Roosting ecology of endangered plant‐roosting bats on Okinawa Island: Implications for bat‐friendly forestry practices

Abstract Roosting information is crucial to guiding bat conservation and bat‐friendly forestry practices. The Ryukyu tube‐nosed bat Murina ryukyuana (Endangered) and Yanbaru whiskered bat Myotis yanbarensis (Critically Endangered) are forest‐dwelling bats endemic to the central Ryukyu Archipelago, Japan. Despite their threatened status, little is known about the roosting ecology of these species and the characteristics of natural maternity roosts are unknown. To inform sustainable forestry practices and conservation management, we radio‐tracked day roosts of both species in the subtropical forests of Okinawa's Kunigami Village District. We compared roost and roost site characteristics statistically between M. ryukyuana nonmaternity roosts (males or nonreproductive females), maternity roosts, and all M. yanbarensis roosts. Generalized linear models were used to investigate roost site selection by M. ryukyuana irrespective of sex and age class. Lastly, we compiled data on phenology from this and prior studies. Nonreproductive M. ryukyuana roosted alone and primarily in understory foliage. Murina ryukyuana maternity roosts were limited to stands >50 years old, and ~60% were in foliage. Myotis yanbarensis roosted almost entirely in cavities along gulch bottoms and only in stands >70 years old (~1/3 of Kunigami's total forest area). Murina ryukyuana maternity roosts were higher (4.3 ± 0.6 m) than conspecific nonmaternity roosts (2.3 ± 0.5 m; p < .001) and M. yanbarensis roosts (2.7 ± 0.5 m; not significant). Model results were inconclusive. Both species appear to be obligate plant roosters throughout their life cycle, but the less flexible roosting preferences of M. yanbarensis may explain its striking rarity. To conserve these threatened bats, we recommend the following forestry practices: (a) reduce clearing of understory vegetation, (b) refrain from removing trees along streams, (c) promote greater tree cavity densities by protecting old‐growth forests and retaining snags, and (d) refrain from removing trees or understory between April and July, while bats are pupping.


| INTRODUC TI ON
Understanding species-specific roosting requirements is critical to effectively conserving bats (Altringham, 2011). Bats are of high conservation concern, with over a third of the world's >1,200 assessed bat species considered threatened or data deficient (Frick et al., 2019). Bats also serve as bioindicators (De Conno et al., 2018;Jones et al., 2009) and provide key ecosystem services including pollination, seed dispersal, and the regulation of insect populations (Kunz et al., 2011). Protecting roosting habitats, where bats rest and reproduce (Racey & Entwistle, 2000), is essential to restoring and maintaining bat populations and communities. Maternity roosts are of particular interest to conservationists (Nad'o & Kaňuch, 2015) because reproductive females are important to population viability and often more sensitive to disturbance (Pryde et al., 2005). Although bats are often associated with underground roosts, many species roost in trees or other plants. Here, we use the term "plant-roosting" to refer to bats that roost in any part of a tree (commonly referred to as "tree-roosting") or any other plants (ferns, shrubs, bamboo, pitcher plants, etc.). This includes bats that roost in woody cavities ("cavityroosting") or foliage ("foliage-roosting"). These plant-roosting bats can be particularly difficult to study as they often roost alone or in relatively small colonies in difficult to find roosts and switch roosts often (Lewis, 1995).
Forestry-driven changes to forest composition and stand age affect bats according to their roosting and foraging ecology (Russo et al., 2016). Cavity-roosting bats usually roost in larger diameter trees and snags (Kalcounis-Rüppell et al., 2005;Nad'o & Kaňuch, 2015). If ecological values are not considered, foresters are incentivized to remove such trees for sale, to improve remaining tree regeneration, or to reduce safety hazards associated with falling debris (Guldin et al., 2007). Less is known about foliage-roosting bats, but some may also prefer larger trees in mature forests (Carter & Menzel, 2007). Studies generally recommend preserving old-growth stands or retaining mature trees and snags in mixed-age stands since many tree-roosting bats appear reliant on these habitats (e.g., Burgar et al., 2015;Webala et al., 2010). Thinning trees or clearing understory foliage to promote the growth of remaining trees may benefit open-space and edge-foraging bats while negatively impacting clutter-foraging species (e.g., Patriquin & Barclay, 2003, Carr et al., 2020. For example, species including the openspace foraging Lasiurus seminolus prefer roosting in relatively open forests , while the clutter-foraging Barbastella barbastellus prefer roosts in unmanaged forests with dense understory (Russo et al., 2004). Forestry practices can reduce negative impacts to bats, or sometimes even benefit bats, by taking their ecological requirements into account (Hayes & Loeb, 2007). However, forestry managers often lack sufficient information to adopt more batfriendly practices (Law et al., 2016). Furthermore, though general trends exist, the interactions between various forestry practices and different bat species may differ between regions and forest types.
Most of our knowledge concerning plant-roosting bat ecology comes from research on tree-roosting bats in North America, Europe, and Australia (e.g., Lacki et al., 2007;Law et al., 2016), while data from other regions are sparse (Kingston, 2010;Racey, 2015).
Threatened but poorly understood species are unfortunately common among the bats of Asia and the Pacific Islands (Wiles & Brooke, 2009, Conenna et al., 2017. The genus Myotis is the most represented in tree-roosting bat studies thanks to extensive research on the relatively small proportion of Myotis species native to Western countries (Nad'o & Kaňuch, 2015), but the majority of Myotis in other regions are poorly studied (Moratelli & Burgin, 2019). Similarly, few Murina species have been the subject of roosting studies (e.g., Fukui et al., 2012;Schulz & Hannah, 1998), and the proportion grows increasingly small when considering the remarkable rate of recent species discoveries in this genus (Yu et al., 2020). In Japan, ecological research concerning endemic plant-roosting bats has been limited despite most being threatened (Preble et al., 2021).
These assessments consider these species reliant on intact mature forests. However, ecological information is limited, especially concerning M. yanbarensis (Funakoshi et al., , 2019, and research has been practically absent on Okinawa. Murina ryukyuana have been found roosting mostly in understory foliage, and only seven M. yanbarensis roosts have been reported, five in tree cavities and two in culverts (Asari et al., 2021;Funakoshi et al., 2019;Watari & Funakoshi, 2013). Neither species is likely to hibernate, but winter records are sparse. Notably, no natural maternity day roosts have been reported for either species, although maternity colonies of M. ryukyuana have used artificially placed dry leaves as day roosts , and mothers with prevolant pups were observed night-roosting in canopy foliage on Amami Ōshima (V. Dinets, unpublished). Both species have been identified as high research priorities because they are threatened, endemic, and poorly understood (Preble et al., 2021).
We radio-tracked M. ryukyuana and M. yanbarensis on Okinawa over two years to clarify the roosting habits of these elusive bats and thereby inform conservation measures and bat-friendly forestry practices. Specifically, we sought to (a) determine what types of roosts and roost sites these bats use to inform conservation of these habitats, (b) determine when and where maternity colonies form so forestry activities can avoid these sensitive roosts, and (c) compare roosting ecology between M. ryukyuana and M. yanbarensis to determine to what degree management strategies must be speciesspecific. We were particularly interested in the influence of stand age as existing local biodiversity conservation plans prioritize the preservation of older forests (Okinawa Prefecture Department of Agriculture, Forestry, & Fisheries, 2013). Our primary hypotheses were that both species roost only within forests, that M. ryukyuana roosts flexibly in both foliage and cavities in stands of various ages, and that M. yanbarensis roosts only in cavities in old-growth stands.

| Study site
Our study was conducted in the forests of the Kunigami Village District (26°45′N, 128°15′E), part of the Yambaru region of Okinawa, Japan. The low human population (~5,000 people) is concentrated around the coasts, and the interior of the roughly 20,000 ha area consists of low mountains (max. ~500 m altitude). Numerous streams weave through subtropical broadleaf forests dominated by Castanopsis sieboldii (Family Fagaceae). Yambaru is one of Japan's most biodiverse regions (Ito et al., 2000). Yambaru's forests have also served as a wood resource since the Ryukyu Kingdom period (15th-19th century). Heavy logging of secondary forests occurred during the 1950s postwar reconstruction period (Saito, 2011). Mature trees were targeted from the 1960s to 1990s, and since around 1990, forestry has greatly declined. As a result, Okinawa's northern forests are mostly of moderate age, with only 3% of total forest area <30 years old and 32% >70 years old ( Figure 1). Large mature trees occur in low densities and are limited to old-growth stands remaining in the Kunigami Village District and the US Military's Northern Training Area. Local forestry is government-subsidized to stimulate the economy, and there is increasing pressure on the government to properly manage forests for biodiversity conservation (Ito et al., 2000;Sugimura, 1988), especially given the recent National Park and UNESCO natural World Heritage site designations.

| Bat capture and radio tracking
We captured bats using a combination of harp traps, mist nets, and acoustic lures across 50 sites between September 2017 and September 2019. Capture devices were usually set perpendicular to flight paths over trails or streams. To increase capture rates, we broadcast synthesized stimuli based on social calls of unrelated species and conspecifics using Sussex Autobat acoustic lures (Hill & Greenaway, 2005). Captured females were categorized as nonreproductive, reproductive (pregnant or lactating), postlactating, or juvenile. Juveniles were identified by their lack of wing joint ossification (Brunet-Rossinni & Wilkinson, 2009). We attached VHF radio transmitters (Holohil LB-2X) to 17 M. ryukyuana (eight male and nine female) and ten M. yanbarensis (seven male and three female; Appendix S1) to identify roosts. We attached transmitters between each bat's shoulder blades using Pros-Aide Adhesive (ADM Tronics Unlimited, Inc.). Transmitter weight was 3.3%-3.8% and 4.7%-5.4% of tracked M. ryukyuana and M. yanbarensis bodyweight, respectively.
We tracked bats to day roosts using a handheld radio receiver and Yagi antenna. When we could not visually confirm roosts, we triangulated their location. From May 2018 onwards, we used a thermal imager to improve visual searching (Pulsar Quantum XQ23V, Yukon Advanced Optics Worldwide). For each roost, we recorded roost type (foliage or cavity), roost height (to nearest 0.5 m), and roost plant species. Understory cover at each roost site (5 m radius plot surrounding the roost) was estimated visually to the nearest 5%. We recorded diameter at base height (DBH) for roost trees, tree ferns, and bamboo. When cavity roosts could be reached, we measured the entrance area and internal dimensions to calculate cavity F I G U R E 1 Left-most histograms show stand age of unique Murina ryukyuana (n = 68) and Myotis yanbarensis (n = 10) roost sites (5 m radius plots around roosts) for which stand age data were available. Right-most histogram shows total area (ha) by stand age within the Kunigami Village District to illustrate the distribution of stand ages potentially available within the study area. * Kunigami Village stand ages are measured from 2014, as reported in the most recently available local forest register, instead of 2020 because stand ages could not be updated across the whole district as they were for roost sites volume following Sedgeley & O'Donnell (1999). Roosts were categorized as maternity roosts if they included multiple bats and juveniles, reproductive females, or postlactating females. We recorded the minimum number of individuals per roost using various methods including visual inspection, emergence counts, and rarely by capturing roosting bats.

| Roost site comparisons
We compared roost type proportions between M. ryukyuana nonmaternity roosts, maternity roosts, and all M. yanbarensis roosts using Fisher's exact tests with Bonferroni corrections. Sample sizes were too low to assess differences between nonmaternity and maternity M. yanbarensis roosts. We used Kruskal-Wallis tests followed by Dunn's tests with Bonferroni corrections to compare roost height and the following roost site characteristics between the aforementioned groups: understory cover, canopy cover, canopy height, "southwestness" (SWness), slope, topographic position index (TPI), stream distance, and stand age. Stand age was calculated as years since the last clear-cutting or substantial logging measured from 2020. Stand age estimates were based on local forest register data from 2014 and revised if necessary based on historical aerial photography (US Air Force 1944, 1946, 1962Geospatial Information Authority of Japan 1973, 1977, 1989; Nakanihon Air Service 2011; NTT Geospace 2013). All other variables were derived from airborne laser scanning LiDAR data (Nakanihon Air Service 2011).
LiDAR-derived variables were calculated as the average value of 1m 2 cells within a 5 m radius of the site center. Canopy cover was calculated as the proportion of cells within the site where the difference between elevation and vegetation height was >3 m. TPI was calculated using an inner and outer radius of 5 and 15 m, respectively.
Stream distance (distance to the nearest stream) was calculated in ArcGIS based on a stream layer created using a minimum flow accumulation of 22,500 m 2 . Aspect was transformed to SWness following Beers et al. (1966), where southwest slopes, most commonly exposed to the sun, were given a value of 2 and northeastern slopes a value of 0. Prior roosting studies have used northness and eastness (e.g., Hammond et al., 2016), but we used SWness to reduce the number of covariates. Lastly, roost fidelity and maximum roostswitch distance were calculated per species by averaging values per individual. Roost fidelity was calculated for each individual as the mean number of days before switching roosts. All analyses except for roost fidelity and roost-switch distance calculations considered only unique roosts to avoid bias toward roosts used multiple times.

| Murina ryukyuana roost site selection
We generated thirteen generalized linear models (GLM) reflecting eleven a priori hypotheses as to what characteristics influence roost site selection by M. ryukyuana, as well as a random model and global model (Table 1). Hypotheses included covariates from the literature known to affect roost selection by tree-roosting bats. Higher snag density, tree diameter, tree height, and lower stream distance and canopy cover are generally preferred by tree-roosting bats that use cavities (e.g., Fabianek et al., 2015;Nad'o & Kaňuch et al., 2015).
Tree-roosting bats that roost in foliage generally prefer higher canopy cover (e.g., Kalcounis-Rüppell et al., 2005). Tree-roosting bats have also been found to prefer warmer microclimates, such as lower canopies and south-facing slopes (in the northern hemisphere) exposed to greater solar radiation (e.g., Hammond et al., 2016;Kerth et al., 2001;Law et al., 2016). We assumed higher stand ages to have higher snag densities and average tree diameters. Canopy height was used as a proxy for average tree height and SWness as a proxy for warmer microclimates. For each M. ryukyuana roost site, a random site was selected from within a 1 km buffer and similar site variables measured using GIS. The response variable for GLMs was site use (roost site = 1, random site = 0), and the explanatory variables included canopy height, SWness, slope, TPI, stream distance, and stand age. The quadratic term for stand age was also included as histograms suggested M. ryukyuana might prefer intermediate values.
Canopy cover was not included as inclusion did not affect relative model performance, and only four roost sites and seven random sites had values <100%. We were unable to include understory cover as we did not have random site values. All explanatory variables were sufficiently uncorrelated (Spearman correlation coefficient <0.6).
All covariates were centered before modeling. We used Akaike's information criterion corrected for small sample size (AICc) to assess model fit (Burnham & Anderson, 2002). We considered ∆AICc ≤ 4.0 to be supported by the data and examined well-supported models for uninformative parameters (Arnold, 2010). Data were too scant to repeat this process for M. yanbarensis.

| Phenology
To estimate the reproductive phenology of M. ryukyuana and M. yanbarensis, we compiled all records of reproductive females (pregnant or lactating) captured, maternity roosts, and female nonmaternity roosts from this and prior studies (Funakoshi et al., 2019;Maeda et al., 2001, V. Dinets, unpublished).

| RE SULTS
All 17 tracked M. ryukyuana were successfully relocated (Appendix S1), resulting in 141 day roost records (mean ± SE; 6.2 ± 0.6 roosts per individual). We visually confirmed 105 roosts representing 73 unique roosts-56 nonmaternity roosts and 17 maternity roosts. From six M. yanbarensis (four male and two female) relocated at least once, we visually confirmed 27 roosts (4.5 ± 1.1 roosts per individual). Ten M. yanbarensis roosts were unique, including two maternity roosts. One M. yanbarensis roost was used as both a nonmaternity roost and a maternity roost. Roost and roost site characteristics are summarized in Table 2. Murina ryukyuana maternity roosts were found in a similar variety of plant species (n = 17; Appendices S2 and S4) but more often in cavities compared with nonmaternity roosts (Figure 2). All foliage maternity roosts were in stands ≥57 years old. Cavity maternity roosts were in stands ≥61 years old, and all but one were located in TA B L E 1 Models for roost site selection by Murina ryukyuana including degrees of freedom (K), difference in AICc relative to the model most supported by the data (∆AIC C ), Akaike weights (w), and Nagelkerke's R 2 Castanopsis sieboldii (Appendix S2). Mean DBH of cavity maternity roosts was 22.8 ± 4.8 cm (n = 13). The mean minimum number of individuals per maternity roost ranged from 2 to 15 (mean 6 ± 1).

| Roost characteristics
Maternity roost entrances were larger (mean 70 cm 2 ; n = 4) than for nonmaternity roosts (25 and 35 cm 2 ), but we were unable to measure internal dimensions.
All ten unique M. yanbarensis roosts were in gulch bottoms, usually directly next to flowing water, and almost entirely in small tree cavities (Figure 2 DBH for all M. yanbarensis roosts was 18.4 ± 3.7 cm. Cavities were generally small; entrance area was 27.4 ± 7.1 cm 2 , and cavity volume was 354.1 ± 110.6 cm 3 . The first maternity roost was 844 cm 3 , but the second was too high to measure. Myotis yanbarensis roost height was 2.7 ± 0.5 m overall, and the maternity roosts were 2.0 and 4.5 m high, respectively.

| Murina ryukyuana roost site selection
The only model supported by the data included canopy height, SWness, and stand age (ΔAIC C ≤ 4; Table 1). Although we did not consider the global model ecologically relevant, the fact that it received 59% of the AICc weight suggests that our a priori hypotheses were not strongly supported by the data. The quadratic stand age term in the warm-tall-intermediate model was considered uninformative as it did not improve model fit (Arnold, 2010  Tokunoshima and other Murina species (Fukui et al., 2012;Funakoshi et al., 2013Funakoshi et al., , 2016. Though its diet is unknown, the low wing aspect ratio and faint frequency-modulated echolocation of M. ryukyuana (Funakoshi et al., 2019;Norber & Rayner, 1987;Schnitzler et al., 2003), as well as the diet of congenerics (Ma et al., 2008;Schulz & Hannah, 1998), suggest that this species likely forages in clutter.

| D ISCUSS I ON
Therefore, understory removal probably temporarily destroys both roosting and foraging habitats.
Myotis yanbarensis roosted primarily in woody cavities along gulch bottoms and streams through forests >70 years old. This roost specialization may be a limiting factor for M. yanbarensis populations that also raises the risk of extinction (Sagot & Chaverri, 2015).
A handful of M. yanbarensis have been observed in tunnels (Asari et al., 2021) and rock crevices along streams (H. Tamura, personal communication), and these potentially important roosts warrant investigation. Many tree-roosting bats prefer to roost near water, presumably to reduce commuting time to foraging grounds (Campbell, 2009;Kalcounis-Rüppell et al., 2005). The higher wing aspect ratio of M. yanbarensis relative to M. ryukyuana and the relatively low understory cover around M. yanbarensis roosts suggest that M. yanbarensis prefers relatively uncluttered corridors within forests (Norber & Rayner, 1987). Dietary information would help clarify the foraging habitat of this species. On Amami Ōshima, individuals have been captured over roads through old-growth forest (Asari & Kimoto, 2016;Funakoshi et al., 2019), and streambeds may simply be the only flyways through old-growth forest on Okinawa.
Old-growth forests appear to be important breeding habitats for both M. ryukyuana and M. yanbarensis. Only old-growth forests were used by maternity colonies of both species even when tracked bats traveled through younger stands very close by. This preference for mature stands has been reported for other cavity-roosting bats (e.g., Burgar et al., 2015; and may be related to higher cavity densities in older stands (Matsumoto et al., 2015). Also like other cavity-roosting bats (e.g., Dietz et al., 2018), both M. ryukyuana and M. yanbarensis utilized old woodpecker cavities, and the recovery of woodpeckers in the central Ryukyu Archipelago bodes well for these bats (Kotowska et al., 2020). Our results are inconclusive as to the importance of roost tree diameter or cavity volume, but M. ryukyuana may utilize more cavity roosts during pup-rearing or cold winter months for their insulating benefits (Klug et al., 2012).
Existing records suggest that pregnancy and lactation in M. ryukyuana and M. yanbarensis occur from April to at least mid-July. Two reproductive female captures, ten maternity roosts, and three nonmaternity female roosts for M. ryukyuana were found in the literature. These included maternity roosts found in artificially placed leaves in August-November on Tokunoshima (Funakoshi et al., 2019), lactating females caught on Tokunoshima in late July (Maeda et al., 2001), and a night roost of mothers and infants seen We found too few roosts to explain the apparent rarity of

| Management implications
Given their roosting habits and the lack of evidence of other threats, forest degradation is likely the greatest threat to M. ryukyuana and M. yanbarensis (Fukui & Sano, 2019a, 2019b. Local forestry guidelines already attempt to mitigate impacts to other threatened taxa and could be updated to also reduce negative impacts to bats. Based on our results, we make the following suggestions: 1. Retain understory vegetation to preserve M. ryukyuana habitat and to avoid disturbing the low roosts of both bat species. Understory removal in Yambaru is already of conservation concern due to adverse impacts on native biodiversity (Azuma et al., 1997;Ito et al., 2000).
2. Riparian trees, including mature trees and snags, should not be removed, particularly in old-growth forests, given the rarity of M. yanbarensis and its apparent reliance on these habitats.
Local forestry guidelines already recommend retaining riparian corridors for wildlife, but currently only advise against removing undergrowth and small diameter trees (Okinawa Prefecture Department of Agriculture, Forestry, & Fisheries, 2013). Riparian exclusion zones are a common forestry practice for maintaining biodiversity, including bats (Lloyd et al., 2006, Law et al., 2016, and the natural succession of riparian trees likely produces more snag cavities than thinning (Pollock & Beechie, 2014).
3. Old-growth forest (especially >70 years old) should be preserved to provide ample cavity roosts, and some mature trees and snags should be retained in harvested or thinned forests. Although the exact cavity densities required by each species are unknown, both bat species formed small maternity colonies and switched roosts regularly, suggesting that they require high roost availability (Russo et al., 2005 We are cautiously optimistic about the outlook of M. ryukyuana and M. yanbarensis given current low levels of logging and increasing interest in sustainable management across their range. Still, further research is needed, particularly concerning M. yanbarensis, and these species should be incorporated into forest management plans rather than just listed as threatened. We hope that this study serves as a reference for both local conservation planning and further roosting ecology research in Asia.

ACK N OWLED G M ENTS
Many thanks to all the volunteers who helped us catch and track bats. Special thanks to D. A. Hill, H. Kamigaichi, K. Nakata, and H.
Tamura for their helpful insights, as well as the Yambaru Wildlife Center staff for their technical advice. We are grateful to Y. Nagai of Idea Consultants, Inc., Okinawa Branch Office, K. Kawai, and M. Motokawa for lending equipment. We would also like to thank two anonymous reviewers whose feedback helped to improve this manuscript. This research was generously supported by Pro Natura

CO N FLI C T O F I NTE R E S T
We have no competing interests to declare.