Island biogeography theory and the urban landscape: stopover 2 site selection by the silver-haired bat ( Lasionycteris noctivagans )

Abstract


Introduction
Animals living in seasonal environments must cope with a spectrum of challenges throughout the year.When faced with seasonal resource limitation, many species respond by migrating to more favourable conditions (Dingle 2014).Consequently, migratory species occupy distinct seasonal ranges and may be characterized as living in 'two worlds' (Greenberg and Marra, 2005).However, migrants also transiently occupy a series of stopover habitats along their migratory route.The biology of migratory species crucially depends on the different habitats and spatial distributions they occupy (Dingle, 2014;Bowlin et al., 2010), and although accounting for a relatively small proportion of the annual cycle, stopover habitats are crucial to the success of migratory species.Migration can represent the period of greatest adult mortality (Sillett and Holmes, 2002), and successful migration may be related to the availability of high-quality stopover habitats (Hutto 1998).Migration periods generally, and stopover biology specifically, remain crucial knowledge gaps for many species (Bowlin et al. 2010).Therefore, developing a holistic understanding of migration biology requires identifying stopover sites and understanding habitat characteristics that migrants rely on (Sheehy et al., 2011).Stopover sites represent important opportunities to study and observe migrants, especially for highly mobile, small-bodied species that are difficult to track (Faaborg et al. 2010).
Some general principles apply to all migrants (Dingle 2014), but migratory requirements vary among taxonomic groups (McGuire and Fraser, 2014).Temperate insectivorous bats are a group that faces different constraints and tradeoffs compared to other migrants (McGuire and Guglielmo, 2009;McGuire et al., 2014;Jonasson and Guglielmo, 2019;Baloun et al., 2020).Temperate insectivorous bats experience a variety of challenges throughout the annual cycle and, when faced with winter as an extended period of resource limitation, most species undertake migration in some form (Fleming and Eby, 2003).For some species, migration distances are relatively short, such as altitudinal or regional migrations (McGuire and Boyle, 2013;Griffin, 1970).Other species undertake latitudinal migrations, travelling hundreds or even thousands of kilometers between seasonal grounds (Fleming and Eby, 2003;Cryan, 2003).Despite this, migration is a poorly understood aspect of bat behaviour (Fleming, 2019).There is an abundance of research focused on the summer active season (e.g., Kurta and Teramino, 1992;Fenton, 1970), and winter hibernation has been extensively studied in regional migrants (Boyles et al., 2020).However, the distribution and ecology of latitudinal migrants during the winter is virtually unknown and there are many outstanding questions about the ecology, behaviour, and physiology of bats during migration.Most studies of bat migration are based on either the mark-recapture of banded bats (Griffin, 1983) or documentation of wind turbine mortality (Baerwald and Barclay, 2009), but little is known about migratory bats while en route.Radiotelemetry studies are beginning to elucidate migratory movement patterns (Roby et al., 2019;McGuire et al., 2012), but migration movement ecology and behaviours are still poorly understood.
Stopover sites are critical ecosystems along the migration path where bats can forage to rebuild fuel stores during the night and roost during the day, sometimes stopping for multiple days before continuing migration (McGuire et al., 2012;Jonasson and Guglielmo, 2019;Fleming, 2019;Moore et al., 1995).Little is known about habitat requirements during stopover, and while it is likely that forested habitats are important for stopover, specific roosting requirements may be more flexible during migration than during breeding seasons (e.g., McGuire et al., 2012).In temperate North America, the three species of latitudinal migrants (hoary bats Lasiurus cinereus Beauvois, 1796, eastern red bats Lasiurus borealis Müller, 1776, and silver-haired bats Lasionycteris noctivagans Le Conte, 1831) all roost in trees, whether in cavities or foliage, further emphasizing the importance of forested areas for stopovers.Therefore, extensive land use change may be an important threat to migratory bats as large areas of forest are cleared for agriculture and urban development.A global shift toward urbanization has transformed many natural landscapes into a mosaic of dense urban areas, agricultural land, and fragmented patches of less-disturbed habitat (McKinney, 2002;Olejniczak et al., 2018).Remaining stopover sites are often subjected to anthropogenic disturbances that reduce habitat quality and species diversity (Ethier and Fahrig, 2011;Moretto and Francis, 2017).Although degraded from the original habitat, remnant urban forests may be important stopover sites for migratory animals.For example, urban parks are important migration stops for short-term refueling in some migratory birds (Seewagen et al., 2011) and may also represent important stopover sites for migrating bats (Coleman and Barclay, 2013).Thus, it is important to understand how bats use these patches to maintain suitable habitat for migratory bats and inform city planning decisions when developing forested areas.
Island biogeography theory (MacArthur and Wilson, 1963) may be a useful framework for making predictions about stopover habitat use in urban landscapes.This framework proposes that species richness increases with island size and decreases with distance from mainland source population, based on the probability of species reaching the island and the amount of resources and niche space available.In temperate forest biomes, remnant forest patches within an urban matrix can represent islands separated from contiguous forest (Olejniczak et al. 2018).In this way, island biogeography theory can be used as a framework to predict the relative number of migratory bats that will stop over in an urban forest, with general expectations of greater bat activity at larger and less isolated forest patches.In addition to the absolute size of forest patches, the amount of edge habitat may affect the use of a patch as many species of bats forage along forested edges (Walsh and Harris 1996;Jantzen and Fenton 2013).Thus, larger patches may provide more roosting habitat, while patches with more edge habitat may provide more foraging opportunities.We used acoustic monitoring during the fall migration period to record bat activity at urban forests of varying sizes.We hypothesized that the island biogeography framework is applicable to stopover site selection by migrating bats.Specifically, we predicted 1) larger forest patches (more roosting habitat) would have higher bat activity than smaller forest patches, 2) forest patches with a higher shape index (more edge habitat for foraging) would have greater activity than more circular patches, and 3) isolated forest patches would have lower activity than those in close proximity to other forests.Study Area and Species-We conducted our research in Waterloo region, Ontario, Canada.Waterloo region has a population of approximately 617,870 (Region of Waterloo, 2019) and is experiencing an increase in urbanization and agricultural output (Bucknell and Pearson, 2006).Much of the original forest in the region has been cleared for agricultural use, creating a "mosaic" fragmentation pattern surrounding the city.Patches of urban forest are present throughout the city in the form of municipal parks, historical sites, and privately owned woodlots.We selected six forested sites with a range of areas, shapes, and degrees of isolation from nearby forest patches (Figure 1).
Southern Ontario is home to eight species of bats, including sedentary species (big brown bats Eptesicus fuscus Beauvois 1796), regional migrants (little brown bats Myotis lucifugus Le Conte, 1831; eastern small-footed bats Myotis leibii Audubon and Bachman, 1842; northern long-eared bats Myotis septentrionalis Trouessart, 1897; and tricolored bats Perimyotis subflavus Cuvier, 1832), and latitudinal migrants (hoary bats Lasiurus cinereus, eastern red bats Lasiurus borealis, silver-haired bats Lasionycteris noctivagans).All research activities were approved by the University of Waterloo Animal Care Committee (Protocol #42374).Forest characterization-To determine forest characteristics, we manually created shapefiles of each study site and the surrounding forest patches within a 2 km radius using QGIS version 3.16 (Bender et al. 2003) Focal forest patches were defined as the contiguous forest closest to the bat detector, ending when gaps between trees were >20 m (Hale et al., 2012).All surrounding forest patches were required to have a core area of at least 400 m 2 to distinguish them from individual trees or sparse tree cover (Table 1).Trees in residential areas (i.e., trees in back yards, gardens) or along roadsides were omitted from the analysis (Krauel and LeBuhn, 2016).When characterizing forest patches, we included a 20 m buffer from the forest edge to account for foraging habitat (Figure 2; Jantzen and Fenton, 2013).Characterizing specific details of edges was beyond the scope of our study, but an interesting opportunity for future research which may consider urban forest patches which may border on habitats ranging from grassy meadows, to wetlands, to high traffic streets.We used Fragstats (version 4;McGarigal et al., 2012) to characterize forest patches by converting the shapefiles of all study sites and surrounding 2 km of forest to raster format and analyzing the default variables for each study site at the "patch metrics" level.Acoustic monitoring-We used acoustic monitoring to quantify fall migration activity, recording for 85 nights from Aug 12 -Nov 5, 2020.We installed Songmeter SM2+ bat detectors (Wildlife Acoustics; MA, USA) along the forest perimeter at the six study sites to record bat echolocation calls, as most bat species have the highest detectability along the forest edge (Brigham et al, 2004).A single microphone was attached to the top of a 3 m pole, >15 m from artificial light and oriented away from buildings and other objects that could cause background noise or echoes.The detectors were programmed to record throughout the night, from 15 min before sunset until 15 min after sunrise, and recordings (sampling rate = 192 kHz) were saved as full-spectrum, timestamped WAV files.Call analysis-All recordings were filtered by SonoBat Batch Scrubber 5.1 using the "high quality" setting with a high pass filter of 5 kHz and default call quality and percent match settings (0.6 and 0.9 respectively).The filtering algorithm excluded recordings that did not contain bat echolocation calls, however high katydid (Orthoptera) activity added background noise to the recordings that could have hidden bat passes from the software.The remaining files were analyzed using SonoBat 4 Software with the North America setting.Activity was quantified as the number of recordings for each species on a nightly basis, allowing us to measure the relative abundance of species at different sites (Brigham et al. 2004).The echolocation calls of E. fuscus and L. noctivagans are very similar, but with modern classification algorithms it is now possible to discriminate the two species in many cases (Lausen et al. In Press).Statistical analysis-We used R (version 3.6.1;R Core Team, 2020) for all statistical analyses.To determine the seasonal period with sufficient activity for analysis (i.e., to exclude low activity period after the majority of migrants had passed through the region), we considered the cumulative distribution of nightly activity, keeping all dates up to and including the 95th percentile of the activity distribution, and excluding nights late in the season after the main period of activity had passed; this occurred between September 22 -30 for all species.
We tested the effect of study site on E. fuscus and L. noctivagans activity using a likelihood ratio test to compare nested generalized linear models with site as the explanatory variable.Since bat activity fluctuated over the sampling period, we used a negative binomial distribution to account for overdispersion of the data (glm.nbfunction from the MASS package; Venables and Ripley, 2002).Given an overall effect of site on bat activity, we determined pairwise differences among sites using the cld function from the multcompView package (Piepho 2004).
Fragstats provides multiple metrics to characterize each of patch area, shape, and isolation.We used principal component analysis (PCA) as a data reduction technique to determine the main characteristics of the forest patches at each study site (Vidal et al. 2016).Given the limited number of sites relative to the number of metrics, we applied a separate PCA to the metrics within the three categories of patch size (area, core area, and perimeter), shape (related circumscribing circle, shape index, and fractal dimension index), and isolation (proximity index and Euclidean nearest neighbour index).We performed each PCA using the prcomp function with centered and scaled data, retaining the first principal component as an integrated measure of size, shape, or area (Table 2).
To quantify the relationship between bat activity and forest characteristics, we performed linear regression on the relative activity predicted by the generalized linear model and each of the three principal components for each study site to determine the species-specific relationships between bat activity and forest area, shape, and isolation.

Results
Site characteristics-Forest area ranged from 11.5 Ha (Columbia Lake) to 108.1 Ha (Grey Silo), and the sites encompassed a variety of shapes and levels of isolation (Table 1; Table 2).Open water or wetland were present at all sites.The two closest sites (Bechtel Park and Woodside) were separated by 1.5 km while the two most distant sites (Grey Silo and Huron Natural Area) were separated by 12.7 km.Stanley Park was difficult to characterize because the park is bisected by a 20 m wide road.We conservatively included the entire park when characterizing our sites, but the site area might effectively be constricted to the smaller section of trees on the same side of the road as the detector.Linear regression analyses (below) yielded higher R 2 values when including the smaller area for Stanley Park but did not qualitatively change the interpretation of our analysis (i.e., no relationships were significant under one area but not the other).Therefore, we conservatively include the larger patch area in our analysis.Acoustic monitoring-After scrubbing for noise, we recorded 21,646 files over the 85 nights of sampling, 16,636 of which were identified to species (Table 3).Activity was recorded on 90.7% of the nights that were sampled, and we detected six of the eight Ontario bat species (Table 3).The site with the most recorded activity was Huron Natural Area (n = 7,425 passes) and the lowest activity was recorded at Woodside (n = 1,447 passes).The most common species was E. fuscus (n = 13,416 passes) which is a resident species commonly found in the region.The migratory L. noctivagans had the second highest recorded activity (n = 2,134 passes).We detected E. fuscus and L. noctivagans at all sites for the majority of the sampling period, therefore providing a comparison of resident and migratory species.Low activity for the remaining four species precluded including them in further statistical analysis.Influence of site characteristics-Activity varied among sites for both L. noctivagans (LR = 77.2,df = 5, p < 0.0001) and E. fuscus (LR = 54.6,df = 5, p < 0.0001).Post-hoc analysis of L. noctivagans identified two groups within the study sites, where the "high activity" group included the two largest sites (Huron Natural Area and Grey Silo golf course), and the "low activity" group included the two smallest sites (Woodside and Columbia Lake; Figure 3).Post-hoc analysis of E. fuscus identified three, less distinct groups with the same three highest and lowest activity sites as L. noctivagans.

Discussion
We detected six out of the eight Ontario bat species at urban woodlots in the Waterloo region, with pulses of L. noctivagans migratory activity observed in late August and mid-September, consistent with previous study in the region (McGuire et al. 2012).As only a small percentage of the historical forest cover remains in the region, urban forests may represent important habitat for both resident and migratory bats.
As predicted, there was a positive relationship between forest area and L. noctivagans activity.Rather than this being a selective process, we may see the most activity at forests with larger areas because they act as a larger catchment for migrating bats (as predicted by island biogeography theory).The weaker relationship between forest shape and activity is consistent with this interpretation.Migrating bats may not select stopover sites based on characteristics such as the amount of foraging habitat available, but rather the largest patches may be the most obvious on the landscape, and thus attract the most bats.
The weak relationship between site isolation and bat activity suggests that the scale of a potential isolation effect is likely to be much greater than the local scale of our study sites.Migrating bats may travel hundreds of kilometers per day, therefore separate forest patches within a city are not truly isolated from each other in the context of migratory movements.Bats can quickly fly between patches, requiring only ~20 min to fly between our two most distant patches (assuming 9 m/s flight speed; McGuire et al. 2012).Stopover sites have been traditionally considered at a small scale, with the assumption that migrants do not move outside of a small stopover patch until they resume migratory flight (Bächler and Schaub, 2007).However, it is now clear that the scale of stopover is substantially greater, and migratory birds and bats may relocate >10 km within a larger stopover landscape before continuing migration (Taylor et al., 2011).Within the scale of our study, a 10 km radius nearly covers the total city area, so it is not surprising that isolation was found to be an unimportant factor within the city limits.
Although not explicitly tested in our study, the movement of resident and migratory species within the city may be influenced by different characteristics, depending on whether they prioritize foraging or roosting.As a migratory species that must forage and roost in a short stopover period (< 3 days;McGuire et al. 2012), L. noctivagans may be more active where foraging areas are closer to roosting sites, and a recent study suggests that access to foraging area from roosting sites might be more important for L. noctivagans than roosting itself (Ethier and Fahrig, 2011).Activity may be more strongly related to fragmentation than forest area, where landscape complementation arises when moderate amounts of fragmentation reduce the distance between roosting and foraging sites (Ethier and Fahrig, 2011).As a generalist species, E. fuscus is not expected to have a high correlation with forest fragmentation since they are better able to make use of the anthropogenic landscape, often roosting in buildings instead of relying on available tree roosts (Brigham 1991).Therefore, it is possible that E. fuscus preferentially selects areas based on roost availability rather than foraging habitat.In this study we contrasted the activity patterns of a migratory and non-migratory species, the two most common species at our study sites.In future studies of systems with more diverse assemblages and greater evenness it may be informative to consider overall assemblage activity patterns with site isolation and fragmentation.
According to island biogeography theory, species richness (or abundance) is determined by the balance of immigration and extinction rates, where factors that make it more likely for a species to arrive at the site (e.g., larger area, lower isolation) will increase abundance and vice versa.As resident species that inhabit our study region year-round, E. fuscus may respond to resource limitation or other desirable/undesirable aspects of forest patches, leading to resettling among urban forest patches.Thus E. fuscus activity at urban forest patches may be best explained by both immigration and extinction in the island biogeography theory context.However, migratory L. noctivagans only occupy stopover sites for brief periods (perhaps only 1 -2 days; McGuire et al. 2012) and therefore island biogeography theory extinction dynamics may be less relevant in this context, and occupancy is best described based on patch area and immigration likelihood.Although our study focused on L. noctivagans, other migratory species (i.e., hoary bats, eastern red bats) are expected to follow this same pattern.
Patch area was the most important predictor of bat activity at our study sites, but it may be important to consider additional factors other than shape and isolation.High activity at Bechtel Park (a medium-sized site) and low activity at Stanley Park (a large site) indicates that either habitat selection is not based solely on patch area, or that bats perceive patches differently than the criteria we used to define patches.We characterized sites a priori, but it is important to understand what types of disturbances influence how bats perceive and use forest patches."Big" forests like Stanley Park that are fragmented and more heavily trafficked acted more like a small patch, and sites with smaller core areas like Bechtel Park but many surrounding patches of trees may represent a larger useable habitat than the contiguous forest that was measured.Additional factors such as artificial lighting, trail density, forest quality, and distance to water may also influence bats, although it is unclear the extent to which these factors affect migratory bats that may only stop over at the site for one or two days.
The complex definition of a forest is an important concept for city planning and land conservation efforts.While some species are tolerant of urbanization, others require large areas of undisturbed natural habitat.These are key considerations when deciding what types of forest to develop or preserve as natural areas.Generalist species that are better able to use urbanized areas will not only use "natural" areas within a city but the entire anthropogenic landscape as part of their functional habitat.Conservation measures (e.g., habitat improvement such as providing artificial roosts) may provide greater benefit to some (often more generalist) species, while others may not benefit to the same degree, emphasizing the importance of understanding habitat requirements of each species (Griffiths et al., 2020).As urbanization continues, it is important to understand the role of natural areas within the urban landscape mosaic, and how this varies among species that occupy these habitats, both permanently and seasonally.Acoustic monitoring is an important component of monitoring programs to identify spatial and temporal variation in activity patterns in urban landscapes, and provides complementary information to support studies using more direct methods such as radio-telemetry, roost monitoring, and foraging ecology.
Our results suggest that urban forest patches, even if degraded compared to continuous forest, are important for maintaining landscape connectivity for migratory and resident bat species.Effective conservation and management practices must include all habitats necessary for the annual cycle, and migratory stopover sites are a critical component of this cycle.Acoustic monitoring of these locations is valuable for informing bat conservation efforts and understanding how both migratory and resident species are impacted by ongoing urbanization at multiple scales.Table 1.Forest patch metrics for the six study sites, grouped into the three major categories of area, shape, and isolation.© The Author(s) or their Institution(s)

Figure Legends Figure 1 .
Figure Legends

Figure 2 .
Figure 2. Example of forest shapefiles created in QGIS.The dark green polygon outlines the forest at the Grey Silo study site, while light green polygons outline the surrounding forest patches and include a 20 m buffer zone around the forest edge.A blue point indicates the location of the microphone (Stamen Design, 2014; NAD83/UTM zone 17N).

Figure 3 .
Figure 3. Relative activity of (a) Eptesicus fuscus and (b) Lasionycteris noctivagans at each study site as predicted by generalized linear models.Error bars represent standard error and post-hoc groupings are shown with superscript letters and bars.Note that the same sites hosted the highest and lowest activity for each species.

Figure 4 .
Figure4.Linear regression between activity and the principal components of each site for Lasionycteris noctivagans, with 95% confidence interval indicated with grey bands.Higher PC scores for shape represent less convoluted (more circular) shapes.Higher PC isolation values represent more isolated sites.The relationship between activity and the three principal components are all in the direction predicted by island biogeography.

Figure 1 .Figure 2 .Figure 3 .Figure 4 .
Figure 1.Map of six study sites and detector locations within Kitchener-Waterloo, ON, acoustically monitored between August 12-November 5, 2020.A = Columbia Lake (within the University of Waterloo Environmental Reserve), B = Grey Silo golf course, C = Bechtel Park (municipal park), D = Woodside National Historic Site, E = Stanley Park (municipal park), and F = Huron Natural Area.Blue dots mark the approximate location of each microphone and grey stippling indicates major municipal parks and greenspaces as provided by the base map (Stamen Design, 2014; NAD83/UTM zone 17N).338x190mm (96 x 96 DPI)

Table 2 :
Principal components of area, shape, and isolation calculated for each study site.Each component was calculated by applying a separate principal component analysis to the metrics within the three categories of patch size (area, core area, and perimeter), shape (related circumscribing circle, shape index, and fractal dimension index), and isolation (proximity index and Euclidean nearest neighbour index).

Table 3 .
Total activity at each site, calculated as the number of files (recorded passes) for each species over the 85 day sampling period.Two Ontario species, Myotis septentrionalis and Myotis leibii, were not detected.