Heard but not seen: Comparing bat assemblages and study methods in a mosaic landscape in the Western Ghats of India

Abstract We used capture (mist‐netting) and acoustic methods to compare the species richness, abundance, and composition of a bat assemblage in different habitats in the Western Ghats of India. In the tropics, catching bats has been more commonly used as a survey method than acoustic recordings. In our study, acoustic methods based on recording echolocation calls detected greater bat activity and more species than mist‐netting. However, some species were detected more frequently or exclusively by capture. Ideally, the two methods should be used together to compensate for the biases in each. Using combined capture and acoustic data, we found that protected forests, forest fragments, and shade coffee plantations hosted similar and diverse species assemblages, although some species were recorded more frequently in protected forests. Tea plantations contained very few species from the overall bat assemblage. In riparian habitats, a strip of forested habitat on the river bank improved the habitat for bats compared to rivers with tea planted up to each bank. Our results show that shade coffee plantations are better bat habitat than tea plantations in biodiversity hotspots. However, if tea is to be the dominant land use, forest fragments and riparian corridors can improve the landscape considerably for bats. We encourage coffee growers to retain traditional plantations with mature native trees, rather than reverting to sun grown coffee or coffee shaded by a few species of timber trees.

land-use change on bats in south India, and the relative merits of two widely used techniques for surveying bats.
All methods for studying bats have potential advantages and disadvantages. Catching bats often allows better species-level identification than acoustic methods and allows the collection of data on biometrics, sex and reproductive status, and genetic material.
However, it is time-consuming, invasive, and can lead to bias as the percentage of the airspace sampled is small and usually close to the ground, and some species are better at avoiding capture than others (O'Farrell & Gannon, 1999).
These drawbacks have been compensated for in recent studies by the use of ultrasound detectors, which detect the echolocation calls of bats. Acoustic recordings can help achieve a more complete species list for the area, and sample some species that are never caught (Kalko, Handley, Handley, Handley, & Handley, 1996;MacSwiney, Clarke, & Racey, 2008). However, some species cannot be separated using echolocation calls alone, low intensity echolocators and nonecholocating bats are under-sampled or not sampled at all (O'Farrell & Gannon, 1999), and ultrasound does not travel far in dense vegetation. Several studies indicate that combining acoustic and capture data give the most complete picture of the bat assemblage (Furey, Mackie, & Racey, 2009;MacSwiney et al., 2008;Murray, Britzke, Hadley, & Robbins, 1999;O'Farrell & Gannon, 1999), yet ecological studies of bats generally report data gathered using only one method.
In this study, we compare and combine results from capture (mist-netting) and acoustic surveys of a palaeotropical bat assemblage to assess the responses of bats to agricultural land uses in South Asia. While harp trapping is widely used in South-East Asia and offers some advantages over mist-netting (Kingston, 2013), we chose to use mist-nets in this study as they are more widely available, easier to transport and deploy in dense understorey, can cover much larger airspaces per unit cost, and are the capture method most commonly used in India. Our early efforts to use 4.2 m 2 two-bank harp traps to catch in forest fragments and coffee plantations were unsuccessful despite high numbers of bats captured in tunnels using these traps (Wordley, Foui, Mudappa, Sankaran, & Altringham, 2014), and we found it difficult to set the traps up in dense understorey, so we reverted to using only mist-nets.
We employed both mist-netting and acoustic sampling in seven habitats (tea (Camellia sinensis) plantations, shade coffee (Coffea arabica and C. canephora) plantations, forest fragments, protected forest, rivers in tea, rivers in tea with riparian corridors, and rivers in protected forests). There were insufficient suitable rivers with coffee planted up to each bank to study bat assemblages in this habitat.
We predicted that acoustic sampling would record more insectivorous species, but that mist-netting would capture more frugivorous species, across all habitats. We predicted that nonecholocating frugivorous species would be detected by capture alone, but that all other species would be more frequently detected by acoustic sampling. We then used these data to compare the relative bat diversity in each habitat.
Despite its size, and the fact that it hosts 10% of the world's bat species, there have been few ecological studies of India's bats. The Western Ghats are, together with Sri Lanka, the eighth "hottest" biodiversity hotspot in the world (Myers, Mittermeier, Mittermeier, da Fonseca, & Kent, 2000), yet only 6% of the land area of the Western Ghats remains under primary vegetation (Sloan, Jenkins, Joppa, Gaveau, & Laurance, 2014). Most of the remaining forest survives as small fragments in a matrix of agricultural land including coffee and tea plantations (Menon & Bawa, 1997). Since 2000, the Nature Conservation Foundation (NCF) has been working to extend and restore the forest fragments in our study area, the Anamalai Hills around Valaparai, and to encourage local coffee growers to shade their coffee with native shade trees rather than commercial timber trees (Mudappa & Raman, 2007). NCF has also been working to understand the relative diversity of different taxa from spiders to mammals in protected forests, forest fragments, and different types of plantations (Kapoor, 2008;Kumar, Mudappa, & Raman, 2010).
Assessment of the value of agroforestry plantations for bats in the palaeotropics has been identified as a key research need (Meyer et al., 2016). Globally, there was a 20% decline in shade-grown agroforestry coffee between 1996 and 2010, such that only 24% of coffee is now managed under diverse shade (Jha et al., 2014). Neotropical studies show that coffee and cacao grown under a canopy of native shade trees can provide a good habitat for many bat species (Faria, Laps, Baumgarten, & Cetra, 2006;Harvey & Villalobos, 2007;Pardini et al., 2009;Pineda, Moreno, Escobar, & Halffter, 2005;Williams-Guillén & Perfecto, 2010. The few studies on bat diversity in coffee from Asia give similar results (Graf, 2010;Wordley, Sankaran, Mudappa, & Altringham, 2015, 2017; however, little is known about the value of this habitat for most palaeotropical bat species. Studies on palaeotropical bats in a range of agricultural land uses, especially large-scale commercial uses, have also been identified as a key research need (Meyer et al., 2016). Tea is a widespread and expanding commercial land use across the palaeotropics (FAOSTAT 2014). In the study site, as is typical, it is grown as clipped bushes with light shade from Australian silver oak trees (Grevillea robusta). Wordley et al. (2015) demonstrated that many bat species avoided areas with a high coverage of tea plantations, and similar patterns have been documented for birds and frogs (Murali & Raman, 2012;Sidhu, Raman, & Goodale, 2010). As climate change is likely to lead to an upslope expansion of the areas suitable for tea and coffee cultivation globally, it is important to understand the likely relative impacts of these two plantation types on biodiversity.
Riparian habitats are important foraging areas for many bat species, due to the abundance of insects. Studies have found that bankside vegetation significantly increased bat activity over rivers, but these have mostly been from temperate regions (Lundy & Montgomery, 2009;Ober, Hayes, & Hall, 2008;Warren, Waters, Altringham, & Bullock, 2000). While Indian law does not currently legislate to promote riparian corridors of vegetation along river banks, in other countries riparian corridors are compulsory in certain land uses, particularly heavily modified plantations, to reduce erosion of river banks, intercept fertilisers, and provide habitats for biodiversity (Marczak et al., 2010;Mayer, Reynolds, McCutchen, & Canfield, 2007;Sweeney et al., 2004).
By studying bat species in a range of habitats in the Valparai plateau and adjacent Anamalai Tiger Reserve, we aim to determine which species survive in human-modified landscapes, and which decline or disappear. We also aim to determine the relative contribution made by human-modified habitats such as fragmented forests, agroforestry plantations, and monoculture plantations to maintaining bat diversity, and which methods are most appropriate for measuring this diversity. We predict that fragmented forests will have lower diversity and altered species composition compared to protected forests; that shade-grown agroforestry coffee plantations will retain a similar but less diverse bat assemblage compared to forest fragments; and that tea plantations will have the lowest diversity of all.
We predict that the presence of riparian corridors on rivers in tea plantations will increase the bat diversity on those rivers compared to rivers without riparian corridors, but that rivers in protected areas will retain the highest diversity. We predict that forest-adapted species such as Megadermatidae, Rhinolophidae, and Hipposideridae will show the greatest declines in all nonprotected habitats, and that fruit bats will be largely absent from tea plantations.

| Study area
This study was conducted on the Valparai plateau and adjacent Anamalai Tiger Reserve in the state of Tamil Nadu in the southern Western Ghats (N 10.2-10.4°, E 76.8-77.0°). The Valparai plateau is an agricultural landscape approximately 800-1,600 m asl dominated by tea plantations interspersed with shade-grown coffee plantations, eucalyptus plantations, rainforest fragments, streams, and riparian vegetation (Mudappa & Raman, 2007). Forest fragments and riparian corridors were remnant forest patches or secondary forest/ overgrown plantations dominated by mature native trees. Several of these fragments have received supplementary planting to restore and extend them (Mudappa & Raman, 2007). The native vegetation is mid-elevation tropical wet evergreen forest of the Cullenia exarillata-Mesua ferrea-Palaquium ellipticum type (Pascal, 1988;Raman, Mudappa, & Kapoor, 2009). For a detailed map of the study area, see Wordley et al. (2015). The average annual rainfall is 3,500 mm, of which about 70% falls during the southwest monsoon (June-September; Raman et al., 2009).

| Data collection
We chose five sites for each of the seven study habitats, and between January and May 2010 to 2013, and in November-December 2014, we spent a total of two nonconsecutive nights at each site capturing bats and recording echolocation calls. We caught bats and recorded them on the same night to reduce the effects of inter-night variation.
At every site, we caught bats using five ground level (6 m x 2.5 m) mist-nets 50-200 m from the nearest acoustic sampling point, and recorded at five points 100 m apart for 15 min per point every night.
We started recording 40 min after sunset as bats begin foraging at different times relative to sunset, so until it is fully dark, each acoustic recording point may be subject to temporal bias. We used a handheld Pettersson D240X ultrasound detector (www.batsound.com) recording onto an Edirol R-09 (www.roland.com) digital recorder.
The detector was set to constantly sample, so a trigger level was not used. The detector was moved in a semicircular arc to record bats from a wider section of the aerospace. Nets were opened at sunset and closed after 2.5 hr. Bats caught in nets were identified to species using Bates and Harrison (1997) and Srinivasulu, Racey, and Mistry (2010). In riparian habitats, all nets were set over the river, and recordings were made on the river banks, pointing at the river, so only species close to the river would be recorded. All rivers were at least 4 m wide at the point of sampling. Forest fragment size varied from 2.2 to 102.8 ha, riparian corridor area from 3.7 to 159.7 ha, and riparian corridor width from 17 to 1,070 m at the widest point. All study sites were at least 1 km apart, and spatial auto-correlation of bat species presence was low (Wordley et al., 2015) so has not been considered here. The elevation range in this study did not affect the likelihood of the presence of any of the species modeled by Wordley et al. (2015), so has not been considered here.

| Sound analysis
Echolocation calls were visualized as spectrograms to measure call parameters using BatSound (www.batsound.com). Calls were manually identified using an echolocation call library for the area (Wordley et al., 2014). At each recording point, a species was marked as present if a call unambiguously attributable to that species was recorded within the 15 min recording. Due to call overlap between species (Wordley et al., 2014), not all species were easily identifiable. Scotophilus heathii and Pipistrellus ceylonicus overlapped extensively in call frequency and had the same call structure, but S. heathii calls were clustered toward the higher end of the P. ceylonicus range with only one call as low as 33.8 kHz; here we have classified calls of 31-34 kHz as P. ceylonicus, and not attempted to

| Species rarefaction curves
We generated individual-based species rarefaction curves combining capture and acoustic data for each habitat, using the R packages "picante" and "vegan," using the formula "rarefaction" (http://www. jennajacobs.org/R/rarefaction.html; Kembel et al., 2010;Oksanen et al., 2013;R Core Team, 2014). We calculated the "Chao" estimated species richness per habitat and per site for each method using the "vegan" package in R.

| Species richness
We combined the data from both nights at each site to avoid pseudo-replication. Generalized linear mixed models would not converge for these data, so we ran a Poisson generalized linear model (GLM) in "lme4" with method (capture, acoustic sampling) and habitat as the predictor variables, and compared models with and without each factor to the full model using a likelihood ratio χ 2 test (Bates, Maechler, Bolker, & Walker, 2014). We ran pairwise comparisons and corrected for multiple testing using the false discovery rate (FDR) method in the "lsmeans" package in R (Lenth, 2014).

| Size effects
Areas of forest fragments and riparian corridors were calculated using ArcGIS (Wordley et al., 2015). Riparian corridor width (perpendicular to river bank) was measured at each acoustic transect point and the mean taken per corridor. Linear regression analyses were performed in R to look at the effects of forest fragment area, riparian corridor area, and riparian corridor width on bat species richness (Table S7).

| Activity
The total number of "records" of bats per method and per site were counted. We counted every bat captured as one record, and every species recorded in a 15-min acoustic recording as a record. We did not count the number of calls per species at each point, to reduce bias from recording the same individual bat multiple times or due to different likelihood of detection of different species. We followed the same procedure as for species richness except that we ran a quasi-Poisson GLM due to over-dispersion.

| Species composition
Using the "PERMANOVA" (permutational multivariate analysis of variance using distance matrices) method executed through the "ADONIS" function in "vegan" with 9,999 permutations, we tested for differences in species composition between habitats, and ran pairwise comparisons using FDR.
For each species with >30 records in total, we used Kruskal-Wallis tests on site level data to test for changes in the activity between habitats, using the "agricolae" package in R to conduct pairwise comparisons with FDR correction (de Mendiburu, 2014).

| Species richness
We recorded 17 species (Table 1). Observed species richness was equal or close to the estimated species richness per habitat and per site when both detection methods were combined (Figure 1, Table S1). When capture data alone were used, species richness and estimated species richness were considerably lower, with no bats captured in tea and higher species richness in tea riparian and riparian corridors than in protected area forest. When acoustic data alone were used, the estimated species richness was slightly lower across the board than when methods were combined, but the overall pattern was very similar; except in forest fragments where estimated species richness was five compared to 8.5 when methods were combined (Table S1). Combined capture and acoustic data and acoustic data alone yielded significantly higher species richness than capture data alone, for all habitats, but not significantly higher than for acoustic data alone (Table S2).
Species richness differed between habitats (deviance = 773.7, df = 7, p < .001) and with the sampling methods used (deviance = 850.9, df = 3, p < .001), but there was no significant interaction (deviance = 16.47, df = 12, p = .17). Species richness was significantly lower in tea plantations than in all other habitats ( Figure 2) and highest in protected forest rivers followed by protected area forest. Rivers through protected area forest had significantly greater species richness than tea plantations, rivers through tea with no riparian corridor (hereafter tea riparian), coffee plantations, and forest fragments ( Figure 2, Table S2).

| Activity
Habitat had a significant effect on overall activity (deviance = −2,286.02, df = −18, p < .001), as did method (deviance = −699.88, df = −14, p < .001), but the interaction was not significant (deviance = −75.26, df = −12, p = .074). Activity was significantly lower in tea plantations and forest fragments than in all other habitats when methods were combined, but with capture data alone, forest fragments had slightly higher activity than coffee plantations (Figure 3, Table S3). Activity using combined data was significantly higher on rivers in protected area forest than in all other habitats except riparian corridors. Activity was significantly lower for capture data than for combined data, whereas activity for acoustic data was not significantly different from combined.

| Species composition
Species composition (using combined acoustic and capture data) differed significantly between habitats (  (Table S4).
When only capture data were used, fewer significant differences in species composition were seen between habitats (Table S5). While F and p values were typically lower using acoustic data only compared to combined data, most significant results remained (Table S6).

| Comparison of methods
We demonstrate that the use of acoustic methods is feasible in India, and that while combining mist-netting and acoustic data maximizes the number of species that will be detected, acoustic methods alone give broadly similar results in terms of richness and composition to those obtained by combining both methods. Despite the fact that several species could not be identified from acoustic recordings due to overlapping call structure, acoustic methods gave consistently higher estimates of species richness and activity than mist-netting ( Figures 1-3), and detected more significant differences in species composition (Table 2).
Very few bats were caught in tea plantations (none, after per habitat singletons were removed); however, several species were recorded in tea acoustically (Table 1). This is likely due to the difficulty of catching

| Comparison of plantations, fragments and forest
This study found broadly similar species richness and composition between bat assemblages in protected area forest, coffee plantations, and forest fragments, indicating that the original bat assemblage need not be lost in a modified landscape so long as F I G U R E 1 Species rarefaction curves with 95% confidence intervals per habitat for capture data, acoustic data, and acoustic and capture data combined  in protected forests than in coffee or forest fragments (Tables 1   and S4); which is expected as they are typical "forest adapted" bats (Wordley et al., 2017). More surprisingly, Miniopterus pusillus, which has the long narrow wings and flexible echolocation calls seen in open air foragers (Wordley et al., 2017), was more frequently recorded in protected forest than in forest fragments. This may be because it can forage within the open understorey of protected area forests more easily than in the dense understorey of forest fragments. We saw that several other species were most abundant in protected forests, but they were recorded too infrequently for statistical analysis (Table 1). Megaderma spasma was only caught in forest fragments and protected forest and was caught more in protected forest; Hipposideros pomona was recorded in all habitats but was most often recorded in protected forests. These species may be the most vulnerable to any land-use change from protected forest.
This underscores the need to examine data beyond richness metrics alone.
Tea is the most heavily modified habitat in the landscape, containing no native bushes or trees. Lower species richness and activity of bats were seen in tea plantations, and bat assemblages were  Habitat Activity While recorded frequently in tea plantations, these two species have been shown to decline in likelihood of occurrence as the percentage tea cover of an area increases (Wordley et al., 2015). Therefore, tea plantations apparently have no clear "winner" species of bat and many "losers," echoing the observations of Maas et al. (2015) that there are few "agricultural specialist" bats.

| Comparison of riparian habitats
The riparian habitats all tended toward having a greater species richness and activity than their nonriparian counterparts, although the only significant difference was between tea and tea riparian.
Species composition also changed between riparian and nonriparian habitats within the agricultural landscape. Other tropical studies reveal that riparian vegetation can be richer in bat species and have higher activity levels than comparable nearby nonriparian vegetation (Monadjem & Reside, 2008;Sirami, Jacobs, & Cumming, 2013;Taylor, Monadjem, & Nicolaas Steyn, 2013), and some bat species show particular preferences for riparian vegetation (Avila-Cabadilla et al., 2012). Riparian corridors represented a better habitat for bats than rivers through tea with no riparian corridor, but a poorer habitat than rivers in protected areas. The benefits to bats of riparian corridors may be enhanced by having native tree cover on both banks of the river, not just one as seen in this study.

| Conservation implications
The high level of protection given to protected forests should be maintained and extended to other intact forests in the Western  Mudappa, Sankaran, & Altringham, 2016). Megaderma spasma, while globally widespread, appears sensitive to disturbance as it was never recorded in tea plantations or tea riparian habitats; likewise,

Rhinolophus beddomei (known only from India and one location in
Thailand) and Rhinolophus rouxii (largely restricted to South Asia, and one location in Burma) were not recorded in tea plantations and were much rarer outside of protected areas.
Shade coffee under native trees has value in a biodiversity hotspot, but the high value for bats seen in this study may rely on some intact forest remaining in the landscape (Faria & Baumgarten, 2007). There is a growing trend toward the use of non-native timber trees in coffee plantations in Valparai and globally (Jha et al., 2014); it is important to develop and implement mechanisms that encourage the use of native species, such as premium prices for coffee planters who retain or replant them.
Tea plantations, however, have a significant negative impact on the diversity of all species studied in them thus far (Murali & Raman, 2012;Sidhu et al., 2010;Wordley et al., 2015Wordley et al., , 2017. If these plantations are to be made compatible with conservation in a biodiversity hotspot, changes to plantation management are needed. For example, in Valparai, NCF is encouraging tea planters to use native trees for shade rather than the exotic Australian silver oak. Shade will always be sparser for tea than for Coffea arabica as tea bushes need more sun, but supplementing the exotic trees with native species may benefit bat diversity. Growers should be rewarded for restoring forest fragments and planting riparian corridors. In areas of high conservation value such as the Western Ghats, it may also be sensible to promote mechanisms to discourage the conversion of shade coffee to tea. The Anamalai Hills were initially planted with shade-grown coffee (Mudappa & Raman, 2007); it is relatively recently that the landscape has become tea dominated. Localized schemes to reward coffee growers although payments for ecosystem services, access to elite international markets, or help with developing ecotourism could be trialed, alongside or instead of legislation or financial penalties for clearing native trees to plant tea.
While riparian corridors are not equivalent to rivers through protected area forest for bats, they have more value than rivers without riparian corridors. Legislation or incentives to encourage plantation owners to leave a buffer of native trees on each side of every river would greatly benefit bats, and other species in the landscape (Gray, Slade, Mann, & Lewis, 2014;Kumar et al., 2010). The Indian government has committed US$10 billion to planting five million hectares of forest, and improving forest quality on another five million hectares (National Action Plan on Climate Change, 2011). While the main focus of this reforestation drive is for large-scale work (5,000+ ha plots), riparian corridors in agricultural land may be a good investment for reforestation due to the biodiversity, hydrological, and erosion reducing benefits (Mayer et al., 2007;Sweeney et al., 2004).
They may help to restore landscape connectivity and have potential for mitigating human-elephant conflict by providing "migration corridors" through the landscape where elephants can drink, feed, and rest in the shade rather than venturing into tea estates (Kumar et al., 2010). In summary, diverse tropical agricultural landscapes can maintain bat diversity, providing sufficient native trees are maintained.