Water and elevation are more important than burn severity in predicting bat activity at multiple scales in a post-wildfire landscape

Bats are among the most widespread mammals on Earth, and are subject to habitat change, loss, and other disturbances such as fire. Wildfire causes rapid changes in vegetation that affect habitat use. However, the spatial scale at which these changes affect bats depends on their use of habitat elements. Three years post wildfire, we assessed how burn severity, water, landform type, elevation, vegetation type, and roads affected use by bats of a forest landscape at multiple spatial scales. We deployed acoustic detectors at randomly selected locations within a 217,712 ha wildfire boundary in Arizona. We classified echolocation calls to species or group and calculated an activity index by adjusting the calls per hour. We conducted a multi-scale analysis of landscape structure and composition around each location from a 90 to 5760 m radius. No scale was selected preferentially by any species or group. Stream density and elevation range were more important predictors for species groups than burn severity. When burn severity was a predictor, agile species had higher activity in areas that were unburned or had low severity burn. A heterogeneous landscape composed of high, medium, and low burn severity patches within a forest altered by large wildfires provided habitat for different bat species, but water density and range in elevation were more important for predicting bat habitat use than fire severity in this arid landscape. More than one spatial scale, representing local to landscape levels, should be considered in managing habitat for bats. In arid areas, such as the western United States, maintaining reliable water sources is important for bats. Managing these factors at multiple spatial scales will benefit bat species with different wing morphologies, echolocation call types, and habitat selections.


Introduction
The diversity of bats, with their nocturnal, volant, and cryptic behaviors, make this taxon challenging to study. Yet bats are among the most widespread mammals on Earth [1], and are a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 were better adapted to habitat use at fine, local scales since they will generally travel farther across the landscape in a night compared to more agile bats (Fig 1) (e.g., [37,38]).

Study area
We conducted our study from 11 June to 7 August 2014 within the perimeter of the Wallow Fire (Fig 2; NAD83, 12S 643903E, 3719106N) on the Apache-Sitgreaves National Forests in southeastern Arizona, USA. The Wallow Fire burned 217,721 ha in 2011 [39]. We surveyed bats in the ponderosa pine vegetation type [40], at sites that ranged in elevation from 2350 to 2690 m. Daily temperature and precipitation averaged 17.0 ± 0.3˚C and 3.5 ± 1.1 mm, respectively, during the study period in the town of Alpine, 12 km from the center of the study area [41]. During the peak in precipitation in late summer (monsoon season; Jul-Aug), daily precipitation averaged 5.6 ± 2.1 mm in contrast to the pre-monsoon season in early summer (May-Jun) when it averaged 1.6 ± 0.8 mm.   [42] which was determined using a relative differenced Normalized Burn Ratio (rdNBR) image from the Monitoring Trends in Burn Severity project [43]. https://doi.org/10.1371/journal.pone.0231170.g002

Acoustic recording
We placed SM3BAT acoustic detectors (Wildlife Acoustics, Maynard, Massachusetts, USA) at 4 points in each of 21 blocks (n = 84 points; Fig 2). Each block averaged~5 km 2 and consisted of one point in each of four burn severity classes, which we determined using the Rapid Assessment of Vegetation Condition after Wildfire (RAVG) layer [44]. This layer used immediate basal area (BA; the sum of the cross-section area of trees at breast height [45]) loss to determine burn severity [44]. Our four burn severity classes included 0-25%, 26-50%, 51-75%, and 76-100% BA loss within a 500 m radius from the acoustic detector point. We used ArcGIS 10.2.2 (ESRI, Redlands, CA, USA) to select points with the following criteria: �1 km from points in other blocks, �1 km from water sources (to avoid bias from bats drinking or foraging over water), �1 km from cliffs and buildings (to avoid bias from bats roosting in these structures), �50 m from roads to reduce risk of equipment loss or damage, and >50% of the area within a 500 m radius around each point was in one of the four burn severity classes.
We randomized blocks so that each point was surveyed twice for three consecutive nights each, once pre-monsoon (11 June to 11 July) and once during the monsoon season (12 July to 7 August). We used omnidirectional SM3-U1 microphones for each detector, which we connected to the SM3BAT detector with a 3 m cable and placed at the top of a 2.5 m pole. We tested each microphone to make sure that the sensitivity was similar across each detector unit. The microphone protected with a soft windscreen and was connected to the pole at a 45-degree angle so that water would drop off instead of pool in the microphone. We used four D-cell batteries to power each of the detectors. We deployed each detector in the center of a forest clear-ing~30 m in diameter to avoid interference during acoustic recording from vegetative clutter. We recorded location, distance to forest edge (m), temperature (˚C), relative humidity (%), and precipitation during deployment. We programmed detectors to record bat calls from sunset to sunrise, and during each 3-day period we concurrently surveyed two randomly-selected blocks.
Each detector recorded 5-sec .wav files when bats were detected with the microphone. We classified each 5-sec file with �2 pulses as a call sequence. We identified each echolocation call sequence using SonoBat 3.2.1 (SonoBat™, Arcata, CA, USA). In the SonoBat Batch Classify tool, we used a decision threshold of 0.95, acceptable call quality of 0.80, acceptable quality to tally passes of 0.20, and maximum number of calls to consider per file of 8. Since characteristic call frequencies overlapped for some species (Table 1), making differentiation unreliable at species level, we placed calls into five groups using SonoBat classification: low frequency (F c <33 kHz), high frequency (F c �33 kHz), Myotis, fringed myotis (Myotis thysanodes), and Mexican freetailed bat (Tadarida brasiliensis) plus hoary bat (Lasiurus cinereus) calls (Table 1). Many bat acoustic studies use high and low frequency groups (e.g., [46,47]), or put calls from the genus Myotis into one or two groups [e.g., [48,49]) due to the difficulty in species discrimination. We felt confident in reliably separating fringed myotis from other Myotis species because of its low characteristic frequency (F c <33 kHz) and short, broadband call compared to other species (e.g., [24]). We felt confident in separating Mexican free-tailed bat and hoary bat from other low frequency species because of their lower characteristic frequencies and call shape, but we were not confident in separating those two species. However, both species are open-air flyers with high wing loading and aspect ratio (Table 1). We averaged calls per hour by group for each point for each 3-day period. The average calls per hour for each of the groups were our response variables.

Landscape variables
We used a moving window analysis in ArcGIS 10.2.2 (ESRI, Redlands, CA, USA) to determine values for burn severity, elevation, range in elevation, land cover, landform type, stream  ), aspect ratio, wing loading, and our prediction of the response to increasing burn severity (Prediction; + = positive response,-= negative response) for bat species occurring in ponderosa pine forests on the Apache-Sitgreaves National Forests, Arizona, USA.  Table 2). We selected these scales because they represented local (e.g., 90 to 720 m) to landscape (�1440 m) habitat use by bats [e.g., 34,51]. We obtained Composite Burn Index (CBI) values [42] post-fire to determine burn severity using a relative differenced Normalized Burn Ratio (rdNBR) image of the Wallow Fire area from the Monitoring Trends in Burn Severity project [MTBS;43]. This provided a continuous value for burn severity. We used the National Land Cover Database [52] to determine land cover. We reclassified land cover into 2 categories: evergreen forest and grassland/shrub/scrub where evergreen forest was used as the reference category. We combined grassland, shrub, and scrub land covers because they were all open land cover types and, individually, they did not make up a large portion of the landscape. We used a landform data layer that consisted of 15 landform classes to determine the landform type [53]. We reclassified landform types into 2 categories: upper slopes/ridge and lower slopes/valley where upper slopes/ridge was used as the reference category. We calculated lake, stream, and road densities using data from the Apache-Sitgreaves National Forests [54]. All continuous values were standardized to z-scores before analysis, and we tested variables for collinearity in program R with Spearman's rho statistic.

Statistical analyses
We used a 2-step process to predict response of each bat group to covariates at the most appropriate scale. We first performed a univariate analysis to determine the best fitting scale for each landscape variable. We used linear mixed model regression with season (i.e., pre-monsoon or post-monsoon) as a random variable in program R using package lme4 (R statistical program v. 3

Results
We collected and classified 63,990 files as bat calls. Of these, 69, 31, 7, 1, and 14%, respectively, were classified as low frequency, high frequency, Myotis, fringed myotis, and Mexican freetailed bat/hoary bat. In the univariate analysis, we found no scale was selected preferentially, therefore scales varied in the multivariate models (Fig 3).
Competitive models consisted of covariates ranging from fine to coarse scales. Scales selected for the low frequency and high frequency groups ranged from 360 m to 5760 m and 90 to 5760 m, respectively (Table 3, Fig 3). Scales selected for Myotis, fringed myotis, and Mexican free-tailed bat/hoary bat ranged from 180 m to 5760 m, 90 m to 2880 m, and 90 m to 5760 m, respectively (Table 3, Fig 3).
The group consisting of low frequency calls responded positively and strongly to stream density and elevation range, and positively and moderately to burn severity (Table 3). Bats in this group selected areas of high burn severity with greater range in elevation and higher stream density (Fig 4a).
Bats with high frequency calls had a strong response to elevation range, landcover, road density, and landform, and responded moderately to burn severity (Table 3). This group responded positively to range in elevation but negatively to burn severity and road density. Bats with high frequency calls selected evergreen forest over shrub/scrub/grassland landcover and lower slopes/valleys over upper slopes/ridges for landform (Fig 4b).
Calls classified as Myotis had a strong response to landform, stream density, road density, landcover, and elevation range (Table 3). Myotis responded positively to elevation range and negatively to stream density and road density. This group selected lower slopes/valleys over upper slopes/ridges for landform and selected evergreen forest over shrub/scrub/grassland landcover (Fig 4c).
Mexican free-tailed and hoary bats responded strongly to stream density, elevation range, and burn severity, moderately to landcover and road density, and weakly to landform (Table 3). These species responded positively to stream density, elevation range, burn severity, and road density. This group selected shrub/scrub/grassland landcover over evergreen forest and lower slopes/valleys over upper slopes/ridges (Fig 4d). Fringed myotis had a strong response to burn severity, lake density, stream density, and elevation range (Table 3). Fringed myotis responded positively to lake density and stream density but negatively to burn severity and elevation range (Fig 4e).

Discussion
Burn severity was important in predicting bat activity in a post wildfire landscape for one species, but other landscape variables were more important for predicting activity of bat groups that included �2 species. Fringed myotis is a small-bodied bat with low aspect ratio and low wing loading, and our results suggest that this species avoids large areas of high severity burn. Jemison et al. [23] also found that small-bodied bats with low aspect ratio and wing loading were negatively affected by wildfire compared to species with big bodies and long, narrow wings. Bats with lower wing loading and a lower aspect ratio are better suited to maneuvering around and gleaning prey from objects than species with high wing loading and aspect ratio [35,58,59]. However, Buchalski et al. [24] found the opposite with Myotis activity higher in areas of high burn severity. Buchalski et al. [24] conducted their study in areas with only moderate-to high-severity burn one-year post fire in mixed conifer forest. Jemison et al. [23] conducted their study two-years post fire, and our study was three-years post fire. Some species of insects increase in abundance directly after a fire [60,61] therefore increasing prey density in burned areas for bats the year immediately after the fire. Brigham et al. [62] found that smallbodied, maneuverable bat species would still use open habitat if that habitat was higher in prey density than more closed habitat. Morphology and call frequency can predict habitat selection, but bats use the landscape differently over time depending on food availability. Burn severity on the landscape helped to predict activity for one bat species, but other landscape variables were key for other bat species groups. Water density was more important than burn severity for most bats in our study. Riparian ecosystems are uncommon in the western United States (e.g., only 0.4% of the land base in Arizona; [63]). Reproductive females especially need water and had higher capture rates at water sources than males or non-reproductive females [27,28,64]. Generally, higher stream density on the landscape benefited bats in our study area, through either increased prey density or availability of drinking water. However, Myotis group selected against increasing stream density at a local scale. At more local spatial scales, stream density may be less important because many streams in this area are intermittent, often fast moving, or covered with vegetation. Bats fly over still water to drink during flight and avoid running water, presumably because of the difficulty and risk of obtaining water from streams or rivers while drinking on the wing [65][66][67]. Human-modified water sources, such as livestock ponds, are common in the southwest and might be more important than natural streams in this area because they are more reliable and non-flowing [68,69]. Myotis are small and have a higher surface area to weight ratio than other bat species in our study. These small bats need reliable water sources to remain hydrated in this arid environment.
Selection for lower slopes and valleys could be associated with elevation range and temperature regulation for reproductive females [70,71]. There was high activity of agile flyers when there was a greater range in elevation over a coarse landscape scale. When females care for young during the summer, they need to roost at lower, warmer elevations, but may choose to drink and forage at higher elevations. In the Black Hills and in parts of the eastern United States, reproductive females were captured less often at high elevations than low elevations compared to males and non-reproductive females [27,70,71]. There were also fewer reproductive females captured at high elevations than low elevations compared to males and nonreproductive females in our study area [64]. A range in elevation on the landscape can support reproductive females as well as non-reproductive females and males. However, we found that one species, the fringed myotis, selected against elevation range at a fine scale. Although Myotis move between roosts frequently [72], fringed myotis do not move far when changing roosts, showing fidelity to a roost area and not to a particular roost tree [73][74][75]. Fringed myotis may be searching for roosting opportunities in close proximity to other roosts, which could facilitate the movement of young to new roosts [76]. Fringed myotis was our only single-species group, and the diverse Myotis group could have selected a wide range in elevation because it consists of more than one species where each species uses a different range in elevation. Because we grouped all Myotis together, the overall range in elevation becomes important.
Selection of a range in elevation may also be related to food availability. McCoy [77] found higher species richness of insects at mid elevations in the United States, suggesting that a greater range in elevation provided more availability for foraging for insectivorous bats. Insects may also be more abundant close to water sources [e.g., 62,27], meaning more prey for bats in riparian areas. Insects were influenced by cattle grazing in southeastern Arizona and were less abundant in grazed grasslands [78]. This may be similar for our study area and grazed grasslands/shrub/scrub provided less prey for bats than evergreen forest.
Roads provided some bat species with travel and foraging areas but other species avoided roads. Forest roads provide open habitat for low frequency bats to use as flight corridors or as open areas around a roost in an otherwise forested area [e.g., 31], and in our study low frequency bats had high activity in areas of high road density at a fine scale. Smaller species that echolocate at higher frequencies, like Myotis, responded negatively to road density, which suggests that these species avoid open spaces such as roads. Myotis and other clutter-adapted species fly closer to the ground over roads than high, open-air flying species and are more likely to be killed at roads [79,80]; roads cause barriers to some bats because of habitat fragmentation [81,82].
Important relationships in how a species selects its habitat will be missed if only one spatial scale is considered [83][84][85]. For bat groups in our study, no scale proved best; the habitat element of interest affected scale importance. This was also true for the single species group in our study, fringed myotis. This indicates that other bat species could also respond to habitat elements across a range of scales. This variability in scale may be attributed to the mobility of bats; they use habitat elements at scales ranging from roost to foraging area. A heterogeneous landscape increases roosting opportunities and variability in foraging opportunities [86].
Studying bats with acoustic devices is less labor intensive than capturing animals, can effectively sample bats that fly above the height of most nets (>7.8 m; [87,88]), detect individuals that learn to avoid nets [89][90][91], and sample more efficiently (e.g., more locations can be sampled than with capture techniques). However, habitat structure [92], temperature, and humidity [93] can affect how sound moves in the environment, which influences detectability of the bat call [88], and acoustics cannot determine individual information (i.e., sex, age group, reproductive condition, species for locations like the southwest United States) or abundance. Despite these limitations, our study is one of few that identified responses of bats to wildfire.

Conclusions
We found that multiple spatial scales were important in determining habitat use. Managing forests for heterogeneity benefits a diversity of bat species. Although burn severity did negatively affect one bat species, water and elevation range were most important, at least in the immediate post-fire years. We might expect the importance of habitat elements such as water (e.g., water use) to differ in other ecosystem types since our study took place in an arid environment. In areas where water is more abundant, burn severity might have a greater effect. Burn severity could also be more important for individual species, and grouping species together in analyses reduces the importance of burn severity. However, maintaining and protecting slow-moving water, like livestock ponds, is important for bats, especially in the arid western United States. Some species groups opportunistically used forest roads, so although some bats responded negatively to roads, roads were not always negative for other bats. Landscape scales of �5760 m should especially be considered when managing habitat for bats that can fly long distances nightly from roosts to foraging areas. The largest spatial scale should be the home range or maximum nightly flight distance for the bat species of interest.
Supporting information S1 Appendix. Candidate model list. The log likelihood (LogLike), AICc, ΔAICc, and weight (w i ) for each of the candidate models with a ΔAICc �4 and the null model for each species group. (DOCX)