandy clay soil in wet seepage areas as well as roadside ditches (Hobbs 1979, Robison 2008, though no information has been published relating to the range or habitat requirements of P. reimeri since 2008. Much of the information that does exist fails to address the relationships between the crayfishes and their habitat. The relevant studies lean heavily on descriptions of the habitat during sampling periods and lack empirical evidence of habitat suitability. However, the current habitat preferences of these animals can be used to make inferences about their historic suitable habitat, undisturbed by anthropogenic change. Human-made, linear right of ways (ROWs) are known to have burrowing crayfish burrows. The habitat in which these animals occurred naturally, before human development, could have been functionally similar to a human-made linear ROW. There are multiple records of burrowing crayfishes occurring in ROWs. These records are from species descriptions and natural history research (Doran and Richards, 1996;Hobbs and Whiteman, 1991;Norrocky, 1991;Page, 1985;Robison and Crump, 2004), and several museum databases (Illinois Natural History Survey Crustacean Database, Arkansas Department of Natural Heritage Database, and National Museum of Natural History Invertebrate Zoology Collection). This association is potentially due to the ease of sampling, as the ROW is highly accessible and provides a sampling area that results in high detectability from roadway maintenance, such as mowing and spraying woody vegetation. This roadway maintenance also contributes to a wet, low-herbaceous microhabitat. There currently is a lack of understanding as to whether the observed occurrence of these species in ROWs is an artifact of the accessibility of these habitats or whether the actual habitat created by ROW activities is preferred.
To assess the possibility F. harpi and P. reimeri prefer the microhabitat of the ROW and to fill in the knowledge gaps associated with their habitat preferences and population statuses, I developed a study with the following objectives: 1) to test a priori hypotheses about the habitat use of two primary burrowing crayfish species (F. harpi and P. reimeri) in relation to their occurrence in roadways, 2) to investigate the use of statistical modeling to predict the occurrence of individuals across the landscape and use these predictions to refine future sampling, and 3) to delineate the geographic distribution of both species using field-validated distribution models.
The body of this thesis is divided into three separate chapters. Chapter 2 describes the habitat preferences of F. harpi and P. reimeri in relation to their occurrence in ROWs. The use of computer modeling to predict the occurrence of both crayfishes is described in Chapter 3. The final chapter consists of range delineations for both species and a conservation assessment.

INTRODUCTION
Human-made linear right-of-ways (ROWs) such as roads, roadside ditches, public utility easements, and railroad lines, and their maintenance are dramatically altered landscape features that traditionally have been observed to have a negative effect on habitat and life-history attributes of animal populations (Rytwinski and Fahrig 2013). For example, ROWs can disrupt wildlife movements (Richardson et al. 1997), fragment habitat (Andrews 1990), and directly cause mortality (Ashley andRobinson 1996, Lode 2000). Roads can act as a physical barrier to movement as well as a behavioral barrier (Oxley et al. 1974, Riley et al. 2006). In addition, the particulate matter emissions from vehicles can be a negative attribute of roadside microhabitat environments (Thorpe and Harrison 2008). Conversely, roadsides can positively contribute to plant and animal persistence by acting as vectors for native and nonnative species dispersal (Gelbard and Belnap 2003) and have led to higher animal densities and diversity than in surrounding habitat (Adams and Geis 1983). Such interactions observed in these areas are directly contributable to the construction and maintenance of the linear ROW.
The objectives of roadside maintenance activities have changed very little since their conversion from dirt trails to paved roads and include maintaining hydraulic capacity of ditches, eliminating vegetative obstructions, and providing wildlife habitat where compatible with roadway traffic (Berger 2005). The ROW environment is disturbed constantly by roadside maintenance (e.g. mowing, spraying herbicide, tree cutting) and remains open, resulting in habitats that resemble early successional stages in natural landscapes. Thus, roadside maintenance can lead to open habitat within a matrix of forested habitat (Watkins et al. 2003), an alteration that can be both beneficial and detrimental to the persistence of wildlife.
The characteristics of animal populations that are vulnerable to negative road effects have been documented as those having high intrinsic mobility, high migration potential, multiple resource needs, low density/large area requirements, and a low reproductive rate; being a forest interior species; and displaying a behavioral avoidance of roads (Forman et al. 2003). Animals that display these traits are inhibited by the physical presence of the road and influences associated with the ROW, such as edge effects. Investigators have studied the responses of various biotic communities to ROWs (see Spellerberg 1998), but the interaction of biotic communities and ROWs is still not fully understood. Fahrig and Rytwinski (2009) reviewed biotic communities and roadsides and found that 59% of interactions resulted in a negative effect on animal abundance. The minority of animal populations that experience some positive effects from roadsides have a small territory range, have a high reproductive rate, and are small bodied.
Investigators have shown a positive response from fauna that exhibit these life-history characteristics (Peris and Pescador 2004, Rosa and Bissonette 2007, Ward et al. 2008. Small populations of endemic habitat specialists often experience negative effects from ROW construction and maintenance (e.g. Altrichter and Boaglio 2004, Pocock and Lawrence 2005, Semlitsch et al. 2007). However, construction and maintenance of ROWs could benefit directly some narrowly endemic habitat specialist taxa by creating suitable habitat (Forman et al. 2003).
One such taxon experiencing these benefits could be North American primary burrowing crayfishes.
All crayfishes are hypothesized to have the ability to construct refugia by way of burrowing into the soil or substrate (Hobbs 1981, Berrill andChenoworth 1982). Construction of burrows and the open space within them allow access to ground or atmospheric water for O2 extraction. Hobbs (1981) described three classes of burrowing crayfishes based on differences in natural history: tertiary, secondary, and primary burrowers. Tertiary burrowers dig shallow burrows only to escape frost or seek shelter when the body of water they are inhabiting dries up.
Secondary burrowers spend much of their lives in their burrows; however, they do move out into open water occasionally, and their burrows normally have a connection to an open, permanent water body. Primary burrowing crayfishes spend their complete life cycle underground. As primary burrowers leave their burrows only to forage and locate a burrow of the opposite sex for mating (Hobbs 1981), their burrows are rarely tied to permanent open water. Rather, these species use subsurface groundwater for moisture and oxygenation.
Prior to the 20 th century, the habitat in which some primary burrowing crayfishes occurred naturally could have been functionally similar to some human-made ROWs.
Specifically, human-made ROWs, such as roadsides, could imitate the hypothesized natural habitat of these animals by creating a landscape that is void of trees, supports a perched water table, and maintains an open, low-herbaceous microhabitat. To study the relationship between crayfishes and ROWs, we examined two narrowly endemic habitat specialists, Fallicambarus harpi and Procambarus reimeri, known from the Ouachita Mountains Ecoregion (OME) of western Arkansas (Woods et al. 2004). These species are vulnerable to population declines and are currently listed as endangered (P. reimeri) and vulnerable (F. harpi) according to . The conservation categories of endangered and vulnerable are based upon the American Fisheries Society Endangered Species Committee, which follows Williams et al. 1993. In addition, these species were included in a recent petition filed by the privately funded Center for Biological Diversity for protection under the US Endangered Species Act.
Both species are known historically from <40 individual sampling sites in restricted areas of the OME. Robison and Crump (2004) reported F. harpi as occurring in wet grassy areas that often had abundant sedges and grasslands such as ditches and pastures. Robison (2008) reported the habitat in which P. reimeri was observed as wet seepage areas and roadside ditches. Based on historic accounts of both species, we predicted some habitat attributes would be more important than others, particularly the presence of sedges and open canopy. We also expected soil composition would be a strong driver of burrow placement. Soil cues are important for other burrowing crayfishes (e.g. Grow and Merchant 1980, Barbaresi et al. 2004, Helms et al. 2013).
To evaluate whether F. harpi and P. reimeri could be experiencing a positive effect from the microhabitat in ROWs, we developed a study based on extensive field sampling and habitat modeling of multiple variables to determine the fine-scale habitat preferences of both F. harpi and P. reimeri in relation to ROWs.

Study site
Our study sites were situated in the Ouachita and Caddo River drainages of southwestern Arkansas. We focused on five counties in the OME that encompassed the entire known range of F. harpi and P. reimeri (Fig. 2.1). The OME has the highest diversity of primary burrowing crayfishes in the state. Six species occur there: Fallicambarus harpi, F. jeanae, F. strawni, Procambarus liberorum, P. parasimulans, and P. reimeri.
The Ouachita Mountains are composed of parallel, folded, east-west ridges underlain by shale and sandstone (Miser 1929). The soils of this region are generally categorized as silty clay and silty loam (Hlass et al. 1998). The most common forest community is mixed pine-hardwood; however, remnant pine-bluestem (Pinus-Schizachyrium) communities do exist (Phillips and Marion 2005). Logging and recreation make up the major land uses of this area, and pastureland and hay fields are found in the broader valleys (Woods et al. 2004). We focused our sampling effort in these broader valleys.

Field collections
All sampling took place in April 2014 during the peak activity period for both species (Robison andCrump 2004, Robison 2008) and thus would result in highest species detection.
The databases of the Illinois Natural History Survey Crustacean Collection, National Museum of Natural History Invertebrate Zoology Collection, and Arkansas Department of Natural Heritage were used to identify known historical locations for both F. harpi and P. reimeri. For each species, we selected historic localities that were accessible and could be validated with geographic positioning information. At each sampling site we positioned three to six 50 m transects ≤100 m from the initial transect. The initial transects were parallel to the road and within the ROW. We placed the initial transect at each sampling site where burrows were present, ensuring the initial transect was situated at the historical museum location. All transects were delineated with a fiberglass measuring tape. We laid out each transect and then checked for the presence or absence of burrows and standing water. After we obtained a global positioning system location and azimuth at the 0-m mark, we placed a 1-m 2 polyvinyl chloride quadrat over the tape every 10 m, which resulted in six 1-m 2 quadrats/50-m transect. After we completed the initial transect, we completed the remaining two to five transects in adjacent habitat in the same manner. We decided the number of transects to be sampled at each site based on habitat heterogeneity. If a site was homogenous, we sampled fewer transects to increase the number of sampling sites that could be visited during our sampling window. We defined adjacent habitat as having significantly more or less canopy cover, seemingly different soil moisture content, higher or lower elevation, or a different dominant vegetation type compared with the initial transect. We excavated burrows at each sampling site and along each transect to ensure any burrows counted at a sampling site harbored the target species. We collected voucher specimens of each target species from all sites with burrows present and deposited them in the Illinois Natural History Survey Crustacean Collection.

Habitat variables
We collected the following habitat variables within each 1-m 2 quadrat: percent tree canopy cover, percent herbaceous ground cover, stem density, number of burrows, and the presence or absence of hydrophilic sedges (Appendix A; Table A.1). We estimated percent tree canopy cover with a concave spherical densiometer (Spherical Densiometer, Model-C; Robert E.
Lemmon, Forest Densiometers, Bartlesville, Oklahoma). We calculated percent herbaceous ground cover by inverting the concave spherical densiometer over the 1-m 2 quadrat. We calculated stem density by counting the stems within a smaller (100-cm 2 ) quadrat placed within the upper right-hand corner of each 1-m 2 quadrat. We scored the presence vs absence of hydrophilic sedges by recording the presence or absence of herbaceous plants that had threeranked leaves, an angular stem, and a spiked fruiting body. At each transect, we collected three evenly spaced soil samples with a soil probe (AMS 7/8 in. [2.2 cm] diameter open-end probe) at a minimum depth of 43 cm and a maximum depth of 66 cm. These depths reflect the column of soil the crayfishes are using for burrow construction (Robison and Crump 2004 and validated in the field). We analyzed the soil samples with laser diffraction on a Malvern Mastersizer 3000 (Malvern Instruments, Malvern, UK) to obtain a percent composition (sand, silt, and clay) for each sample. We computed the isometric log-ratio transformation for these soil data (Egozcue et al. 2003). We constructed a soil texture plot with the package soiltexture in R ( Fig. 4; Moeys 2015). The selected habitat variables reflect habitat characteristics associated with F. harpi (Robison and Crump 2004) and P. reimeri (Robison 2008), other primary burrowing crayfish species (Hobbs 1981, Welch and Eversole 2006, Loughman et al. 2012, and biological intuition.
The habitat variables and a description of each model term used in the statistical analysis are shown in Table 2.1.
We mapped the quadrat locations in ArcGIS (version 10.2; Environmental System Research Institute, Redlands, California). We then calculated the following habitat variables for each quadrat with ArcGIS: elevation, distance to nearest waterbody, compound topographic index value (CTI), and solar radiation value. We measured elevation as the height in meters above sea level and distance to nearest waterbody as the Euclidean distance from all permanent waterbodies. We assessed CTI with the Geomorphometry and Gradient Metrics toolbox for ArcGIS (version a1.0; Evans et al. 2010). This metric is a function of both the slope and the upstream contributing area per unit width orthogonal to the flow direction. CTI is a steady state wetness index, where a larger CTI value represents areas that are topographically suitable for water accumulation. We measured solar radiation by calculating the watt-hour/m 2 of the delineated sampling area using the Area Solar Radiation tool in ArcMap. We calculated all ArcGIS values with digital elevation maps (National Elevation Dataset; http://ned.usgs.gov/) and surface water maps (National Hydrography Dataset; http://nhd.usgs.gov/index.html) at a resolution of 10 m in an attempt to minimize autocorrelation. We combined the habitat variables collected in the field and those calculated with ArcGIS into one data set for a fine-scale analysis of habitat features affecting crayfish burrow placement on the landscape.

Modeling analysis
We conducted all fine-scale statistical analyses in R (version 3.1.1; R Project for Statistical Computing, Vienna, Austria). We made the isometric log-ratio transformations with the package compositions (van den Boogaart et al. 2014) and used generalized linear mixed models to analyze the data (package lme4; Bates et al. 2014). The response variable in each model was the number of burrows within each 1-m 2 quadrat and was modeled with a Poisson error distribution and log link. We modeled burrow counts separately for each species. To account for potential site effects, we modeled transects nested within sites as a random effect in each model. We scaled and centered all habitat variables by subtracting the variable mean from each respective value and dividing by the standard deviation of that variable. We tested for overdispersion for each model before comparison of all models. We assessed model convergence and fit and then adjusted the optimization algorithm as needed. We did not include covariates having a Spearman correlation coefficient of >0.60 in the confined candidate model set in an attempt to avoid multicollinearity of our variables in the modeling suite. The full candidate model set and each hypothesis tested is shown in Table 2.2. We compared candidate models with Akaike's Information Criterion corrected for small sample sizes (AICc; Akaike 1974). We examined the relative support for each model and calculated unbiased model-averaged parameter estimates from the top models (ΔAICc < 4) with the package MuMIn (Barton 2014) by means of model selection and averaging methods described by Burnham and Anderson (2002) and Luckacs et al. (2009).

Field collections
Our search of museum databases resulted in 57 unique historic capture records (24 for F. harpi and 33 for P. reimeri). The records ranged from 1967 to 2008, and the oldest record we visited was from 1973. We sampled 11 of these localities (35 transects, 210 quadrats) for F. harpi and 9 of these localities (37 transects, 222 quadrats) for P. reimeri. Most (75%) of these localities were in the ROW of secondary, local, and private roads. Other sampling sites (25%) were situated in yards, pastures, and adjacent habitat farther from the ROW (up to 90 m).

Modeling analysis
Fallicambarus harpi and P. reimeri had similar patterns of habitat selection. For both species, canopy cover was the most important habitat variable, and it was present in all top models (ΔAICc < 4, Table 2.3). Model-averaged parameter estimates for both species are shown in Table 2.4. The number of burrows in a quadrat was negatively associated with canopy cover ( Fig. 2.2A, 2.3A) while the presence of hydrophilic sedges was positively associated with the number of burrows in a quadrat (Fig 2.2B, 2.3B). The transformed soil variables and stem density variable were also present in the top models. Burrows were generally present in quadrats with little to no canopy cover (mean ± SD, 4.4% ± 17.7, n = 110). No burrow was observed in a quadrat with complete canopy cover (100%). Sedges were present in 83% of the quadrats that harbored burrows of either species (n = 110).

DISCUSSION
We developed a suite of models to assess our predictions regarding the habitat preferences of F. harpi and P. reimeri. We found support for some of our predictions, whereas some results were counterintuitive. Open-canopy habitat and the presence of sedges were important for burrow placement across the sampled landscape. Our predictions that these variables would be preferred by both species of crayfish were supported by the models fitted with generalized linear mixed-model analysis. The presence of hydrophilic sedges is an indication of a seepage area or that the water table is relatively close to the ground surface (Schütz 2000). The absence of tree canopy cover also contributes to these wet seepage areas (Eastham et al. 1994). Our prediction that soil would be a strong predictor was not supported.
This outcome was potentially a result of the way we spatially segregated our soil samples. We collected soil samples that fell into only three distinct soil textural classes (silt loam, loam, and sandy loam; Fig. 2.4), which did not capture the variation seen across the entire OME. Overall, however, our findings point to the preference of ROW-like habitat for F. harpi and P. reimeri.
The habitat in which animals were most abundant was treeless, wet seepage areas with abundant low grasses and sedges. The soil composition at these occupied sites was primarily loam and silt loam (Soil Survey Division Staff 1993; 90% of F. harpi quadrats, 92% of P. reimeri quadrats).
The burrows of F. harpi and P. reimeri were complex, 0.5-1 m in depth, and connected to groundwater. Our results highlight the specific importance of these wet, open-canopy habitats as a preferred environment for both species.
Previous studies of other primary burrowing crayfish species have revealed the existence of habitat specialists and habitat generalists. Specialist species occur in habitats ranging from pitcher plant bogs (Fallicambarus gordoni [Johnston and Figiel 1997]) to sand ridges (Distocambarus crokeri [Welch and Eversole 2006]), whereas generalist species such as Procambarus gracilis (Hobbs and Rewolinski 1985), Fallicambarus devastator (Hobbs and Whiteman 1991), Fallicambarus fodiens (Norrocky 1991), Cambarus catagius (Mcgrath 1994), Cambarus dubius (Loughman 2010), and Cambarus thomai (Loughman et al. 2012) can be found in both forested floodplains and open habitat throughout their respective ranges. Based on the modeling and field observations, F. harpi and P. reimeri can be considered habitat specialists. They occur in wet, open herbaceous areas and not in the adjacent forested habitat. We believe the microhabitat of the roadside ditch is acting as suitable habitat for these specialists within a matrix of unsuitable habitat.
We sampled transects adjacent to known localities to better model habitat preference for each species. The habitats sampled by these transects generally differed in composition from the ROW (Appendix A; Table A.1) but were spatially proximate so as to be accessible to crayfishes. These sites composed 25% of the sampling locations and were not in the ROW. We designed this sampling scheme with the knowledge that it would be unnecessary and cost ineffective to sample the entirety of the OME randomly. Primary burrowing crayfishes rarely, if ever, inhabit permanent open water (Hobbs 1981) such as streams, lakes, and swamps, or highgradient slopes found in the larger OME. A review of over 2000 freshwater crayfish collections made in the state of Arkansas (Illinois Natural History Survey Crustacean Collection and National Museum of Natural History Invertebrate Zoology Collection) revealed no observations of F. harpi or P. reimeri in either of these habitat types. Thus, we think the microhabitat available to primary burrowing crayfishes is spatially restricted because of their life-history characteristics (Hobbs 1981), and those available habitats were represented in our sampling design.
The microhabitat found in roadside ditches where these animals occur is a result of the physical presence of the road and roadside maintenance. The surface of the road is less permeable than the surrounding habitat, which diverts precipitation into the surrounding terrain (MacDonald et al. 2001). The roadside ditch also intercepts groundwater flow, adding more water to the roadside microhabitat (Forman et al. 2003). Roadside maintenance halts succession by removing woody stems and constantly disturbs the herbaceous community with mowing and herbicide application. The removal of woody stems also increases the soil moisture in the roadside habitat (Eastham et al. 1994) because woody stems have deeper root systems and are capable of transpiring more water from the soil than are herbaceous plants. Through these roadmaintenance activities, the ROW habitat is uniformly distributed where the road occurs. These attributes of the roadside appear to have created suitable habitat for these two species of primary burrowing crayfishes.
The characteristics of animal populations vulnerable to negative road effects have been documented as having a high level of mobility, behaviorally avoiding roads, and being habitat generalist or forest interior species. The negative responses observed in the biota that have these characteristics are not seen in F. harpi or P. reimeri. These two species do not show a high level of mobility or a behavioral avoidance of roads: some of the ROW sites revisited in our study were first discovered >40 years ago. These animals are also not perceived as being habitat generalists or forest interior species. They were observed in open, wet, herbaceous areas, which suggests these species are habitat specialists within the broader matrix of forested habitat that dominates the OME. They may also be avoiding other direct negative effects of inhabiting ROWs by spending most of their life underground. Other species of burrowing crayfishes have been observed evading the effects of pesticide use by occupying a buffer zone within their burrows during pesticide application (Sommer 1983).
Our findings add to the understanding of the interactions between ROWs and the biota that live within them. Previous research has shown positive and negative responses of biota to ROWs (e.g. Adams and Geis 1983, Forman et al. 2003, Fahrig and Rytwinski 2009). Our study is the first to show a positive interaction between a narrowly endemic habitat specialist and a ROW habitat that is commonly seen as highly altered and detrimental to endemic wildlife populations. We are confident that we captured different potential habitat types available to F. harpi and P. reimeri by sampling the adjacent habitat. The adjacent habitats were out of the ROW and generally did not have the habitat characteristics of the roadside ditch. We think these crayfishes prefer the ROW microhabitat because of the lack of canopy and presence of sedges, which presents a moist, low-herbaceous environment. These data support the benefit of ROWs to the persistence of these narrowly endemic habitat specialists. The use of this habitat by these species could also encourage dispersal along these linear corridors. Future work is needed to assess this possibility and to investigate locations within the OME within and well beyond the roadside ditch, where these animals are not known to occur.

INTRODUCTION
Understanding the factors influencing species distributions and habitat selection are critical to researchers (Baldwin 2009). For example, rare or narrowly endemic plants and animals are difficult to monitor and conserve when their total distribution and habitat preferences are not completely known. One method used in this research is species distribution models (SDMs).
SDMs are correlative models using environmental and/or geographic information to explain observed patterns of species occurrences (Elith and Graham 2009). SDMs can provide useful information for exploring and predicting species distributions across the landscape (Elith et al. 2011). Models estimated from species observations can also be applied to produce measures of habitat suitability (Franklin 2013). This information can be useful for detecting unknown populations of rare, endemic, or threatened species (e.g. Williams et al. 2009 Maxent is a presence-only modeling algorithm using a set of known occurrences together with predictor variables such as topographic, climatic, edaphic, biogeographic, and remotely sensed variables (Phillips et al. 2006, Phillips andDudík 2008). These data are used to predict the relative occurrence rate of a focal species across a predefined landscape (Fithian and Hastie 2013). Recent studies focusing on the performance of Maxent have revealed it to perform well in comparison to other SDMs (Elith et al. 2006). Maxent also performs well with small sample sizes (e.g. Hernandez et al. 2006, Pearson et al. 2007, Wisz et al. 2008, rare species (e.g. Williams et al. 2009, Rebelo andJones 2010), narrowly endemic species (e.g. Rinnhofer et al.

2012)
, and when used as a habitat suitability index (Latif et al. 2015).
In all of the above applications, the potential for the inaccurate execution and interpretation of an SDM is well documented (e.g. Araújo and Guisan 2006, Baldwin 2009, Yackulic et al. 2013. Specific issues surrounding the interpretation of Maxent analyses include sampling bias (Phillips andDudík 2008, Boria et al. 2014) and the lack of techniques to assess model quality (Hijmans 2012), overfitting of model predictions North America has the highest diversity of crayfishes worldwide ).
Within North America, 22% of the species listed as endangered or threatened in a recent conservation review of crayfishes were primary burrowing crayfishes ). These listings were based upon the American Fisheries Society Endangered Species Committee, which follows the criteria of Williams et al. (1993). It is hypothesized all crayfishes have the ability to construct refugia by way of burrowing down into the soil or substrate (Berrill and Chenoworth, 1982;Hobbs, 1981). Hobbs (1981) described three classes of burrowing crayfishes based on differences in natural history: tertiary, secondary, and primary burrowers. Tertiary burrowers dig shallow burrows only to escape frost or seek shelter and when the body of water they are inhabiting dries up. Secondary burrowers spend much of their lives in their burrows; however, they do move out into open water occasionally, and their burrows normally have a connection to an open, permanent water body. Primary burrowing crayfishes spend their complete life cycle underground. As primary burrowers leave their burrows only to forage and find a burrow of the opposite sex for mating (Hobbs 1981), their burrows are rarely tied to permanent open water.
Amongst the three types of burrowers, the least is known regarding the natural history of primary burrowing crayfishes (Moore et al. 2013;) due to the challenges in sampling these largely fossorial animals (Larson and Olden 2010). However, the narrowly endemic nature of North American crayfishes is well documented (Morehouse and Tobler, 2013;Page, 1985;). Primary burrowing crayfishes in Arkansas are no exception  The rarity of and difficulties surrounding the collection of natural history information, specifically habitat suitability, make primary burrowing crayfishes ideal candidates for SDM. To test the ability of SDMs to predict the distribution of suitable habitat for two narrowly endemic habitat specialists, we constructed SDMs for F. harpi and P. reimeri and validated the models using independent sampling data. These species are vulnerable to population declines and are currently recorded under the Endangered (P. reimeri) and Vulnerable (F. harpi) conservation status categories ) based on modifications to or reductions of habitat in their already restricted ranges. Additionally, these species were included in a recent petition filed by the privately funded Center for Biological Diversity for protection under the Federal Endangered Species Act.

Study Area
Our study sites were located in the Ouachita and Caddo River drainages of southwestern Arkansas in the Ouachita Mountains Ecoregion (OME; Woods et al. 2004). The remnant pinebluestem (Pinus-Schizachyrium) communities (Phillips and Marion 2005) and silty loam soil (Hlass et al. 1998) make this region of Arkansas ideal habitat for primary burrowing crayfishes and in fact has the highest diversity of primary burrowing crayfishes in the entire state with six species (Fallicambarus harpi, F. jeanae, F. strawni, Procambarus liberorum, P. parasimulans, and P. reimeri). We sampled thirteen counties encompassing the known range of both species of primary burrowing crayfish: from east to west, those counties were Pulaski, Saline, Perry, Garland, Hot Spring, Clark, Yell, Montgomery, Pike, Scott, Howard, Polk, and Sevier ( Fig. 3.1).

Presence data and environmental variables
To determine habitat requirements of F. harpi and P. reimeri, we queried natural history museums or databases (Illinois Natural History Survey Crustacean Collection, the National Museum of Natural History Smithsonian Institution, and the Arkansas Department of Natural Heritage) for historic locations of both species, and a subset of those locations were visited (Rhoden et al. in press). At each location, we measured the following suite of habitat variables: percent tree canopy cover, percent herbaceous ground cover, stem density, the number of burrows, presence of standing water at the site, remotely sensed variables and the presence or absence of hydrophilic sedges. We found canopy cover and the presence of hydrophilic sedges were the most important factors in predicting crayfish abundance (Rhoden et al. in press).
The environmental variables used for the SDM analysis consisted of canopy cover, elevation, distance to nearest waterbody, compound topographic index value (CTI), and solar radiation value (  (Evans et al. 2010).
CTI is a steady state wetness index, where a larger CTI value represents areas that are topographically suitable for water accumulation. We measured solar radiation by calculating the watt-hour/m 2 of the delineated sampling area using the Area Solar Radiation tool in ArcMap (Table 3.1). These habitat variables reflect habitat characteristics associated with F. harpi (Robison and Crump 2004) and P. reimeri (Robison 2008), and other primary burrowing crayfish species (Hobbs 1981, Welch and Eversole 2006, Loughman et al. 2012. These values were calculated using digital elevation maps (National Elevation Dataset http://ned.usgs.gov/ accessed 07/21/2014) and surface water maps (National Hydrography Dataset http://nhd.usgs.gov/index.html accessed 07/21/2014). The entire OME was used as a delineation for both species of crayfish in the SDM analysis. Each surface was resampled to a common resolution of 30 m to match the resolution of the canopy surface.

Maxent analysis
Upon gathering the presence localities and calculating rasters of pertinent habitat variables (canopy cover, CTI, elevation, solar radiation, and distance to nearest waterbody), we created our distribution models, refined our models, tested the models against a null model and generated two final SDMs. The SDM algorithm used was Maxent (version 3.3.3k; Phillips et al. 2006). Along with the presence localities of each species, we incorporated 2500 background points to create our models. These points were randomly generated within a 10 km 2 polygon situated around each historic museum locality sampled in the field. A 10 km 2 buffer was used due to the distance between sites and the relative size of the area in which we projected our Maxent predictions (the OME). This approach followed Peterman et al. (2013) and was implemented to reduce model bias as described by Phillips (2008). The initial model fit was assessed using the area under the receiver operator curve (AUC).
Upon calculation of the initial Maxent models, a comparison of spatial predictions was conducted with the ENMeval package (Muscarella et al. 2014) in program R (R Development Core Team 2014). ENMeval analysis of F. harpi identified a betamultiplier of 2.5 and a linear quadratic hinge feature to provide the most parsimonious fit to our data. ENMeval analysis of P.
reimeri identified a betamultiplier of 1.5 and a linear quadratic hinge feature to provide the most parsimonious fit to our data. Spatial predictions were then re-run using the refined regularization multiplier and feature classes to increase the rigor in building and evaluating our SDMs based on presence-only data.
The refined models' performance was determined using the null model approach of Raes and ter Steege (2007) using ENMtools (Warren et al. 2010). We generated two groups of 999 random data sets containing 56 and 50 samples, which corresponded to the number of presence locations used for F. harpi and P. reimeri (respectively) in the initial model. These points were drawn without replacement from the OME delineation used in the initial model. Both model AUC values were compared to the 95 percentile of the null AUC frequency distribution.
The final Maxent models were calculated with the maximum number of iterations set to 5000 and the analysis of variable importance was measured by jackknife and response curves.
The bootstrap form of replication was used. These settings, the refined regularization multiplier and feature classes, and the recommended default values were used for our final Maxent model runs. Due to the narrowly endemic nature of both species and the small amount of presence locations in the initial model, we did not include a bias file or spatial filtering.

Field sampling and validation
The refined Maxent models (one for F. harpi and one for P. reimeri) were used to select 80 semi-random sampling sites for each species within the OME. These sites were semi-random because we restricted our sampling to areas of public access (roadside ditches). The Maxent output for both species was placed into four categories based on the relative occurrence rate (ROR; Fithian and Hastie 2013). The first category ranged from 0 to the lowest presence threshold (LPT) of each species (Pearson et al. 2007). The LPT is the smallest logistic value associated with one of the observed species localities. The second class ranged from the LPT to 50% of the maximum ROR of each species. The third category ranged from 50% of the maximum ROR to 75% of the maximum ROR of each species. The fourth category ranged from 75% of the maximum ROR to the maximum ROR of each species.
The final Maxent model outputs for both species were placed into the described categories in ArcMap. A polygon in ArcMap represented each category. Any polygon representing a single pixel or island (one 30 m x 30 m area) was removed. All category polygons were then overlaid with a layer representing the public right of ways and other public areas (state parks, natural areas, etc.). We generated 40 random points in each category polygon using the final polygon layer. All points within each category polygon had a spatial buffer of 2 km and were checked before sampling to ensure accessibility. If a point was inaccessible in the field, the next closest accessible point within the respected category was chosen and sampled. To assess the accuracy of the Maxent predictions, we calculated the receiver operating characteristic (ROC) and the AUC for the ROR of occupied quadrats vs. the ROR of unoccupied quadrats (Fawcett 2006)  Field sampling occurred in March and April of 2015, the period of peak activity for both F. harpi and P. reimeri (Robison and Crump 2004). At each sampling point, one 50-m linear transect was searched for the presence of burrows in six 1-m 2 quadrats placed at 10 m intervals along each transect. Within a sampling polygon, the area surrounding the transect was also thoroughly searched for burrows ( Fig. 3.2). If burrows were present along the transect, quadrat, or within the vicinity of the transect, animals were captured by hand excavation using a hand shovel to slowly dig around the burrow entrance and inserting one's arm into the burrow feeling for the crayfish. This method was chosen over other methods due to the success rate and limited amount of time spent at each burrow location (Ridge et al. 2008). Voucher specimens of crayfishes collected were deposited into the Illinois Natural History Survey Crustacean Collection.

Presence data
The presence locations used for the Maxent analysis, based on the field surveys of 2014, consisted of 58 locations for F. harpi (of which 56 were used for the SDM analysis) and 53 locations for P. reimeri (of which 50 were used for the SDM analysis). To minimize spatial autocorrelation, a subset of the original presence data was used. All duplicate presence locations falling within the same cell of a 30 m resolution raster were removed before the SDM analysis.
The selected presence locations used for the SDM analysis were near (<90 m) primary and secondary roadways.

Maxent analysis
The AUC converged to 0.959 and 0.976 for the final F. harpi and P. reimeri models, respectively. Both models were significantly better than the random AUC estimations from the null models (p<0.001). Of the parameters included in the model, canopy cover was the variable with the highest percent contribution for both species (48.8% and 47.2% F. harpi and P. reimeri, respectively; Table 3.2). Both species showed a steady decline in the probability of presence as canopy cover increased. The variable with the highest gain when used in isolation was elevation for both species (Table 3.2). An elevation between 150 m and 200 m was most suitable for F.
harpi and between 300 m and 350 m was most suitable for P. reimeri. The concentration of the highest ROR was centered around the presence locations for both species (Fig. 3.3). The LPT was 0.07 for F. harpi and 0.26 for P. reimeri. In the F. harpi model, 10% of the area in the OME was predicted to be above the LPT. In the P. reimeri model, 2% of the OME was predicted to be above the LPT (Table 3.3).

Field sampling and validation
Most (89% for F. harpi and 98% for P. reimeri) of the land area in the OME was in the first (lowest ROR) category (  (Fig. 3.4). Procambarus reimeri was captured in 41 of the 480 quadrats surveyed for the species (Fig. 3.4). We counted 70 burrows for each F. harpi and P.
reimeri. The historic range of F. harpi was extended by 2.8 km to the north and 3.2 km to the south of its historic range while the historic range of P. reimeri was extended by 51.6 km to the east, 12.1 km to the south, and 19.2 km to the west of its historic range. Thus, the total range for both species was approximately 265 km 2 for F. harpi and 1467 km 2 for P. reimeri using a minimum convex polygon approach in ArcGIS encompassing all known capture localities from both years and historic museum data.
The AUC for the F. harpi field validation was 80.96 (73.94-87.98). The AUC for the P.
reimeri field validation was 70.5 (63.33 -77.67). The ROC plot, AUC, and 95% confidence intervals for each model are shown in Figure 3.4. The threshold values (prediction with the highest specificity and sensitivity) were 0.68 and 0.57 for F. harpi and P. reimeri, respectively.

DISCUSSION
We found SDMs to be a useful tool to predict the occurrence and distribution of suitable habitat of two narrowly endemic, burrowing crayfish species in Arkansas. We used Maxent along with a suite of functions to assess model fit and safeguard against potential pitfalls associated with the Maxent program (Phillips and Dudík 2008, Elith et al. 2010, Warren and Seifert 2011, Hijmans 2012). We also used biologically relative habitat information at a constrained geographic scale to increase the accuracy of our predictions (Guisan and Thuiller 2005). These habitat variables and the scale at which we delineated them were a result of previous field sampling and analysis of habitat preference of both species (Rhoden et al. in press), which revealed both crayfish to be microhabitat specialists; using open, low-herbaceous microhabitats. We validated the model through a stratified sampling of our Maxent model predictions based on the LPT and the maximum ROR. We then equally sampled each category across the entire OME. This validation resulted in the range expansion of both species and the discovery of new populations. The models performed well by directing sampling efforts to treeless areas on the landscape that tended to have greater predicted probabilities of occurrence. However, the models did a poor job of identifying the wet, low-herbaceous microhabitats most frequently associated with occurrence in the field and previous studies (Robison andCrump 2004, Robison 2008, Rhoden et al. in press).
The use of the LPT to determine the threshold between the probability of presence or absence at any given predicted output location (Pearson et al. 2007) is well documented (Rinnhofer et al. 2012, Fois et al. 2015. We successfully used this value in our field validation techniques: no animal was captured in an area predicted below the LPT (Table   3.3). The land area above the LPT for the F. harpi model comprised 10% of the Ouachita Mountain Ecoregion (OME) and 2% for the P. reimeri model in Arkansas. The ROC analysis identified threshold values of 0.68 and 0.57 for the F. harpi and P. reimeri models, respectively, which optimized the sensitivity and specificity of our model (Robin et al. 2011). These values are far more conservative than the LPT and are based on the field validation results from both species. Using the threshold metrics, the area predicted as suitable habitat for F. harpi and P.
reimeri is less than 1% of the OME. We recommend the use of this threshold based on the ROC analysis for a more fine-tuned sampling effort for high-quality habitat for both species in the future.
Our SDMs used fine-scale (30 m) rasters of relative biological variables (canopy cover, CTI, solar radiation, elevation, and distance to waterbody). In the past, it has been common to use coarse (≥1 km) climatic data to construct models (e.g. Peterson 2001, Chunco et al. 2013, Barnhart and Gillam 2014. The use of coarse-scale habitat variables in Maxent has been addressed in previous studies Guisan 2006, Jiménez-Alfaro et al. 2012). Others using fine-scale inputs have found new populations of other rare species such as the discovery of new breeding ponds for a salamander species in east central Illinois (Ambystoma jeffersonianum; Peterman et al. 2013). Using fine-scale rasters of specific habitat variables for narrowly endemic habitat specialists was more appropriate than the more general approach of coarse-scale climatic data due to the resolution one gains with specific habitat information and fine-scale inputs. We believe this resolution was necessary to capture elements of the microhabitat the crayfishes prefer, differentiating between suitable and unsuitable habitat at a scale to assess animals occurring within anthropogenically altered habitat situated in natural landscapes (roadside ditches).
The habitat attributes of sites in which animals were present consisted of treeless, wet, low-herbaceous microhabitats. The average canopy cover for the categories above the LPT (category 2, 3, and 4) was 17% for both species. The presence quadrats for both species had an average canopy cover of 5%. Hydrophilic sedges were present in over 90% of the quadrats having F. harpi and P. reimeri, however sedges were present in less than half of the quadrats predicted above the LPT (category 2, 3, and 4). The sites recorded as being above the LPT (categories 2, 3, and 4) not having the target species were treeless for the most part, however those sites did not exhibit a moist microhabitat. The Maxent models did not capture the perched water table observed across the landscape that has been associated with other primary burrowing crayfishes (Welch et al. 2007). It is likely the model did not capture these moist, low herbaceous habitats due to the variables chosen for the Maxent analysis (canopy cover, CTI, elevation, solar radiation, and distance to nearest waterbody). Future studies could incorporate remotely sensed data to better identify these unique habitats.
Our study shows Maxent is an appropriate tool to analyze and discover populations of narrowly endemic species in the Ouachita Mountains of Arkansas. Our method of initially collecting habitat data using museum records in the spring of 2014 added precision to our presence locations used for analysis. Our initial surveys also added valuable information regarding the habitat preferences of both F. harpi and P. reimeri, which in turn guided the selection of our habitat variables for both models. Our concentrated search efforts resulted in the discovery of five new populations of F. harpi and 16 new populations of P. reimeri and range expansions of approximately 91 km 2 and 1404 km 2 , respectively. The discovery of new populations support the contention both species appear to be locally abundant where habitat is suitable for persistence will aid in the conservation of these rare species. This is accomplished by narrowing the knowledge gap in distribution information, adding localities for monitoring persistence in roadside ditches, and providing habitat preference information. All of these attributes are required for the refinement of conservation and management decisions.
Constructing models followed by ground validation has added valuable habitat information to two spatially restricted, understudied species of primary burrowing crayfish in southwestern Arkansas and illustrates the effectiveness of such a strategy for other rare habitat specialists.

Figure 3.2: Map representing the sampling scheme based on the predictions from a Maxent analysis of two primary burrowing crayfish species (Fallicambarus harpi and Procambarus reimeri) in southwestern Arkansas in the spring of 2015.
Each color (green, yellow, orange, and red) represents a relative occurrence category (1, 2, 3, 4) upon which the field validation sampling procedure was based. The black lines in the lower graphic depict 50-m transects used to assess presence or absence of the target species at each site. The linear, focused colors in the bottom graphic represent the accessible polygons in which the transect sampling was carried out.

Figure 3.3: Projection of the Maxent models for (A) Procambarus reimeri and (B)
Fallicambarus harpi onto the environmental variables (Table 3.

Figure 3.4: ROC analysis for two primary burrowing crayfish species, Fallicambarus harpi (A) and
Procambarus reimeri (B), in southwestern Arkansas. Input data derived from the predictions of two Maxent models and presence/absence data from field surveys conducted in the spring of 2015. The cross represents the 95% confidence interval and shaded bars represent the associated error for each ROC curve. AUC values are given with 95% confidence intervals in parenthesis. A higher AUC value depicts a better classification. Inset bar chart represents the percentage of quadrats (y-axis) in each category 1, 2, 3, and 4 (x-axis) that harbored each species of burrowing crayfish.

INTRODUCTION
North America has the highest diversity of crayfishes worldwide .
Within North America, 22% of the species listed as endangered or threatened in a recent conservation review  were primary burrowing crayfishes. Primary burrowing crayfishes differ from stream-dwelling crayfishes in their life history traits. They spend their complete life cycle underground, leaving their burrows only to forage and search for a burrow of the opposite sex for mating (Hobbs 1981 The narrowly endemic nature of North American crayfishes is well documented (Morehouse and Tobler, 2013;Page, 1985;, and primary burrowing crayfishes in Arkansas are no exception . The endemic nature of any taxa makes them more vulnerable to localized extirpation. Because these animals occur at such a constrained geographic scale, it is important to accurately describe their habitat preferences and range to ensure the persistence of local populations through the management of suitable habitat and monitoring of existing populations. Two such primary burrowing crayfishes in Arkansas in need of these conservation assessments are Fallicambarus harpi (Ouachita Burrowing Crayfish) and Procambarus reimeri (Irons Fork Burrowing Crayfish).
Since their formal descriptions, little information has been published regarding the habitat preferences and range assessments of F. harpi and P. reimeri. Both crayfishes are endemic to the OME in southwestern Arkansas. These species are vulnerable to population declines and are currently recorded under the Endangered (P. reimeri) and Vulnerable (F. harpi) conservation status categories based on modifications to or reductions in habitat in their already restricted ranges ). Also, these species were included in a recent petition filed by the privately funded Center for Biological Diversity for protection under the Federal Endangered Species Act. To assess these conservation concerns, we developed a study with the following objectives: 1) to determine the holistic range of both species (F. harpi and P. reimeri) within the OME of Arkansas and 2) to refine the description of suitable habitat for both F. harpi and P.
reimeri.  (Hobbs and Robison 1985). Robison and Crump (2004) reviewed and updated the status of this endemic crayfish, revealing 12 new populations in Montgomery, Hot Spring, Garland, and Pike counties in Arkansas. Since 2004, no other information has been published relating to the range or habitat requirements of this species. This primary burrowing crayfish is known from wet seepage areas with abundant sedges such as roadside ditches and other right of ways (Robison and Crump 2004). Fallicambarus harpi most closely resembles both F. strawni and F. jeanae but differs by possessing a free, never adnate, cephalic process on the first pleopod of the first form male (Hobbs and Robison 1985). information has been published relating to the range or habitat requirements of this species. This primary burrowing crayfish reportedly constructs relatively simple burrows in sandy clay soil in wet seepage areas and roadside ditches (Hobbs 1979, Robison 2008. This species most closely resembles P. gracilis and P. liberorum but differs from both by possessing a broader areola and lacking tubercles on the annulus ventralis (Hobbs 1979).

METHODS
Field surveys were conducted in the spring of 2014 and 2015 for these two primary burrowing crayfishes in southwestern Arkansas. We sampled in the spring of both years as this is the time of peak activity for both species, which correlates with the time of highest species detection. In 2014, we sampled historic localities from museum databases. Historical records for both species were queried from the Illinois Natural History Survey Crustacean Collection, In 2014, three to six 50-m transects were positioned at and near the historical museum locality. The first transect at each sampling site was placed where burrows were present, ensuring the first transect was indeed situated over the historical museum location. After a GPS location and azimuth were taken at the 0-m mark, a 1-m 2 PVC quadrat was placed over the linear transect every 10 m resulting in six 1-m 2 quadrats per 50-m transect. After the initial transect was completed over the historical location, two to five transects were completed in the same way in adjacent habitat at the sampling site. Habitat heterogeneity determined the number of transects sampled at each site. If a site was homogenous, fewer transects were sampled to increase the number of sampling sites to be visited during our sampling window.
In 2015, we sampled a semi-random group of sites based on two species distribution models, one for F. harpi and one for P. reimeri. These species distribution models were constructed with Maxent (Phillips et al. 2006). Maxent is a presence-only modeling algorithm using a set of known occurrences together with predictor variables such as topographic, climatic, edaphic, biogeographic, and remotely sensed variables (Phillips et al. 2006, Phillips andDudík 2008). These data are used to predict the relative occurrence rate of a focal species across a predefined landscape (Fithian and Hastie 2013). At each sampling site, one 50-m transect was positioned at a GPS location selected a priori from the output of the Maxent model. After a GPS location and azimuth were taken at the 0-m mark, a 1-m 2 PVC quadrat was placed over the linear transect every 10 m resulting in six 1-m 2 quadrats per 50-m transect. In both years, supplemental sampling was conducted in the vicinity (<100 m) of each sampling site to confirm or deny the presence of each species.
The density of individuals per 1-m 2 quadrat was estimated with the following equation: In the case that two individuals were found in several burrows (n=3), average quadrat density was calculated for burrow populations of n=1 and n=2.
In both years (2014 and 2015) all quadrats along each transect were searched thoroughly for the presence of active burrows. A burrow was considered active if it had freshly deposited mud at the entrance or the presence of a chimney and lack of debris or spider webs at entrance (Simon 2004 Akaike 1974). A subset of active burrows were hand excavated at each sampling site and along each transect to ensure any burrow counted at a sampling site harbored the target species. Hand excavation consists of using a hand shovel to dig slowly around the burrow entrance and following the main burrow tunnel, feeling for the crayfish as one works their way through the burrow complex. This method was chosen over other methods due to the success rate and limited amount of time spent at each burrow location (Ridge et al. 2008). Voucher specimens of each target species were collected and preserved in 70% ethanol from all sites with burrows present and deposited in the Illinois Natural History Survey Crustacean Collection.

RESULTS
The field surveys from 2014 and 2015 resulted in the sampling of 690 quadrats across 91 sites for F. harpi and 702 quadrats across 89 sites for P. reimeri (Fig. 4.1; Appendix C, D;

Fallicambarus harpi
Of the 91 sites sampled for F. harpi, the species was found at all historical sites (n = 11) and 5% of sites sampled in 2015 (n = 5) ( Fig. 4.2). We counted 214 active burrows in both years.
The modeling analysis revealed the number of burrows in a quadrat was negatively associated with canopy cover, herbaceous ground cover, and percent clay composition of the soil. The presence of hydrophilic sedges and amount of solar radiation was positively associated with the number of burrows in a quadrat. Burrows were generally present in quadrats with little to no canopy cover (n = 79, µ = 4.8%, σ = 18.7). Sedges were present in 76% of the quadrats with burrows present (n = 60) and present in 40% of quadrats where burrows were absent (n = 247).
Assuming a burrow population of one individual, the density of individuals per 1-m 2 quadrat harboring burrows of only F. harpi was 1.82 (±0.06) individuals per m 2 of occupied quadrat.
Assuming a burrow population of two individuals the density of individuals per 1-m 2 quadrat harboring burrows of only F. harpi was 3.64 (±0.12) individuals per m 2 of occupied quadrat.
Fallicambarus harpi was captured in one transect with P. parasimulans present and one transect with P. simulans.

Procambarus reimeri
Of the 89 sites sampled for P. reimeri, the species was found at all but one historical site sampled (n = 8) and 14% of the sites sampled in 2015 (n = 16) ( Fig. 4.2). We counted 140 active burrows in both years. The modeling analysis revealed the number of burrows in a quadrat was negatively associated with canopy cover. The presence of hydrophilic sedges and amount of solar radiation was positively associated with the number of burrows in a quadrat. Burrows were generally present in quadrats with little to no canopy cover (n = 90, µ = 4.1%, σ = 12.8). Sedges were present in 94% of the quadrats with burrows present (n = 85) and present in 44% of quadrats where burrows were absent (n = 269). Assuming a burrow population of one individual the density of individuals per 1-m 2 quadrat harboring burrows of only P. reimeri was 0.93 (±0.01) individuals per m 2 of occupied quadrat. Assuming a burrow population of two individuals the density of individuals per 1-m 2 quadrat harboring burrows of only P. reimeri was 1.86 (±0.02) individuals per m 2 of occupied quadrat. Procambarus reimeri was captured in one transect where P. tenuis was present.

DISCUSSION
Our study assessed the range and habitat of two primary burrowing crayfish species in southwestern Arkansas. We recorded F. harpi at all of the historic sites sampled and P. reimeri at all but one historic site. Although we did not visit all historical locations, historic sites sampled encompassed the (then) known range of both species. Our sampling scheme resulted in most (91% and 93% for F. harpi and P. reimeri) of the sampling sites to be near or in the right of way of primary, secondary, and tertiary roadways. Although most of the sampling was conducted in these highly altered habitats, our field surveys revealed new populations of both species. Both species of primary burrowing crayfish showed common patterns of habitat preference, which correspond with early accounts of both species preferred habitat (Hobbs 1979, Hobbs and. Based on this study, we recommend F. harpi retain its conservation status category of Vulnerable and P. reimeri be downgraded to Vulnerable based on . The range of F. harpi was expanded marginally by approximately 91 km 2 with five new populations, including one found in a county (Clark) not previously documented in its range ( Fig. 4.2). The range of P. reimeri was expanded by a larger margin, approximately 1404 km 2 , including 16 new populations. Once again, one of these populations was discovered in a county (Montgomery) previously not known in its range ( Fig. 4.2). The range of both species continues to be relatively small compared to most of the other primary burrowing crayfishes in the OME (e.g. F. jeanae, F. strawni, P. liberorum, and P. parasimulans). We are confident the ranges of F. harpi and P. reimeri are now accurately delineated in Arkansas. P. reimeri was captured near (<1 km) the state line of Oklahoma in Arkansas ( Fig. 4.2), leading to the conclusion P. reimeri may occur in Oklahoma. We recommend more field surveys to assess this possibility.
Both F. harpi and P. reimeri showed common patterns of habitat use. The habitat we found animals to be most abundant in was composed of treeless, wet seepage areas with an abundance of low grasses and sedges. The soil composition at these occupied sites was primarily loam and silt loam (Soil Survey Division Staff 1993), present in 92% of F. harpi quadrats and 94% of P. reimeri quadrats. The burrows of F. harpi and P. reimeri excavated in the field were complex, 0.5 to 1 meter in depth, and connected to groundwater. Our results confirm the specific importance of these wet, open canopy habitats as a preferred environment for both species.
The success in finding F. harpi at all of the historical sites sampled and five new populations indicates this crayfish is geographically constrained but relatively stable. Its total range is estimated to be 265 km 2 using a minimum convex polygon encompassing all known populations. The area within this range deemed as suitable habitat is estimated as 109 km 2 . This area was calculated by summing the area predicted above the lowest relative occurrence rate associated with any one of the presence localities from 2014 and 2015 from the Maxent analysis.
These results suggest this crayfish continue to be categorized as Vulnerable due to its restricted range in accordance with . The success of finding P. reimeri at all but one historical site and 16 new populations, one of which was greater than 50 km away from the previously known range, indicates this crayfish is somewhat geographically constrained but is more widespread than originally thought. Its total range is estimated to be 1467 km 2 using a minimum convex polygon encompassing all known populations. The area within this range deemed as suitable habitat is estimated as 178 km 2 . This area was calculated by summing the area predicted above the lowest relative occurrence rate associated with any one of the presence localities from 2014 and 2015 from the Maxent analysis. These results suggest P. reimeri should be categorized as Vulnerable as opposed to Endangered as dictated by .

Figure 4.2: Historical and current capture locations of Fallicambarus harpi (right) and
Procambarus reimeri (left) in southwestern Arkansas. Some triangles and squares represent more than one presence location.   My study of the habitat preferences F. harpi and P. reimeri revealed both to be habitat specialists, keying in on the open canopy, wet microhabitats. These habitats were similar to or within roadside ditches, and field observations showed the animals largely did not occur in the adjacent habitat. I conclude that these results support the benefit of the right of ways (ROWs) to the persistence of these narrowly endemic habitat specialists. My study on the ability of species distribution models to predict suitable habitat outside of the historically known range of both species revealed Maxent to be a suitable tool for this endeavor. The results of my study revealed new populations of both species, expanding the range of F. harpi marginally (<100 km 2 ) and the range of P. reimeri by a larger margin (>1000 km 2 ), both into counties from which they were not previously known. The conservation status of both crayfishes can be assessed with my data, and it revealed that while both F. harpi and P. reimeri are geographically constrained, they are locally stable in many roadside ditches. I conclude that the roadside ditch is functioning as suitable and preferred habitat within a matrix of unsuitable habitat for these two species within the Ouachita Mountains Ecoregion.