Bats partition activity in space and time in a large, heterogeneous landscape

Abstract Diverse species assemblages theoretically partition along multiple resource axes to maintain niche separation between all species. Temporal partitioning has received less attention than spatial or dietary partitioning but may facilitate niche separation when species overlap along other resource axes. We conducted a broad‐scale acoustic study of the diverse and heterogeneous Great Smoky Mountains National Park in the Appalachian Mountains. Between 2015 and 2016, we deployed acoustic bat detectors at 50 sites (for a total of 322 survey nights). We examined spatiotemporal patterns of bat activity (by phonic group: Low, Mid, and Myotis) to test the hypothesis that bats partition both space and time. Myotis and Low bats were the most spatially and temporally dissimilar, while Mid bats were more general in their resource use. Low bats were active in early successional openings or low‐elevation forests, near water, and early in the evening. Mid bats were similarly active in all land cover classes, regardless of distance from water, throughout the night. Myotis avoided early successional openings and were active in forested land cover classes, near water, and throughout the night. Myotis and Mid bats did not alter their spatial activity patterns from 2015 to 2016, while Low bats did. We observed disparate temporal activity peaks between phonic groups that varied between years and by land cover class. The temporal separation between phonic groups relaxed from 2015 to 2016, possibly related to changes in the relative abundance of bats or changes in insect abundance or diversity. Temporal separation was more pronounced in the land cover classes that saw greater overall bat activity. These findings support the hypothesis that niche separation in diverse assemblages may occur along multiple resource axes and adds to the growing body of evidence that bats partition their temporal activity.


| INTRODUC TI ON
Diverse species assemblages theoretically partition along an indeterminate number of resource axes, but the nature and relative importance of these axes remain a hot topic in ecology. We typically study resource partitioning along three dominant resource axes: space, diet, and time. Assemblages may partition along these axes simultaneously (Hearn et al., 2018;Luiselli, 2006;Wilson, 2010), and more diverse assemblages should partition along multiple axes to maintain niche separation among all species. Temporal partitioning has received less attention than spatial and dietary partitioning and relatively few studies have attempted to understand species interactions across both time and space (Frey et al., 2017;Kronfeld-Schor & Dayan, 2003).
Given our theoretical understanding of temporal partitioning, we should expect bats to partition time; however, few studies examine temporal partitioning in bats and the results sometimes conflict. Temporal partitioning facilitates niche separation among sympatric predators, particularly where prey exhibit peaks of activity (Ramesh et al., 2012). Because insect activity is periodic within a night (Lewis & Taylor, 1965;Rydell et al., 1996), insectivorous bats are well-situated for temporal resource partitioning.
However, bats do not always exhibit clear patterns of temporal partitioning, and it may not be an important mechanism of niche separation where bats exhibit strong patterns of spatial separation (Arlettaz, 1999;Fenton & Rautenbach, 1987).
Under ideal conditions, temporal partitioning may be a strategy of flexibility rather than necessity, employed by species occupying similar spatial or dietary niches. For example, Lasionycteris noctivagans expands its temporal niche in the presence of Eptesicus fuscus (Reith, 1980), a species with similar ecomorphology (Norberg & Rayner, 1987) and diet (Whitaker, 2004), and Myotis sodalis shifts its foraging activity to earlier in the evening in the presence of the ecologically similar M. septentrionalis (Lee & McCracken, 2004). If temporal partitioning is a flexible strategy, its presence may not be apparent in small datasets, especially if we sample a subset of species assemblages at a few similar sites. Moreover, temporal partitioning may not always be a viable strategy for predators, as it relies on the temporal separation of their prey. It stands to reason that heterogeneous landscapes that support a diverse insect assemblage should provide greater opportunity for bats to partition temporally; however, this has not been formally tested, thus emphasizing the importance of scope and heterogeneity in temporal partitioning studies.
Large-scale disturbances can influence temporal partitioning in diverse ways. For example, understory-foraging frugivorous bats alter their temporal activity patterns in response to reducedimpact logging (Castro-Arellano et al., 2009). Likewise, changes in bat or insect assemblages could prompt bats to modify their spatiotemporal strategies. Recent evidence suggests white-nose syndrome (WNS), a fungal disease that has caused declines in several Myotis species, has caused shifts in resource use by bats.
For example, there is evidence that the spatiotemporal (Jachowski et al., 2014;Teets, 2018) niches of bats unaffected by WNS have relaxed in WNS-positive landscapes, presumably due to the decline of Myotis in those assemblages. These findings support the hypothesis that bats may exhibit flexibility in their spatiotemporal niches. However, neither of those studies examined the fine-scale patterns of temporal partitioning that occur across years and land cover classes.
We examined spatiotemporal partitioning in bat activity across a large area with a significant elevation gradient in the ecologically diverse Appalachian Mountains. We evaluated the relationship between land cover class, distance to water, and phonic group activity to assess spatial partitioning and quantified the degree of temporal overlap between phonic group pairs. We also investigated the consistency of patterns of spatiotemporal activity across the 2-year study period and examined how temporal partitioning varied with land cover class. We hypothesized that bats would partition time and space in this diverse landscape. Specifically, we predicted that Myotis and Low bats would be the most spatially and temporally separated given their distinct ecomorphology. Mid bats, with intermediate ecomorphology, should be more generalist in their spatiotemporal resource use. During this study, populations of four bat species in our area were in decline due to the effects of WNS (O'Keefe et al., 2019).
We predicted that we would observe changes in spatiotemporal resource use from 2015 to 2016, particularly along the temporal axis, due to decreased bat abundance and, therefore, competition. We predicted that phonic group activity patterns would be the most disparate in land cover classes with greater overall bat activity, given temporal partitioning is likely to increase with spatial overlap.

| Study area
We conducted this study in Great Smoky Mountains National Park (GRSM; 35°36′42″N, 83°29′22″W, Figure 1), a 211,000 ha protected area set in the Appalachian Mountains of the eastern United States, during the summers of 2015 and 2016. Approximately 97% of the park is vegetated and elevation ranges from 259 to 2,026 m.
The park additionally contains a variety of early successional openings (ES; 6%) unrelated to elevation, both natural and anthropogenic in origin. Climate also varies with topography. From May to August of 2015 and 2016, the high peaks received more rain (59 and 74 cm, respectively) than the lowlands (50 and 44 cm, respectively) and were 9°C cooler and 8%-14% more humid (NPS, 2019).

| Survey design
We conducted a comprehensive, spatially distributed acoustic bat survey of GRSM. We selected our survey sites using the Generalized Random-Tessellation Stratified method, as this method yields a spatially balanced, stratified design while allowing for substitution of unsuitable sites (Stevens & Olsen, 2004). The GRTS sampling was conducted using the spsurvey package within R and spatial data processing was undertaken in QGIS v2.8 (QGIS Development Team, 2015). We plotted 50 potential sites, with 50 oversample F I G U R E 1 Bats were sampled acoustically at 50 sites (ringed circles) within five counties in Great Smoky Mountains National Park in the Appalachian Mountains on the border of Tennessee and North Carolina, USA (see insert). Map shows dominant land cover classes: conifermixed hardwood (light green), northern hardwood (red), spruce-fir (blue), and early successional (pink). The black linear features represent sources of water within the park boundary sites, each of which belonged to 1 of 8 categories, defined by both land cover class (ES, CM, NH, and SF; Table 2 for descriptions) and proximity to permanent sources of water (third order and larger streams; near: <1,000 m from a stream or river; far: >1,000 m from a stream or river), although we treated distance from water as a continuous variable in our analysis. We constrained water sources this way because bats are more likely to use larger, calmer sources of water relative to smaller, faster-flowing sources of water for drinking and foraging (Razgour et al., 2010;Seidman & Zabel, 2001;von Frenckell & Barclay, 2011). We measured site characteristics with a laser range finder and densiometer and characterized vegetation using digital GRSM National Parks vegetation maps (Madden et al., 2004; Table 2). We used the detailed vegetation data to assign each stand to one of the four land cover classes described above: CM, NH, SF, or ES. We measured proximity to permanent water features (all linear) in the National Hydrography Dataset NHD Plus v2 (McKay et al., 2012). We ground-truthed all land cover class assignments and used the oversample to make substitutions when necessary (e.g., if site access was too dangerous), to achieve a total of 50 sites. Our dataset included some sites near each other, as eliminating these would have violated the GRTS method. The closest sites were 102 m apart, while the farthest were 83 km apart. Overall, spatial auto-correlation in nightly bat activity was low (Moran's-I = 0.05, p = .57).

| Bat surveys
We recorded bat echolocation calls with a Pettersson D500X detector for 2-15 nights/site (6.55 ± 2.69, mean ± SD) from May to August in 2015 and 2016, for a total of 322 survey nights. At any given time, we surveyed 1-7 random sites simultaneously. Of the TA B L E 1 Summary of bat phonic groups used in our study: species composition, their relative capture rate, ecomorphological indices (Norberg & Rayner, 1987), and foraging strategy Note: Foraging strategy is informed by Norberg and Rayner (1987) and Denzinger and Schnitzler (2013). Capture rate is based on 841 captures from 2015 to 2016 (O'Keefe et al., 2019, Carpenter, 2017. High aspect ratio (the square of the wingspan divided by wing area) is associated with long, narrow wings. High wing loading (weight divided by wing area) indicates a heavier body size relative to wing size. Slow, maneuverable flight is generally associated with low wing loading and aspect ratio. Fast, efficient, agile flight is generally associated with high wing loading and aspect ratio.

TA B L E 2 Dominant land cover classes in Great Smoky Mountains and some environmental characteristics of sites sampled
Land cover class  We deployed detectors on hiking trails or in early successional openings to standardize probability of detection across land cover classes and made the following assumptions in our analyses: that all bat passes were independent, all phonic groups had an equal probability of detection, and that bats were distributed randomly in vertical space. These assumptions are congruent with many acoustic studies (see Sherwin et al., 2000). We attached each directional microphone to its detector using a 7.5 m-cable, raised to a height of 3.0 m above the ground using PVC piping, and supported by a dowel rod extending 0.3 m away from the top of the post at a 5° decline from horizontal to prevent water from pooling on the surface of the microphone. We programmed our detectors to record from 30 min before sunset to 30 min after sunrise, according to data for a point in GRSM, 35°36′30″ North/83°56′09″ West and discarded any partial-night data. We did not deploy detectors when there was a strong chance of rain in the forecast. We measured average nightly temperature and relative humidity for the duration of each deployment using Honest Observer by Onset Pro units deployed in a weatherproof case within 10 m of the acoustic detector and deployed at the same height of 3 m.
We identified acoustic files using the SonoBat 3.2.2 (Szewczak, 2013) automatic classifier at its default settings. To address the difficulty of assigning species to acoustic calls (see Russo et al., 2018), we re-assigned all calls that were identified to the species or genus level to one of three phonic groups: Low, Mid, or Myotis. Phonic groups are closely related to clutter tolerance and foraging strategy in bats (see Table 1 for a summary). The Low group primarily contains open-space aerial hawkers; the Mid group contains edge-adapted aerial hawkers, and the Myotis group primarily contains narrow-space aerial hawkers (Norberg & Rayner, 1987).
Two bats do not fit these classifications: Myotis grisescens, an edgeadapted, trawling-hawking Myotis, and Corynorhinus rafinesquii, a gleaning Mid bat. It is unlikely this influenced our work, as M. grisescens is only rarely captured in the park (Table 1) and C. rafinesquii produce low-intensity echolocation calls that are less likely to be recorded (Fenton, 1982). We defined bat activity as the number of nightly bat passes. To examine the degree of temporal overlap in activity between phonic group pairs, we used a nonparametric kernel density estimation procedure (Linkie & Ridout, 2011;Ridout & Linkie, 2009).

| Statistical analysis
Following the methods outlined in Ridout and Linkie (2009), we converted the timestamp of each acoustic file to radians and used a kernel density estimation to generate a probability density distribution for phonic group pairs. We used the overlap package in Program R (Meredith & Ridout, 2014) to generate an overlap term from the mutual area under the two activity curves (Δ 4 ), ranging from 0 (no activity overlap) to 1 (total activity overlap). We calculated 95% confidence intervals for all estimates from 1,000 bootstrap resamples, using them to determine whether the degree of overlap between phonic group pairs varied with year; if estimates did not overlap, we considered the degree of temporal overlap significantly different.
We used the same nonparametric kernel density estimation procedure to explore temporal activity trends among the three phonic groups in different land cover classes. We quantified temporal overlap using the aforementioned methods but report the average Δ 4 for each land cover class (calculated by averaging the overlap values for each of the three phonic group pairs). This method prevented us from calculating confidence intervals but made interpretation more succinct by eliminating 66 pairwise comparisons.

| Evidence for spatial partitioning
Year, land cover class, and proximity to water were important pre- Despite their similarities-larger size and lower predicted clutter tolerance-the Low and Mid bats used space differently. Neither year, land cover class, nor proximity to water were important predictors of Mid bat activity, as none of the models outperformed the null model (Figures 2b and 3b). However, year, land cover class, and proximity to water were important predictors of Low bat activity.

The most plausible model for Low bats included year as an interac-
tive effect with land cover class and water availability. Depending on the year, Low bat activity was the greatest overall in either conifermixed hardwood forests or early successional openings (Figure 2c).
Low bat activity was lowest in either northern hardwood or sprucefir forests (Figure 2c). In 2015, Low bat activity was 64% greater in conifer-mixed hardwood forests than in early successional openings, 54% greater in early successional openings than in spruce-fir forests, and 177% greater in spruce-fir forests than northern hardwood forests (Figure 2c). In 2016, Low bats shifted to using early successional openings more than the forested land cover classes. Low bats were 156% more active in early successional openings than conifer-mixed hardwood forests, 15% more active in conifer-mixed hardwood forests than northern hardwood forests, and 26% more active in northern hardwood forests than spruce-fir forests (Figure 2c). Holding other variables at their means, Low bat activity was up to 14 times lower at sites 3,000 m from water versus sites 0 m from water; proximity to water was a better predictor of Low bat activity in 2015 versus 2016 (Figure 3c).

| Evidence for temporal partitioning
Temporal patterns in phonic group activity varied across the night and years, with differential activity peaks and troughs driving All phonic groups shifted their mean nightly activity patterns from 2015 to 2016, but not to the same extent. We observed the greatest activity shift in Low bats, followed by Mid, and then Myotis bats. Broadly, the degree of temporal overlap between each phonic group pair was greater in 2016 ( Figure 4). In 2016, the primary Low bat activity peak still occurred 2-3 hr after sunset, but it was much less dramatic, and Low bats remained relatively active throughout the night (Figure 4e,f). In 2016, there was no discernible primary peak in Mid bat activity, although activity pulsed 3, 6, and 10 hr after sunset. This final pulse corresponded with a period of reduced Low bat activity (Figure 4e). Myotis also exhibited no discernible primary peak in activity, but activity pulsed approximately 2, 9, and 11 hr after sunset (Figure 4d

| D ISCUSS I ON
In a diverse and heterogeneous landscape, bat activity varied in both space and time in ways consistent with the prediction that diverse assemblages should partition resources along multiple resource axes to coexist. Myotis and Low bats were the most spatially and temporally dissimilar, while Mid bats employed a more generalist spatiotemporal strategy (Figure 6). Phonic groups changed their temporal and, to a lesser extent, spatial resource use across the 2- year study. Notably, we found evidence for relaxation in the temporal niches of all phonic groups from 2015 to 2016. The degree of temporal separation between bat phonic groups varied across the landscape-temporal partitioning was more apparent in the conifermixed hardwood forests and early successional openings, which also saw greater overall bat activity. Overall, these findings support the hypothesis that niche separation in predator assemblages may occur along multiple resource axes, adding to the growing body of evidence that bats partition their temporal activity, and reinforcing the notion that temporal partitioning affords bats the flexibility to respond to environmental cues.
Low bats used low-elevation conifer-mixed hardwood forests and early successional openings near water, consistent with expectations based on ecomorphology and previous research. Many bat species in the Low phonic group are predicted to have lower maneuverability in cluttered environments (Aldridge & Rautenbach, 1987;Norberg & Rayner, 1987), implying increased foraging efficiency in open or sparsely vegetated sites. This is congruent with our finding that Low bats were more active in early successional openings and conifer-mixed hardwood forests, which have more vertical space and are less cluttered than the northern hardwood forests and spruce-fir forests. Eptesicus fuscus (the most commonly captured Low bats at our study site) are often associated with uncluttered sites (Brooks  , while L. cinereus and Myotis, but not E. fuscus, partition vertically in the boreal forests of Canada (Kalcounis et al., 1999). The prevalence of vertical partitioning likely depends on a variety of local conditions, including forest structure and prey availability.
Land cover class and water availability were not strong predictors of Mid bat activity in this landscape, which may result from the inclusion of a common generalist bat, L. borealis, the eastern red bat, in this phonic group or from limitations in our study design.
L. borealis do not exhibit clear patterns of landscape-level habitat selection (Carter, 1998;Elmore et al., 2005), except to avoid urban development (Walters et al., 2007). Where evidence of selection is present, L. borealis select for linear features that facilitate flight, including roads, ridgetops, and streams (Amelon et al., 2014), rather than specific land cover classes. Though our findings agree with the general trends of published literature, they may be biased because we deployed our detectors on trails. Alternatively, any evidence of spatial partitioning within the Mid bat group may be obscured by the fact that some of the bats within the Mid bat phonic group are ecologically dissimilar (Table 1).
Myotis were comparatively distinct in their spatial activity patterns, avoiding early successional openings in favor of all the forested land cover classes. The apparent preference of Myotis for forested land cover classes matches expectations in light of their ecomorphology-short, broad wings, low wing loading, and broadband, high-frequency calls (Aldridge & Rautenbach, 1987;Norberg & Rayner, 1987). Several Myotis species that occur in our study area exhibit strong selection for forested land cover (e.g., M. leibii : Johnson et al., 2009;M. septentrionalis: Henderson & Broders, 2008;M. sodalis: Menzel, Ford, et al., 2005;Sparks et al., 2005) and respond negatively to fragmentation or loss of forests (M. septentrionalis: . However, the use of openings by Myotis species is less clear, with Myotis avoiding them in some landscapes but selecting for them in others. Myotis activity is often greater in intact forest patches than clear-cuts (Owen et al., 2004;Patriquin & Barclay, 2003) or forest edges (Morris et al., 2010), and there is evidence that M. sodalis avoid agricultural fields and pastures when commuting from roosting to foraging habitat (Murray & Kurta, 2004).
However, small natural gaps may provide important foraging opportunities for Myotis, particularly within densely stocked or regenerating forests. For example, M. septentrionalis are more likely to be recorded at sites with low or medium density vegetation than sites with dense vegetation (Loeb & O'Keefe, 2006).  Farney and Fleharty (1969). The characteristic frequencies and forearm lengths represent a range of means for the bat species contained within each phonic group. Characteristic frequencies were informed by Szewczak (2013), and forearm lengths were produced from our capture database (O'Keefe et al., 2019). Spatial and temporal overlap charts were generated from the results of this study and the characteristics of the opening (Brooks et al., 2017). Because clutter-adapted Myotis are not restricted to foraging in complex or cluttered habitats, we might expect Myotis to exhibit some flexibility in their foraging strategy, moving into early successional openings to take advantage of resource pulses in an environment where flight costs should be comparatively low. However, our data suggest this may be a rare occurrence in this landscape. Insect productivity is 2.5 times higher in the forests than the grassy balds of this landscape (Whittaker, 1952) so forested sites may, on average, provide more insect-rich foraging opportunities than early successional openings.
Insect productivity is also 1.4 times greater in mid-versus lowelevation forests and 2 times greater in high-versus low-elevation forests (Whittaker, 1952), suggesting the mid-elevation northern hardwoods and high-elevation spruce-fir forests may indeed be quality foraging habitats for bats that can effectively forage in these more cluttered forest types. Myotis, being adept at flying in clutter, may preferentially exploit these accessible and insect-rich forests with little need for early successional openings. Conversely, the less clutter-adapted Low bats may be biomechanically constrained from efficiently foraging in the cluttered, high-elevation forest types (Mayberry et al., 2020 which also saw the highest overall levels of bat activity (Figure 5a,b).
Temporal segregation was comparatively low in northern hardwood and spruce-fir forests, which saw low overall bat activity ( Figure 5c,d). These observations support the hypothesis that temporal partitioning may be a flexible strategy employed to minimize interspecific competition in more crowded environments (Adams & Thibault, 2006;Razgour et al., 2011), though it bears mentioning that the ubiquity of interspecific competition in bat assemblages is unclear (Salinas-Ramos et al., 2020). The temporal separation we observed could instead reflect the variable activity patterns of the preferred prey of each phonic group (Rydell et al., 1996). For example, low-frequency bats prey more heavily on beetles (e.g., Clare et al., 2014b;Wray et al., 2021), while Myotis mainly depredate small-bodied moths and flies (e.g., Clare et al., 2014a;O'Rourke et al., 2021).
The prevalence of temporal partitioning in bat assemblages could easily be underestimated if spatiotemporal context is not considered. Overall bat activity peaked just after sunset and each phonic group was active during this period; however, pulses of activity by phonic group were more segregated after this initial activity period. Consequently, partial-night surveys may be insufficient to study partitioning in bat assemblages. Additionally, temporal partitioning was inconsistent across years, land cover classes, and phonic group pairs, emphasizing the importance of scope in temporal partitioning studies; by narrowing the scope of a study to a few nights, sites, or species, we would likely miss temporal partitioning. These factors may explain the lack of temporal partitioning in several studies (e.g., Adams & Fenton, 2017;Arlettaz, 1999;Fenton & Rautenbach, 1987). However, scope is not enough, as we might wrongfully conclude from our dataset that this assemblage does not partition their temporal activity had we simply quantified the degree of overlap across all years and land cover classes, as the average degree of overlap was relatively high (65%). Context was important too, as temporal partitioning was more dramatic in 2015 than in 2016 and certain land cover classes. The variable and contextual nature of temporal partitioning in bat assemblages may provide an excellent system for studying the conditions and mechanisms that lead to temporal partitioning in species assemblages.
Temporal segregation, which relaxed from 2015 to 2016, was more flexible than spatial segregation which may be due to a range of dynamic processes. We may have observed niche relaxation associated with a change in the competitive landscape from 2015 to 2016, such as might be caused by the decline of one or more species from an assemblage. For example, there is evidence that the spatiotemporal (Jachowski et al., 2014;Mayberry et al., 2020;Teets, 2018) and  Schoener, 1974), and given the mounting evidence that insects are declining (Wagner, 2020), bats may be expanding their niches through time in response to declines in prey availability. Either scenario is outside the scope of our 2-year study. It is equally possible these shifts were driven by annual fluctuations in weather or insect availability.
This work has important implications for how we survey bat populations-namely that partial-night surveys may not capture a representative sample of the local bat assemblage. It is a common practice in mist-net and driving transect surveys to exclusively sample the first 1-5 hr following sunset (Loeb et al., 2015;U.S. Fish & Wildlife Service, 2020). However, if Myotis are more active in the latter part of the night, then these approaches may miss the majority of Myotis activity (Johnson et al., 2011).
Temporal partitioning among bats may be more common than previously thought, and this phenomenon is worth exploring in other settings, ideally with more sampling locations and over a longer period. Multi-night, passive acoustic studies are relatively easy to implement and lend themselves to studying spatiotemporal trends in activity (Frick, 2013;Obrist, 2020). We recommend researchers use their acoustic datasets to examine patterns of spatiotemporal resource use to determine how widespread temporal segregation is and to describe its role in structuring bat assemblages. If temporal partitioning is revealed to be widespread among bats, they may serve as model organisms for studying temporal partitioning-its importance relative to other forms of partitioning, and the conditions that spur temporal segregation (Kronfeld-Schor & Dayan, 2003).

CO N FLI C T O F I NTE R E S T
None declared.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available