Applying mobile acoustic surveys to model bat habitat use across sinuous routes

Mobile acoustic surveys allow estimates of overall bat activity, relative abundance, and species richness across large areas. Protocols for estimating relative abundance recommend using non ‐ sinuous routes to ensure individual bats are only recorded once. We conducted mobile acoustic surveys along 12 sinuous routes in the mountainous terrain of northeastern Tennessee. Our objectives were to 1) determine if more calls were recorded in sinuous segments of mobile survey routes, thus violating assumptions of mobile surveys, and 2) analyze mobile transect data to assess bat habitat use. To test for effects of road sinuosity, we divided transects into ~1.6 ‐ km segments, calculated a sinuosity value, and summed the number of identified call files recorded for each segment. Using generalized linear models, we determined sinuosity did not affect the number of identified acoustic files. We assessed habitat use for 3 bat phonic groups (Low, Mid

USA; Fisher-Phelps et al. 2017) and others are less likely to be detected (e.g., Myotis spp. in South Carolina, USA; Neece et al. 2019). Some bats may avoid roads while others are attracted to them, particularly forest roads (Zurcher et al. 2010, Loeb and O'Keefe 2011, Bennett and Zurcher 2013. In addition, high-frequency echolocation calls attenuate more rapidly than low-or mid-frequency calls and, thus, high-frequency bats foraging along road sides may not be detected from a vehicle (Tonos et al. 2014). Braun de Torrez et al. (2017) noted that stationary surveys may be more effective for assessing composition of the local bat assemblage in places where road access is limited.
Wind and other noise associated with surveys from a moving vehicle could yield poor quality call files identifiable only to guild, but in some environments mobile surveys yield more identifiable calls (Tonos et al. 2014, Whitby et al. 2014). Even if call files are not identifiable to species, they can be classified to phonic group (Kaiser and O'Keefe 2015a) and are still useful for identifying important conservation areas (Luck et al. 2013) and population trends (Roche et al. 2011).
When the aim is to estimate relative abundance, Britzke and Herzog (2009) recommend conducting mobile acoustic surveys in a straight line and in one direction. However, in mountainous terrain, roads are often curved.
When the road is not a straight line, the assumption that one call file equals one individual bat could be invalid.
North American Bat Monitoring Program guidelines suggest that if a route is curved, points along opposite sides of these curved sections should be >100 m apart (Loeb et al. 2015). Ensuring survey points are >100 m apart is not always feasible in mountainous terrain. If we exclude sinuous roads, we may be unable to adequately assess trends, occupancy, and habitat relationships for bats using mobile transects.
We applied data from mobile acoustic transect surveys in the northern districts of the Cherokee National Forest in northeastern Tennessee to assess the relative effects of land-cover types on the habitat use of 3 phonic groups of bats. Many of the mobile transects were conducted on roads that climbed steep grades on the sides of mountains and contained many vertical curves. Thus, we tested the effects of road sinuosity on the number of call files recorded within transect segments during a mobile transect survey to determine if sinuous roads should be used in NABat monitoring. We predicted that road sinuosity within a segment would inflate the number of call files recorded. In northeastern Tennessee, there was a regular juxtaposition of agriculture, forest, and developed areas and there are few data on habitat associations for bats. Northeastern Tennessee hosts at least 12 species of bats (Appendix A), including several Myotis species, and low-and mid-frequency bats (Bernard et al. 2020). Although Myotis bats may be detected less frequently than other phonic groups overall, we predicted that these forestadapted, high-frequency echolocators would most often use forested areas O'Keefe 2011, Altringham andKerth 2016). In contrast, we predicted that bats in the low-frequency phonic group would be found more often in open areas (agriculture/pasture land; Lee andMcCracken 2002, Loeb andO'Keefe 2011) and in developed regions because big brown bats (Eptesicus fuscus) and Mexican free-tailed bats (Tadarida brasiliensis) often use anthropogenic roosts (e.g., buildings, bridges; Fenton 1986, McCracken et al. 2018). Mid-frequency bats, most likely to be eastern red bats (Lasiurus borealis) in our study area (Rojas et al. 2019), are adept at navigating in both cluttered and open areas (Furlonger et al. 1987, Elmore et al. 2005, Loeb and O'Keefe 2011; thus, we predicted we would find them using both forested and agricultural landscapes. We recorded a low number of calls per phonic group in each transect segment, so we modeled occupancy rather than abundance.

STUDY AREA
Our study was conducted in a 7-county region that encompasses the northern ranger districts of the Cherokee National Forest (NCNF) in northeast Tennessee (Figure 1). The 140,350 hectares of the NCNF are within the southern Appalachian Mountains. Mobile transect routes ranged in elevation from 241 to 1482 m (x ̅ = 679 m) above sea level and were composed of approximately 37% hardwood forest (no conifer component), 11% forest with conifer, 36% agriculture, and 11% development. The major forest type was chestnut oak (Quercus montana), with oak (Quercus)-yellow pine (Pinus subgenus Diploxylon) and poplar (Liriodendron)-oak components (Southeast Gap Analysis Project 2014; Appendix B). Agricultural land surrounding the forest was mainly hay pasture. Anthropogenic development was more prominent in the northern region, but most was low-to moderate-intensity development (Southeast Gap Analysis Project 2014); the population in most towns was <5,000, but the largest city, Johnson City, was just over 66,000 (U.S. Census 2011).
Five transects were in the southern portion (Unaka Ranger District) of the study area and 7 were in the northern portion (Watauga Ranger District). Transects were generally ≥6.5 km apart and did not intersect, excluding one transect that overlapped the end of another; we excluded the overlapping portion (11 km) from 1 of those 2 transects.
To create driving routes, we used ArcMap (v10.3.1;ESRI 2015), topographic maps, Garmin Basecamp™ software (v4.3.5, Garmin 2014), and Google Maps (Google 2014). We used U.S. Forest Service data to confirm roads were paved and accessible with low-clearance vehicles; we avoided sections of gravel longer than 1 km. We positioned routes to sample multiple land-cover types National Land Cover Data, 30-m resolution, USGS 2014. Use of Google Maps enabled us to cross-check routes before testing in the field, as sometimes these maps were the most up to date and highest image resolution of digital road views. After generating potential routes, we conducted trials during the day to assess the suitability, accessibility, safety, and length of the routes. Once we F I G U R E 1 Twelve mobile acoustic transects in a 7-county region in northeastern Tennessee, USA, were surveyed during June and July 2014-2015. One transect ended in Virginia. Each transect was surveyed twice each year, for a total of 4 surveys per transect. The inset map shows segment sinuosity along one transect; lower values are more sinuous and higher values are more linear. confirmed transect suitability, we uploaded the route onto a GPS (Garmin nüvi 2555LMT) using Garmin Basecamp™ software to aid in navigation when conducting the survey.
To assess summer bat populations, transects were surveyed twice each year for 2 consecutive years (2014)(2015) during June and July for a total of 4 surveys per transect. Surveys included pre-(June) and postvolancy (July) stages for juvenile bats born during the maternity season (Jones 2000). For all transects, we followed guidelines set forth by Britzke and Herzog (2009), who suggested the routes be driven at approximately 32 km/h and begin 30 minutes after sunset (start time~21:15 EDT) on evenings suitable for bat activity (no rain or fog, wind ≤24 km/h, temperature ≥12°C). Transects were driven in the same direction each time (Britzke and Herzog 2009).
Start and end points varied in terms of site characteristics but starting point characteristics were not the same for every transect. The second survey of each transect occurred approximately 3-4 weeks after the initial survey each year. We discarded data recorded when we pulled to the side to allow another vehicle to pass, or if we were off the transect due to a missed turn.
To record bat calls we used a Titley Anabat SD2 ultrasonic recording device (Titley Scientific, Columbia, MO, USA) with an Anabat Hi microphone mounted on the roof (oriented 0°from vertical and with the surface of the microphone 23 cm above the vehicle's surface). We connected and synced a GPS receiver (BR-355S4; USGlobalSat, Inc., Chino, CA, USA) to the Anabat to record location coordinates every second.

Acoustic analyses
We used Bat Call Identification software (v2.7c, Bat Call Identification BCID 2016) followed by manual vetting to identify recorded call files to phonic group. Preliminary examination of the data using BCID indicated we recorded few calls identifiable to species and, therefore, we used the phonic group designations produced by the software.
The software classifies call files mainly using 4 parameters: average minimum frequency (F min ), slope, average frequency at the knee of the call (F k ), and duration (C. R. Allen, BCID, personal communication). For bats in the lowfrequency group F c (characteristic frequency of the pulse, typically equal to F k , Gannon et al. 2004) was <30 kHz, for mid-frequency F c was 30-60 kHz, and Myotis were usually F c ≥ 40 kHz (Appendix A) with higher F min , steeper slope, and shorter duration than other phonic groups. To identify calls to phonic group, we required a 3-pulse minimum within 15 s, a 70% group confidence level, and minimum discriminant probability of 0.35 (Romeling et al. 2012, Kaiser andO'Keefe 2015a); otherwise, files were marked as unknown. Following BCID automated identifications, we manually vetted Myotis calls to confirm phonic group identification. To verify Myotis identifications, both V. Rojas and J. O'Keefe independently assessed all files marked as Myotis by BCID (Romeling et al. 2012, Braun de Torrez et al. 2017. We kept the Myotis designation for files which we agreed upon and marked all others as unknown. We excluded files that BCID could not clearly assign to phonic group from all analyses.

Environmental data
We recorded temperature (°C) and the Beaufort wind scale value (Barua 2005) at the start and end of each transect. We averaged start and end temperatures for the temperature detection covariate used in our occupancy analysis. We used ArcMap (v10.3.1; ESRI 2015) to reclassify National Land Cover Data into 3 dominant land-cover types (agriculture, development, forest; Appendix B). Road transect points were often classified as developed nonconcrete open space, but this designation ignored land cover adjacent to the road; hence, we excluded the developed non-concrete land-cover type from our analysis.
We divided each transect into 30 segments using the Split tool in ArcMap. Segments averaged 1.6 km, ranging from 1.2 to 1.9 km. All segments were used in our sinuosity analysis; however, to limit the influence of spatial autocorrelation in our occupancy analysis we systematically selected alternating segments for each transect (Russ et al. 2003). We added a 250-m buffer to each side of the segment and calculated the proportion of forest and agriculture within the buffer area (Appendix B). If alternating segment buffers overlapped due to a curve along the transect, we randomly selected one segment to exclude from the analysis, with a final count of 162 segments in our occupancy analysis. For each segment, we measured the distance from the center point of the segment to the nearest city or town boundary (m; Tennessee Department of Finance and Administration 2017) and nearest lake or river (m; hereafter, water; U.S. Geological Survey Hydrography 2013). All transect segments were relatively close to streams (average of 200 m away); therefore, we only included navigable lakes and river water bodies for the water covariate. We calculated the mean elevation (m) for each segment using GPS data points collected during the survey.

Sinuosity
Using the Calculate Sinuosity Python code add-on tool (ESRI 2011), we calculated the sinuosity value of every segment on every transect. We used generalized linear mixed models (lme4 package; Bates et al. 2015) in R (v3.3.2; R Core Team 2015) to test the effects of sinuosity on the number of acoustic files identified to phonic group. Due to inadequate call file numbers per segment (Low x ̅ = 2.6 call files per segment, Mid x ̅ = 2.1, Myotis x ̅ = 0.3), we did not analyze separate models for each phonic group and instead grouped all files for the sinuosity analysis. We compared 3 models for which the response variable was the number of acoustic files per segment (count); each model included transect as a random effect (intercept). In Models 1 and 2, the slope was set equal to 1. The simplest model, Model 1, had no fixed effects [count~(1 | transect)]. In Model 2, sinuosity was a fixed effect [count~sinuosity + (1 | transect)]. In Model 3, we modeled sinuosity as a fixed effect and with slope ≠ 1 [count~sinuosity + (sinuosity | transect)]. When we assumed a Poisson distribution in our models, we detected significant overdispersion and, hence, we ran our final models with a negative binomial distribution, which assumes conditional means ≠ conditional variance (Lindén and Mäntyniemi 2011). After running the models, we conducted an analysis of variance in R to test the 3 negative binomial distribution models against one another and present chisquared values to compare model significance.

Habitat use
We used occupancy models rather than conducting an analysis of relative abundance because most segments had ≤1 detection for each phonic group (Low, Mid, Myotis). For each phonic group, we created a binary detection history (1 = detected, 0 = not detected) for the segment and each of the 4 surveys. We considered a phonic group present along a segment if ≥1 call file was detected. Noting that bats might randomly occupy the transects we surveyed, we interpret modeled occupancy estimates as indicators of probability of habitat use (MacKenzie 2005). To facilitate comparisons of important predictors of habitat use among the phonic groups, we tested the same suite of probability of detection and occupancy covariates for each phonic group.
We developed hypotheses based on published literature (Furlonger et al. 1987, Jones 2000, Elmore et al. 2005 . We did not include other weather covariates, as we only conducted surveys during times of low wind and no precipitation. We used day of year as a detection covariate, expecting more detections once juveniles were volant later in the year (Jones 2000). We tested single-season occupancy models with year as a detection covariate to account for interannual variation. Initially, we tested year as an occupancy covariate, however year produced unreliable parameter estimates and lowered model weights; therefore, we excluded year from our final occupancy analyses. We did not detect any significant changes in available habitat from 2014 to 2015 and, thus, we did not expect occupancy to vary by year over our study. We chose single-season occupancy models rather than multiple-season models because our objective was to assess patterns of habitat use within season, we had limited temporal replication, and we were not interested in occupancy dynamics (i.e., extinction and colonization; Fuller et al. 2016). To test for covariate correlation, we used Spearman's rank tests (R v3.3.2; R Core Team 2015) and visually assessed plots. Detection covariates were not highly correlated (|r | < 0.2).
We used the following spatial covariates in our occupancy models: proportion of forest and proportion of agriculture in the 250-m buffer surrounding the segment, distance to lakes/rivers (m; water), distance to city or T A B L E 1 Ranked Akaike's Information Criterion (ΔAIC c ), model weights (w i ), and number of model parameters (K) for 6 detection models per phonic group for bat acoustic data collected during 12 mobile transect surveys in northeastern Tennessee, USA, June to July 2014-2015. We tested each detection model with a 3-variable occupancy (ψ) model containing proportion of forest in a 250-m buffer on each side of the road (for), distance to city boundary (m, city), and mean elevation (m, elev). Probability of detection (p) covariates were temperature (temp), day of year (date), and year. The null model lacked detection covariates.

Model
ΔAIC c w i K town (m; all ≤~66,000), and elevation (m). Two pairs of covariates were correlated: proportion of forest and agriculture (|r | > 0.49, P < 0.001), and distance to water and elevation (|r | > 0.5, P < 0.001). We did not include the correlated variables in the same models. We normalized all spatial covariates using the normalize function in program Presence (v.12.17, v.13.13;Hines 2016).
Using Presence (Hines 2016), we fit 6 probability of detection models, each with 1-3 detection covariates and a set of 3 non-correlated occupancy covariates (MacKenzie et al. 2018); we also tested a null model (Table 1). We required models to have substantial support (ΔAIC c ≤ 2.0) to be included in the plausible set. From plausible models, we identified informative detection covariates as those for which 85% confidence intervals for parameter estimates did not cross zero (Arnold 2010); informative covariates were used in occupancy models. For low-frequency bats, date was in both plausible detection models and temperature was only in one model (Table 1); given that warmer nights occur later in the summer, we opted to use only date as a detection covariate in occupancy models. Date and year were in both plausible detection models for mid-frequency bats, and both were retained as detection covariates. For Myotis bats, date was in all 3 plausible models and year was in 2 of 3 plausible models. There was model uncertainty and the effect of year was ambiguous; therefore, we only retained date as a detection covariate.
For each phonic group, we fit 10 probability of occupancy models with 1-3 covariates each, plus a null model (Table 3). We followed the same requirements described above for models to be included in the plausible set of models. To identify important factors predicting habitat use, we evaluated estimates for parameters in plausible models (Burnham and Anderson 2002;

RESULTS
We drove 2,328 km total in 73 hours of active surveying. Over the entire length of all transects, most call files recorded were in the low-frequency group (51% of all files), followed by the mid-frequency group (42%), and Myotis (7%; Appendix C). Transects T3, T2, T1, and T5 had the highest activity; all had ≥75 call files for at least one transect survey night or nearly 2 bat call files/km. Although Myotis detections were low overall, T1 and T2 had the highest Myotis activity (0.4 and 0.5 call files/km, respectively; Appendix C). On average, we identified 5 call files per segment (range 0-23).

Sinuosity
Segment sinuosity ranged from 0 to 1 (x ̅ = 0.77), with the segment being straighter as the value approached 1 ( Figure 1). Across all transects, the mode for sinuosity within a segment ranged from 0.53 (T5) to 0.95 (T4) and the mean sinuosity by segment ranged from 0.71 (T3) to 0.89 (T4). Sinuosity did not affect the number of calls recorded in each segment, as there was no significant difference between the model that did not account for sinuosity (Model 1) and models that did (Model 2 χ 2 (4) = 0.5472; Model 3 χ 2 (6) = 0.5053). As models did not differ, we did not consider sinuosity in the detection and occupancy analyses.

Probability of detection and use
Low-frequency phonic group For low-frequency bats, date had a clear, positive effect in both models ( Table 2). The probability of detecting low-frequency bats was higher later in the summer, and survey-specific detection estimates were 0.21-0.61 in the top-ranked occupancy model. Proportion of agriculture was a covariate in all 3 plausible occupancy models (Table 3) 8 of 20 | ROJAS ET AL. and was the only informative covariate (Table 4). As the proportion of agriculture within a 250-m buffer around the transect segment increased, the probability of use by low-frequency bats decreased (Figure 2A).

Mid-frequency phonic group
For mid-frequency bats, the estimated effects of date and year were similar ( Table 2). The probability of detecting mid-frequency bats was higher later in the summer and during the second year of surveys; detection estimates were 0.23-0.58 in the top-ranked occupancy model. All occupancy models had low AIC c weight (≤0.17) and there was uncertainty as to the best model, as 8 of 11 models were plausible based on ΔAIC c values, including the null model (Table 3). Further, none of the parameters were informative regarding habitat use by mid-frequency bats (Table 4).

Myotis phonic group
For Myotis bats, date had a clear negative effect (Table 2). In contrast to what we observed for low and midfrequency bats, we were less likely to detect Myotis bats on later survey dates within a season. Detection estimates for Myotis bats were 0.03-0.30 in the top-ranked occupancy model. We excluded 2 models (city and agriculture + water + city) from model comparisons and rankings due to convergence failures and negative standard errors; this problem likely arose due to the overall low detection rates for Myotis. Although there were 8 plausible occupancy models containing various parameters (Table 3), forest was the only informative parameter (Table 4).
The proportion of forest within a 250-m buffer surrounding the transect segment had a positive effect on habitat use by Myotis bats (Figure 2B).
T A B L E 2 Estimates, standard errors (SE), and 85% confidence levels for parameters in plausible probability of detection models (see Table 1). Date was important for all phonic groups as 85% confidence intervals for this parameter did not cross zero.  T A B L E 3 Ranked Akaike's Information Criterion (ΔAIC c ), model weights (w i ), and number of model parameters (K) for 11 occupancy models per phonic group for bat acoustic data collected during 12 mobile transect surveys in northeastern Tennessee, USA, June to July 2014-2015. We tested these occupancy (ψ) covariates using data for alternating transect segments: proportion of forest (for) and agriculture (ag) in a 250-m buffer on each side, distance to city boundary (m; city) and nearest lake or river (m; water), and mean elevation (elev). Day of year (date) was a detection (p) covariate in each model; year was also included in mid-frequency bat models. Null models lacked occupancy covariates. mid-frequency bats, habitat use followed our predictions based on ecomorphology (Norberg and Rayner 1987), but this was not true for low-frequency bats.

Model
Sinuosity of a transect segment did not affect the number of call files detected. Transects had varying degrees of sinuosity but dividing transects into segments allowed the issue of sinuosity to be addressed more precisely.
There is concern that surveys conducted on sinuous routes risk detecting an individual bat more than once, thus inflating the number of calls in sinuous segments and reducing the efficacy of mobile transects for estimating bat abundance in mountainous terrain Herzog 2009, Loeb et al. 2015). However, in our study, sinuous segments did not have inflated numbers of call files relative to straighter segments. We recommend additional testing of mobile transects along sinuous routes in other landscapes.
Probability of detection estimates along segments in top occupancy models were higher for low-and midfrequency bats than for Myotis bats. Date (day of year) was an informative detection parameter for all phonic groups. We expected detections to be higher later in the summer due to the presence of volant young of the year (Jones 2000). Low-and mid-frequency bat detections were higher later in the summer; however, we were less likely to detect Myotis later in summer. Lower detection rates for Myotis bats could be related to within-summer declines in Myotis bats, as was observed for a population of little brown bats (Myotis lucifugus) affected by white-nose syndrome (Reichard and Kunz 2009). Year also had a positive effect on detection of mid-frequency bats. There was no apparent explanation for the differences among years, as temperatures and captures of mid-frequency bats were similar between years during a concurrent study (average 21°C and 20 bats vs. 22°C and 23 bats in 2014(average 21°C and 20 bats vs. 22°C and 23 bats in and 2015(average 21°C and 20 bats vs. 22°C and 23 bats in , respectively, Rojas et al. 2019. In that study (2013)(2014)(2015), nearly all mid-frequency captures were eastern red bats (Lasiurus borealis, n = 94), except 4 eastern tri-colored bats (Perimyotis subflavus). We agree with Li and Wilkins (2014) and Loeb et al. (2019) that it may be important to conduct surveys at least twice each year to account for within-season variation in detection.
Complete transects passed through multiple land-cover types but dividing transects into segments allowed us to assess the effects of land cover on habitat use and to reduce spatial autocorrelation bias. We used an average segment length (1.6 km) that was equal to segment length for mobile surveys in Ireland (Roche et al. 2011); their approach was to survey 15 individual 1.6-km segments from a 24-km survey. Future studies could consider how varying segment lengths affect the perceived relationship between land cover and habitat use.
Factors that affected habitat use largely matched our expectations based on ecomorphology, although this was not the case for low-frequency bats. Capture data from a concurrent study showed that big brown bats were the most common low-frequency bat in our study area (Rojas et al. 2019). We expected that proportion of open area would have a positive effect on low-frequency bat use, but in fact low-frequency bats were less likely to occupy segments with higher proportions of agricultural land cover (mainly pastureland in this region T A B L E 4 Estimates, standard errors (SE), and 85% confidence levels (CI) for parameters in plausible probability of occupancy models (see Table 3). Agriculture was the only important parameter for low-frequency bats and forest was the only important parameter for Myotis, as 85% confidence intervals did not cross zero for either parameter. No parameters were important for mid-frequency bats.  (Brigham and Fenton 1986) and far from preferred patches of roosting habitat. Low-frequency bats may favor foraging along road corridors within forested areas because such edge habitat is uncluttered relative to the forest interior but offers similar insect abundance to the forest (Grindal and Brigham 1999).
As expected, mid-frequency bats, which were most likely to be eastern red bats in our study area, were detected across a variety of land-cover types; however, we did not identify any informative parameters regarding habitat use.
Eastern red bats are capable of foraging in a variety of land-cover types (Furlonger et al. 1987), with varying preferences dependent upon the landscape context. For example, in an intensively managed pine landscape in Mississippi, eastern red bats do not show a preference for stand type, given the options of young and open pine, closed canopy pine, thinned pine, unmanaged mixed pine/hardwood, and mature pine/hardwood (Elmore et al. 2005). In southwestern Ontario, eastern red bats are more active in forests and fields than over open water (specific feature types not specified; Furlonger et al. 1987), while in forests in Kentucky, eastern red bats forage over aquatic habitat (streams, ponds, and lakes) more often than expected (Hutchinson and Lacki 1999). Probability of habitat use by Myotis increased with higher proportions of forest in transect segments. Increased use was expected for these clutter-adapted species, as Myotis bats tend to forage in more forested areas (Henderson andBroders 2008, Loeb andO'Keefe 2011).
We were not able to use individual species identifications but instead based our analyses on phonic groups. In areas where bats are difficult to detect and calls may be poor quality (e.g., highly cluttered areas), the use of phonic groups may be a more feasible approach. The use of phonic groups allows the use of less conservative call identification settings (e.g., using a 3-pulse rather than 5-pulse minimum per call file) and enables a surveyor to quickly parse calls into groups. Our approach allows resource managers to quickly identify areas where a phonic group of interest is present, following up with acoustic, mist net, or roost surveys. For example, the majority of Myotis detections on our mobile surveys were aggregated along certain segments of just a few transects, mainly in the more forested southwestern region of the NCNF. Mist net surveys confirmed Myotis presence near some of these segments (Rojas et al. 2019), but detections along other segments point to potential sites for future mist net surveys.

MANAGEMENT IMPLICATIONS
Mobile transects are an efficient way to survey a relatively large landscape and can be used in areas with a mixture of private and public lands. Further, mobile transect surveys allow the estimation of relative abundance whereas individual point surveys do not (e.g., Evans et al. 2021). Estimating relative abundance may be impossible when bat detections are particularly low. In such cases occupancy modeling may be a viable alternative. When mobile acoustic surveys are conducted across a range of land-cover types, occupancy models can be used to assess general bat-habitat relationships.
To account for temporal variation in detection probability, we recommend that surveys be conducted both early and late in the summer, and across multiple years. Because sinuosity of the route did not affect the number of calls recorded in our study, it is possible that sinuous roads in mountainous terrain can be included in future NABat monitoring efforts.
Using mobile acoustic surveys to assess spatiotemporal patterns of habitat use by phonic groups or species in mountainous regions should aid in monitoring the status of bat populations and informing conservation decisions.

ACKNOWLEDGMENTS
Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government. The authors declare that they have no conflict of interest. We thank B. Walters for administrative support and these numerous field technicians for their dedication and assistance: J. Hoeh, A. Bender, J. Cox III, R.