bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Convergent acoustic community structure in South Asian dry and wet grassland birds 2 Sutirtha Lahiri1, Nafisa A. Pathaw1 and Anand Krishnan1,* 3 4 5 1- Department of Biology, Indian Institute of Science Education and Research (IISER) 6 Pune, Pashan Road, Pune 411008, India. 7 *- Author for correspondence- [email protected] 8 9 10 Running title: Acoustic Communities in grassland habitats 11 Keywords: Grasslands, India, acoustic community, community bioacoustics, signal space, 12 birds 13 14 15 Abstract 16 Although the study of bird acoustic communities has great potential to provide valuable 17 conservation data, many aspects of their assembly and dynamics remain poorly understood. 18 Grassland habitats in South Asia comprise distinct biomes with a unique avifauna, 19 presenting an opportunity to address how community-level patterns in acoustic signal space 20 arise. Similarity in signal space of different grassland bird communities may be due to 21 phylogenetic similarity, or because different bird groups partition the acoustic resource, 22 resulting in convergent distributions in signal space. Here, we quantify the composition, 23 signal space and phylogenetic diversity of bird acoustic communities from the dry semiarid 24 grasslands of Northwest India and the wet floodplain grasslands of Northeast India. We find 25 that acoustic communities occupying these distinct biomes exhibit convergent signal space. 26 However, dry grasslands exhibit higher phylogenetic diversity, and the two communities are 27 not phylogenetically more similar than expected by chance. The Sylvioidea encompasses 28 half the species in the wet grassland acoustic community, with an expanded signal space 29 compared to the dry grasslands. Thus, dry and wet grassland communities are convergent 30 in signal space despite differences in phylogenetic diversity. We therefore hypothesize that 31 different clades colonizing grasslands partition the acoustic resource, resulting in convergent 32 community structure across biomes. Many of the birds we recorded are highly threatened, 33 and acoustic monitoring will support conservation measures in these imperiled, yet poorly- 34 studied habitats. 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

56 Introduction 57 The acoustic signals of different bird species may diverge to minimize competitive overlap, 58 leading to overdispersion or uniform distribution of species within acoustic signal space. 59 Each species within the acoustic community occupies a different region of this space 60 (Nelson and Marler 1990; Luther 2009; Krishnan 2019). However, the role of community 61 phylogenetic structure in driving these signal space patterns remains poorly understood. To 62 illustrate, if two biogeographically distinct acoustic communities partition the acoustic 63 resource, they are predicted to exhibit convergent distributions in signal space (trait space), 64 in spite of dissimilar phylogenetic compositions. Alternatively, communities may possess 65 similar or divergent distributions in signal space simply as a consequence of phylogenetic 66 similarity or dissimilarity. Quantifying the species compositions of different bird acoustic 67 communities, their respective signal (or trait) spaces, and phylogenetic similarity are 68 necessary to test these predictions (Webb 2000; Webb et al. 2002; Cavender-Bares et al. 69 2004). 70 Grassland habitats are widespread across both tropical and temperate regions of the world, 71 yet are highly threatened by habitat destruction (Nerlekar and Veldman 2020). Because 72 birds are a visible and vocal component of grassland fauna, they are a valuable indicator of 73 ecosystem health (Vickery and Herkert 1999). Most grassland birds nest on or close to the 74 ground, and many possess highly specific habitat requirements. As a result, habitat 75 destruction is causing steep declines in grassland bird populations (Vickery and Herkert 76 1999; Dutta et al. 2011; Rahmani 2012; Hill et al. 2014; Correll et al. 2019). However, until 77 recently, grassland birds, particularly in tropical regions, have received relatively little study 78 or conservation attention compared to forest birds (Krishnan 2019). This lacuna is 79 particularly pronounced in the Indian Subcontinent, which possesses diverse grassland 80 habitats (Ratnam et al. 2016) occurring along a range of rainfall regimes. These include the 81 semiarid dry grasslands of the Thar desert in the northwest, and the wet alluvial floodplain 82 grasslands of the Gangetic-Brahmaputra floodplains in the northeast (Dabadghao and 83 Shankarnarayan 1973). Although occurring in different ecoregions, bird species are thought 84 to have speciated across the boundary between dry and wet grasslands (Ripley and Beehler 85 1990), and their acoustic communities remain unstudied. An examination of community 86 composition and phylogenetic similarity across these two biomes presents an opportunity to 87 understand how acoustic communities are assembled, and also use this data as a baseline 88 for non-invasive monitoring. 89 Here, we study the avian acoustic communities of dry grasslands in Northwest India, and of 90 wet floodplain grasslands in Northeast India. First, we assess the species compositions, 91 distributions and diversity of birds in these acoustic communities. Secondly, we quantify the 92 signal space occupied by vocal birds in each habitat, and assess whether the two exhibit 93 convergent community structure (i.e. distributions in signal space). Finally, we test whether 94 community structure arises from phylogenetic similarity between the communities (similar 95 species or close relatives in each habitat), or from different bird groups expanding to fill the 96 same signal space (i.e. phylogenetically dissimilar communities). Our findings, some of the 97 first detailed acoustic data from these habitats, have great value in long-term conservation 98 monitoring of threatened grassland biomes and their unique avifauna. 99 100 Materials and Methods 101 Study sites 102 We conducted fieldwork in two protected areas, which contain some of the best-preserved 103 semiarid and floodplain grasslands in India (which we refer to as “dry grassland” and “wet 104 grassland” henceforth) (Figure 1). For wet grasslands, we carried out acoustic sampling in 105 the D’Ering Wildlife Sanctuary in East Siang District, Arunachal Pradesh (27°51’-28°5’ N, 106 95°22’-95°29’ E). This sanctuary consists of seasonally inundated riverine islands, or 107 chaporis, with a Phragmites-Saccharum-Imperata type grassland structure (Dabadghao and 108 Shankarnarayan 1973; Rahmani 2016a). Within this habitat, two grassland types exist: tall 109 (3-4m) grassland that does not undergo burning, and shorter (1m) grassland that is 110 managed using annual controlled burns. The latter also contains forested patches and bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

111 scattered trees such as Bombax ceiba (Rahmani 2016a). The grasslands are in close 112 proximity to water, with human habitation on the other side of the Siang river. We sampled 113 tall-grass habitat in the Borghuli range of the sanctuary, and short-grass habitat in the 114 Anchalghat range on a different chapori. Our landscape-level analysis considered the overall 115 acoustic community across these sites, as tall and short grassland often exist in close 116 proximity in other areas of the floodplain, and therefore a landscape-level analysis is the 117 most appropriate assessment of the overall acoustic community. 118 For dry grasslands, we sampled in the Tal Chhapar Wildlife Sanctuary in Churu District, 119 Rajasthan (27°47′ N, 74°26′ E). This sanctuary is located on the edge of the Thar Desert, 120 containing xerophilous Dichanthium-Cenchrus-Lasiurus type grasslands, with intermixed 121 scrub of species such as Capparis and Prosopis (Dabadghao and Shankarnarayan 1973; 122 Kaur et al. 2020) and man-made water bodies. The village of Chhapar exists at the border of 123 this relatively small sanctuary, and human habitation is adjacent to the grasslands. We 124 sampled sites within the sanctuary, as well as two sites at a community grazing area (or 125 gaushala) 2Km from the sanctuary boundary. This second location contains short grass 126 habitat mixed with Prosopis cineraria trees, and provides additional habitat for grassland and 127 scrub birds. 128 129 Fieldwork and acoustic sampling 130 We sampled winter acoustic communities in both grassland habitats, partly to include winter 131 migrant birds (Grimmett et al. 1998; Rasmussen and Anderton 2005) in our analysis of 132 acoustic communities, but also because the grasslands at D’Ering WLS are inundated and 133 inaccessible for much of the wet season (Rahmani 2016a). Therefore, the winter season is 134 the best time to directly compare both acoustic communities at a similar time of the year. 135 Our study at Tal Chhapar was conducted in late November-early December 2019, and our 136 study at D’Ering shortly afterward in early January 2020. 137 Following previously established methodology (Krishnan 2019), we recorded the dawn 138 singing activity of grassland habitats and used this to quantify the composition, diversity and 139 structure of acoustic communities. Bird song was recorded using either a SongMeter SM4 140 (Wildlife Acoustics Inc., Maynard, MA, USA) or a Sennheiser (Wedemark, Germany) ME62 141 microphone connected to a Zoom (Tokyo, Japan) H6 recorder. In total, we sampled six 142 distinct recording locations in D’Ering WLS, three in the Anchalghat range and three in the 143 Borghuli range. At Tal Chhapar WLS, we sampled five sites, three in the main sanctuary and 144 two in the gaushala (the latter were combined for site-wise analysis). Sites were selected 145 based on the availability of suitable vegetation (extensive patches of grassland/scrub, and a 146 safe place for the recorders). At each site, we began recording at 6AM (early dawn, just after 147 sunrise), and recorded for at least two hours (and often over three, depending on the time 148 taken to reach recorders and stop recording; we censused up to three hours of recording 149 wherever available). We periodically monitored recorders owing to the risk of theft or 150 damage by mammals. 151 To calculate the acoustic abundance index at the landscape level (see below), we used two 152 hours of censused recording to ensure that each sample was equally represented, but for 153 site-wise comparisons (also below), we used the entire censused dataset for each location. 154 In D’Ering WLS, we sampled each site twice, resulting in a total of 36 hours of audio over 12 155 individual recordings. At Tal Chhapar, because there were a smaller number of sampling 156 locations where we could place recorders, we sampled two sites 4 times each, a third 3 157 times and obtained two samples from the gaushala. This was a total of 13 individual 158 recordings representing approximately 36 hours of audio. In one of these samples, an 159 equipment malfunction left us with only one hour of audio, and so we excluded it from the 160 landscape-level abundance (although we included it in site-wise analyses). Therefore, we 161 obtained a comparable amount of audio data from both dry and wet grasslands. 162 163 Acoustic community analysis 164 After dividing each recording into 10-minute segments, we identified all bird species 165 vocalizing in each segment and constructed presence-absence matrices (one matrix for bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

166 each individual recording, i.e. 12 from D’Ering and 13 from Tal Chhapar), where a value of 1 167 meant the species’ vocalizations were detected, and a 0 indicated absence. Two of us 168 censused the data to minimize individual biases in detection, and also cross-verified species 169 identity to minimize identification uncertainties. Using recordings, our observations during 170 sampling, and online song databases, we were able to identify most vocalizations down to 171 species. The few remaining unidentified vocalizations were almost all single detections 172 (much less than 1% of the total) that do not, therefore, influence our community-level 173 analysis. For further analysis, we considered species that had been detected in at least 5% 174 of total 10-minute samples (considering the first two hours of all recordings). Next, we 175 categorised these species (based on published information about their habitats) into 176 grassland residents, grassland migrants and non-grassland species (both resident and 177 migrant) (Ali and Ripley 1997; Grimmett et al. 1998; Rasmussen and Anderton 2005; del 178 Hoyo et al. 2014). The first two categories consisted of species that regularly use grassland 179 habitats, even if not entirely confined to them. The third category contained forest birds such 180 as barbets and orioles, flyovers such as starling flocks and parakeets, waders and waterfowl 181 which were detected near water, and birds from nearby human habitations, such as crows, 182 peafowl and sparrows. Additional notes on this are in Supplementary Data. To control for 183 any subjectivity arising from this classification, we performed analyses both on the entire 184 acoustic community, as well as the subset of grassland species. 185 For each species within the two acoustic communities, we determined an “acoustic 186 abundance index” at the landscape level, using the first two hours of each individual 187 recording (roughly 24 hours of audio total from each community and 48 hours in total, 12 10- 188 minute samples per matrix, across 12 matrices for each community), based on published 189 methods (Krishnan 2019). We used the first two hours of each recording in this analysis to 190 ensure a consistent sample size to calculate acoustic abundance; bird singing activity was 191 typically higher in this time period. The acoustic abundance index represents the probability 192 of detecting a species’ vocalizations if one 10-minute sample were drawn at random from 193 each of the 12 presence-absence matrices (one for each sample). We performed this 194 random draw 10,000 times, and took an average of the proportion of samples that contained 195 the species (Krishnan 2019). Next, we constructed ranked-abundance curves for each 196 habitat, and calculated Jaccard’s similarity index (i.e. shared vocal species between dry and 197 wet grasslands), and the evenness metrics EQ and Evar, (Smith and Wilson 1996) using the 198 codyn (Hallett et al. 2020) package in R (R Core Team 2013). For both metrics, a value of 0 199 implies a non-even community and 1 a perfectly even community. 200 Because we sampled different sites in each sanctuary, and because the number of sampling 201 sites was smaller in the dry grasslands owing to the smaller area of Tal Chhapar (although 202 overall sampling effort was comparable), we also compared species abundance across 203 recording sites. This enabled us to assess how patterns at each site influenced the 204 abundances observed at the landscape level. We calculated, for each species at each 205 recording site, the proportion of 10 minute samples (in the total censused dataset, not just 206 the first two hours) in which that species was detected (referred to as site-wise abundance 207 for convenience). We carried out this analysis for six sites at D’Ering WLS and four at Tal 208 Chhapar WLS (we pooled the two gaushala sites for this analysis). For visual representation, 209 we plotted the coefficient of variation (CV) of this proportion across sites (larger CVs 210 indicated more variance, and more heterogeneous detections across sites, indicating 211 specificity to certain sites or habitats). To quantitatively compare how heterogeneous the 212 distributions of vocal birds were across recording sites, we used the RAC_difference 213 function in the codyn package to calculate the pairwise evenness difference between 214 recording sites (Avolio et al. 2019), as well as the rank change of each species, where larger 215 values indicate more heterogeneity between sites. We performed this analysis separately for 216 both wet and dry grasslands. 217 218 Signal parameter space and acoustic community structure 219 In order to quantify signal space of wet and dry grasslands, we calculated note parameters 220 for each species in the acoustic community using recordings from the databases Xeno- bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

221 Canto (https://www.xeno-canto.org/) and AvoCet 222 (https://avocet.integrativebiology.natsci.msu.edu/) to ensure a good signal:noise ratio when 223 calculating parameters. We used recordings from the same geographical region wherever 224 possible. After labeling the notes (10-20 per species, depending on the number of notes in 225 the recordings) in Raven Pro 1.5 (Cornell Laboratory of Ornithology, Ithaca, NY, USA), we 226 calculated the following 8 parameters: average, maximum and minimum peak frequency, 227 center frequency, bandwidth at 90%, average entropy, note duration, and relative time of 228 peak (i.e. the point during the note at which the peak frequency occurred) (Krishnan 2019; 229 Chitnis et al. 2020). Taking a species average for each of these parameters, we performed a 230 principal components analysis on the correlation matrix using MATLAB (Mathworks Inc., 231 Natick, MA, USA), and used the principal component scores to test for patterns within 232 community signal space (Luther 2009). We performed three statistical analyses: first, we 233 compared the principal component scores of dry and wet grassland acoustic communities to 234 100 randomly drawn uniform distributions spanning the same range of scores, using 235 Kolmogorov-Smirnov tests (Krishnan 2019). By measuring the percentage of times each 236 community fit to a uniform distribution (in this case, P<0.05 means they differed significantly 237 from uniform), we tested whether they showed signatures of dispersion versus clustering in 238 signal space, consistent with divergent acoustic signals within the community. Second, we 239 tested whether the two communities occupied similar regions of signal space by testing 240 whether their respective point clouds in principal component space were drawn from the 241 same distribution, using a MANOVA with Wilk’s lambda as the test statistic on the first 3 242 principal components, and separately on the original parameters as well. The last test we 243 performed built upon this previous one to also test whether the organization of points within 244 principal component space (or community structure) was similar across the two acoustic 245 communities. For this, we calculated the nearest-neighbour distance (NND) for species 246 across communities (Luther 2009; Krishnan 2019). For each species in wet grassland, we 247 calculated the Euclidean distance to its nearest neighbour in the dry grassland communities. 248 If the two communities exhibited similar patterns of community organization in signal space, 249 then the average NND should be lower than that of a randomly drawn community. 250 Therefore, we constructed 10000 randomly drawn “null communities” (Chek et al. 2003) 251 spanning the same range of principal component scores, calculated the average NND to 252 each of these communities, and calculated the Z score of the observed NND versus this 253 distribution of null values. All three statistical tests were performed twice, once for the entire 254 acoustic community and once for only the grassland species. 255 256 Phylogenetic diversity 257 After testing for patterns in community structure using the signal space of both acoustic 258 communities, we tested whether both acoustic communities were phylogenetically similar. 259 For this, we downloaded a meta-tree from the open-source Bird Tree of Life Project 260 (https://birdtree.org/) (Jetz et al. 2012), containing 100 possible phylogenetic hypotheses for 261 all the species within both acoustic communities. In the case of mixed-species starling 262 flocks, we were unable to calculate separate abundance indices for each species because it 263 was difficult to separate their vocalizations. Therefore, we included only one species from 264 each habitat (Acridotheres sp.) in phylogenetic diversity analyses, although we included both 265 in signal space calculations. When measuring phylogenetic diversity, we calculated values 266 for each of the 100 possible trees, giving us a distribution of values. We then compared 267 diversity metrics to each other using paired Wilcoxon signed-rank tests. Using the picante 268 (Kembel et al. 2010) package in R, we calculated three phylogenetic alpha-diversity (within- 269 community) indices (Faith’s phylogenetic diversity (PD), Mean Pairwise Distance (MPD) and 270 Mean Nearest Taxon Distance (MNTD)), and two beta-diversity (between-community) 271 indices (Community Pairwise Distance (CPD), and Community Nearest Taxon Distance 272 (CNTD)) for each acoustic community (Tucker et al. 2017). In addition to the raw values, we 273 also weighted phylogenetic diversity indices by the acoustic abundance index. Finally, to 274 quantify whether the two acoustic communities exhibited congruent phylogenetic structure, 275 we calculated the Phylogenetic Community Dissimilarity (PCD) between the two bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

276 communities (Ives and Helmus 2010). Using consensus trees for both the dry and wet 277 grassland acoustic communities, we calculated PCD by comparison to 10000 randomly 278 reshuffled trees using the inbuilt permutation test in the pcd function of the picante package. 279 This function outputs an overall PCD value, which is close to 1 if the two communities are as 280 similar as expected by random chance, >1 if they are more dissimilar than expected, and <1 281 if they are more similar. The function additionally calculates the contribution of non- 282 phylogenetic components (shared species, PCDc) and phylogenetic components (PCDp). 283 This allowed us to assess whether the PCD value obtained arose because of shared 284 species within the community versus similarities in phylogenetic structure. Again, we 285 calculated all phylogenetic diversity metrics both for the entire acoustic community, and for 286 grassland species. 287 288 Results 289 Avian acoustic communities of wet and dry grasslands 290 We identified vocalizations of 52 bird species in wet grasslands, and 68 in dry grasslands 291 (Supplementary Data). Of these, 31 and 37 species (32 and 38 if both starling species are 292 considered separately, see above) were recorded in >5% of 10-minute samples, and we 293 considered these to constitute the acoustic community (see Figure 2 for examples). These 294 species, classified according to habitat, are listed in Figure 3. Jaccard’s similarity between 295 the dry and wet grassland acoustic communities was 0.06, as they only shared four species 296 (Pycnonotus cafer, Prinia inornata, Streptopelia decaocto and Dendrocitta vagabunda), thus 297 indicating large differences in community composition. 298 Ranked-abundance distributions (Figure 3) demonstrated that higher-ranked species in dry- 299 grassland habitats possessed acoustic abundance indices (across all sites) above 0.75, 300 whereas only two species in wet grasslands had abundance indices above 0.6. Evenness 301 metrics were slightly higher for wet grasslands (after normalizing for overall species 302 diversity). This suggests that the higher abundance of vocally common species in the dry 303 grasslands resulted a steeper drop-off of the abundance distribution, and thus slightly lower 304 evenness across the acoustic community (EQ: wet grasslands 0.22, dry grasslands 0.19, Evar: 305 wet grasslands 0.66, dry grasslands 0.55). 306 307 Heterogeneity across habitats in grassland bird acoustic communities 308 In order to investigate spatial heterogeneity in the acoustic community across recording sites 309 and habitats, we quantified site-wise abundances for each species in both dry and wet 310 grasslands. Site-wise abundances possessed generally higher coefficients of variation in wet 311 grasslands than in dry grasslands, indicating more spatial heterogeneity in the wet grassland 312 acoustic community than in the dry grasslands (Figure 4A-B). However, in general, many 313 species exhibited heterogeneous distributions across both grasslands, as can be seen by a 314 detailed comparison. 315 In the wet grasslands of D’Ering WLS, multiple grassland species (for example Prinia 316 flaviventris, Prinia gracilis, Paradoxornis flavirostris, Chrysomma altirostre and Alaudala 317 raytal) were recorded only within tall grassland. Others (Cisticola exilis, Graminicola 318 bengalensis, Orthotomus sutorius and Francolinus gularis) were recorded either entirely or 319 almost entirely in short grassland, and species such as Turdoides earlei and Timalia pileata, 320 although recorded frequently at all sites, were more frequent in short grassland (Figure 4C). 321 In the dry grasslands of Tal Chhapar WLS, some grassland species (Turdoides malcolmi, 322 Francolinus pondicerianus, Sylvia curruca, Streptopelia decaocto, Lanius excubitor and 323 Pycnonotus cafer) were more or less homogeneously distributed, whereas others (such as 324 Lanius vittatus, Prinia buchanani, Turdoides caudata, Pterocles exustus and Pycnonotus 325 leucotis) were more heterogeneously distributed across the landscape, with different 326 detection levels across recording sites (Figure 4D). 327 In general, although pairwise evenness differences between each recording site were 328 comparable across dry (average of the absolute evenness difference= 0.05) and wet (0.03) 329 grasslands, rank difference (a measure of change in community composition) (Avolio et al. 330 2019) was higher on average for the wet grasslands (0.25, compared to 0.17 for dry bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

331 grasslands), with the three tall grassland sites differing in composition from the short 332 grassland ones (full pairwise comparisons in Supplementary Data). The dry grassland sites 333 also differed little from each other in species composition (Supplementary Data), suggesting 334 that the lower heterogeneity in these habitats is genuine, and not a result of sampling fewer 335 sites. 336 337 Wet and dry grasslands exhibit convergent acoustic community structure 338 Next, we compared the community signal space of wet and dry grassland acoustic 339 communities. The first three principal components (PCs) accounted for about 85% of total 340 variation (Supplementary Data), loading positively on frequency parameters (PC1), average 341 entropy and bandwidth (PC2), and relative time of peak (PC3). Both communities exhibited a 342 dispersed community structure, although the wet grassland community fit 100 randomized 343 uniform distributions better than the dry grassland community (Kolmogorov-Smirnov tests 344 against 100 randomized uniform distributions: wet grasslands, all species= 99%, grassland 345 species=94%, dry grasslands, all species= 72%, grassland species= 64%; percentages 346 indicate number of times out of a 100 where the observed distribution was similar to a 347 uniform distribution at P greater than 0.05) (Supplementary Data) (Krishnan 2019). This 348 dispersion is consistent with divergent signals and low interspecific overlap. Further, 349 MANOVA on both the total acoustic community (Figure 5A) and the grassland species alone 350 (Figure 5B) showed that the two acoustic communities occupied the same regions of signal 351 space (all species: dF within groups=68, dF between groups=1, total dF=69, Wilk’s lambda 352 on PCs= 0.967, P=0.53, on song parameters: lambda= 0.82, P=0.13; grassland species: dF 353 within groups=43, dF between groups=1, total dF=44, Wilk’s lambda on PCs= 0.99, P=0.93, 354 on song parameters: lambda= 0.81, P=0.41). This suggests that the slightly higher clustering 355 observed in the dry grassland community is because of higher species diversity within the 356 same region of signal space. Both communities accumulate species at broadly similar rates 357 with distance from the origin of PC space, further supporting this assertion (Supplementary 358 Figure 1). 359 Finally, a comparison of the observed between-community NNDs to those obtained from 360 10000 randomized null communities showed that the NNDs between dry and wet grassland 361 acoustic communities were significantly lower than expected by chance (Z=-3.43, P<0.01). 362 This result held true even if we only considered grassland species (Z=-3.9, P<0.01), 363 suggesting that both these grassland biomes, in spite of possessing very few species in 364 common, exhibit a convergent community structure in acoustic signal space. 365 366 Dry grassland acoustic communities are higher in phylogenetic diversity 367 Summarizing our results so far, dry grassland acoustic communities exhibit slightly higher 368 species diversity, but convergent community structure with wet grassland acoustic 369 communities. We next calculated measures of phylogenetic diversity (Figure 6), to 370 understand whether similarity in community structure was due to phylogenetic similarity of 371 the acoustic communities. Three measures of alpha diversity, PD, MPD, and MNTD, were all 372 significantly higher for dry grasslands than wet grasslands (Wilcoxon signed-rank test, 373 N=100; W=0, P<0.001 for all three), even after weighting for abundance (Figure 6B-C). The 374 same pattern held when considering only grassland species (Wilcoxon signed-rank test, 375 N=100; W=0, P<0.001 for all three) (Figure 6E-F). Phylogenetic beta-diversity (CPD and 376 CNTD) metrics were of roughly the same magnitude (or slightly lower, see Supplementary 377 Data) as the corresponding alpha diversity metrics (Figure 6D,G). Weighting for abundance 378 significantly lowered CPD but increased CNTD (Wilcoxon signed-rank test of raw versus 379 weighted values, N=100; all species: CPD: W= 4973, CNTD: W= 1504, P<0.001, grassland 380 species: CPD: W= 5050, CNTD: W= 0, P<0.001). Overall, though, between-community 381 phylogenetic distances remained similar to within-community distances. A PCD value of 1.71 382 (all species) and 1.69 (grassland species) indicated greater dissimilarity between the 383 communities than expected by chance. However, the PCDc value (1.87 in both cases) 384 suggested that this difference was driven largely by differences in species composition, as 385 PCDp (the phylogenetic component) was 0.92 (all species) and 0.91 (grassland species) bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

386 (Supplementary Data). Thus, the acoustic communities were neither phylogenetically more 387 similar nor more dissimilar than expected by random chance (Ives and Helmus 2010). 388 Higher phylogenetic diversity in dry grasslands, together with a lack of phylogenetic similarity 389 suggests that phylogenetic considerations alone do not explain convergent community 390 structure. Because these acoustic communities are largely composed of passerine birds, we 391 considered all non-passerines as one group, and the major passerine clades represented in 392 our dataset (Corvoidea, Muscicapoidea, Passeroidea and Sylvioidea) as the other four 393 groups. The dry grassland acoustic community contained 11, 6, 5, 8 and 8 species from 394 each of these five groups respectively. However, the wet grassland acoustic community 395 contained 6, 4, 3, 3, and 16 species respectively (considering the two starling species 396 separately in each). The Sylvioidea (babblers, warblers, bulbuls and larks) (Alström et al. 397 2006) accounted for half the species in the wet grassland acoustic community, with a 398 concomitant expansion in signal space compared to their dry grassland counterparts (Figure 399 7). This expansion of the Sylvioidea in wet grasslands appears to drive convergent signal 400 space and community structure, in spite of the lower phylogenetic diversity in wet 401 grasslands. 402 403 Discussion 404 To summarize, we recorded diverse acoustic communities from both wet and dry grasslands 405 in India. Although the two communities were not significantly phylogenetically similar, both 406 exhibited convergent structure in acoustic signal space. We suggest that this is because the 407 Sylvioidea have expanded in wet grassland signal space to occupy the same acoustic 408 resource as the dry grassland community, even though overall phylogenetic diversity is 409 lower in wet grasslands. 410 411 Community bioacoustics and conservation in South Asian grasslands 412 Because of the multitude of threats facing tropical and subtropical grasslands in India 413 (Ratnam et al. 2016), and the generally poor status of knowledge about their birds, 414 bioacoustics studies have great potential to illuminate their natural history and conservation 415 status (Blumstein et al. 2011; Campos-Cerqueira and Aide 2016; Raynor et al. 2017; Sugai 416 et al. 2019). This is particularly true of the wet grasslands of the Brahmaputra floodplains, 417 which have suffered extensive conversion to agriculture (Rahmani 2016b; Jha et al. 2018). 418 Many threatened birds in this landscape remain very poorly studied. We recorded multiple 419 globally threatened species (Laticilla cinerascens, Paradoxornis flavirostris, Chrysomma 420 altirostre, Pellorneum palustre, Francolinus gularis and Graminicola bengalensis) (Rahmani 421 2012). Protected areas such as Tal Chhapar and D’Ering sanctuaries represent strongholds 422 for many grassland bird species. 423 Grassland bird communities are often spatially heterogeneous (Vickery and Herkert 1999; 424 Hamer et al. 2006; Hovick et al. 2014, 2015; Jacoboski et al. 2017; Londe et al. 2019), and a 425 mix of habitats sustains the greatest diversity of birds. Broadly, our study confirms this 426 pattern in both dry and wet grasslands in India, with more heterogeneity across recording 427 sites in wet grasslands. However, some dry grassland species, particularly those partial to 428 scrub cover (Ali and Ripley 1997; Rasmussen and Anderton 2005), also exhibit different 429 abundances across recording sites. D’Ering WLS possesses distinct tall and short 430 grasslands, compared to Tal Chhapar WLS where both grass and scrub are interspersed 431 across sites. The dominance of the Sylvioidea in the wet grassland acoustic community, 432 many of which are specific to certain grassland types, may also be a factor in these 433 differences. These heterogeneous distributions are likely responsible for the slightly higher 434 evenness in the wet grassland acoustic community, as bird species specific to either tall or 435 short grass have similar abundances at the landscape level. For example, we recorded 436 Graminicola bengalensis at all three short grassland sites, and Chrysomma altirostre at all 437 three tall grassland sites, and thus their abundance indices at the landscape level were 438 similar. Although our study primarily focused on the acoustic community at the landscape 439 level and not on habitat variables, acoustic community structure arising from local habitat 440 heterogeneities is an interesting future study. Our baseline data demonstrate that a bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

441 community-level methodology can uncover spatial patterns within grasslands. We thus 442 highlight the need for detailed spatial studies of bird acoustic communities in India using an 443 occupancy framework (Dorazio et al. 2010; Furnas and Callas 2015; Campos-Cerqueira and 444 Aide 2016; Iknayan and Beissinger 2018; Wood et al. 2019; Abrahams and Geary 2020), 445 both to clarify habitat requirements and to study smaller-scale spatial organization in the 446 acoustic community. 447 448 Acoustic community structure and partitioning of the acoustic resource 449 Although phylogenetic diversity is higher in dry grasslands, acoustic community structure is 450 convergent with wet grasslands, both communities exhibiting overdispersion in signal space. 451 This overdispersion is somewhat less pronounced in dry grasslands, likely because a slightly 452 higher number of species are occupying the same space. However, inter-community NNDs 453 are lower than expected by chance, which is indicative of each species having a counterpart 454 in the signal space of the other community. This may occur because grasslands filter the 455 kinds of birds that occur in them, thus resulting in closely related species replacing each 456 other across habitats. Their signals may resemble each other owing to their shared ancestry, 457 thus leading to convergence in signal space (Cardoso and Price 2010; Tobias et al. 2014). 458 However, our phylogenetic analyses suggest the converse. Firstly, dry grassland acoustic 459 communities have higher phylogenetic diversity, and second, the two grassland communities 460 are no more or less phylogenetically similar than expected by chance (PCDp close to 1). 461 The results of PCD hold even when considering all recorded species (and not just those 462 recorded often enough to form the acoustic community, PCDp=0.9, see Supplementary 463 data), so our results are robust to the definition of the acoustic community that we employ 464 here. This suggests that although some close relatives may replace each other between 465 habitats (further supported by the values of between-community CPD and CNTD being 466 within the same range as the within-community MPD and MNTD), convergence is also 467 responsible for the observed similarity in acoustic community structure. Some species are 468 replaced by congeners between dry and wet grasslands (eg. Francolinus, Turdoides) (Ripley 469 and Beehler 1990; Ali and Ripley 1997; Rasmussen and Anderton 2005). However, others 470 are unique to each type of grassland. For example, Pterocles is found only in semiarid 471 habitats, and Paradoxornis only in wet grasslands (Ali and Ripley 1997; Grimmett et al. 472 1998; Rasmussen and Anderton 2005; Rahmani 2012). The fact that acoustic community 473 structure is similar in spite of these phylogenetic differences is consistent with convergence 474 in signal space. A similar pattern holds even when considering just grassland species, 475 indicating that the presence of villages or wooded patches does not change patterns in 476 community structure. 477 The Sylvioidea (babblers, warblers, bulbuls and larks in our dataset) comprises half of the 478 wet grassland acoustic community. This overrepresentation compared to the dry grasslands 479 is accompanied by an increase in the signal space occupied by this clade (Figure 7), such 480 that the overall signal space of the two communities is convergent. The Eastern Himalayas 481 and their foothills possess the highest babbler diversity in the Indian Subcontinent 482 (Srinivasan et al. 2014). This biogeography may partly explain why the acoustic community 483 of the wet grasslands (occupying this region) is dominated by this clade. The expansion of 484 the Sylvioidea’s signal space, resulting in convergent community structure with Northwestern 485 dry grasslands, is a compelling indicator of acoustic resource partitioning. This is further 486 supported by our analyses suggesting that species within each community are dispersed 487 across acoustic signal space (Figure 5). Thus, we hypothesize that grassland bird acoustic 488 communities assemble by occupying the available acoustic signal space (analogous to filling 489 available niches) (Price et al. 2014), resulting in convergent community structure across 490 different grassland habitats. This pattern may arise to minimize acoustic interference 491 between relatives (Schmidt et al. 2013), or as an indirect outcome of morphological 492 divergence between coexisting species (Krishnan and Tamma 2016). The drivers of acoustic 493 community assembly are thus a compelling subject for future study. Quantifying community- 494 level signal space is a valuable way to establish patterns using passive acoustic data. Thus, bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

495 we suggest combining these methods with phylogenetic analyses to study spatiotemporal 496 change in global biodiversity. 497 498 Acknowledgments 499 We are deeply indebted to the forest departments and chief wildlife wardens of Arunachal 500 Pradesh (Letter number CWL/G/173/2018-19/Pt.VII/1585-86) and Rajasthan (Letter number 501 F19() permission/cwlw/2017/1598) for permission and support to carry out acoustic 502 sampling. For Arunachal Pradesh, we thank Mr. Tasang Taga, DFO, D’Ering WLS, Mr. 503 Maksam Tayeng, Arunachal State Biodiversity Board, Taksh Sangwan and the range forest 504 officers and forest guards of Borghuli and Anchalghat for logistical support, hospitality and 505 accompanying us during sampling for safety. For Rajasthan, we thank Ram Mohan, Mr. 506 Yogendra Rathod, the ACF’s office and the staff at Tal Chhapar WLS for hospitality and 507 logistical support. We also thank Savithri Singh and Anand Prasad for assistance during 508 sampling, Ram Mohan, Taksh Sangwan, Roon Bhuyan and Siva R for permission to use 509 photos, Ramana Athreya and Uma Ramakrishnan for logistical advice, Viral Joshi for help 510 with identifying bird calls, Krishnapriya Tamma and members of our research group for 511 feedback. 512 513 Funding 514 AK is funded by an INSPIRE Faculty Award from the Department of Science and 515 Technology and an Early Career Research Award (ECR/2017/001527) from the Science and 516 Engineering Board (SERB), Government of India. SL received funding for this project from 517 an Oriental Bird Club Conservation Grant, equipment support from Idea Wild, and a Rufford 518 Foundation Small Grant (awarded to Ram Mohan, in collaboration with SL who sampled 519 birds). 520 521 References 522 Abrahams, C., and M. Geary. 2020. Combining bioacoustics and occupancy modelling for 523 improved monitoring of rare breeding bird populations. Ecological Indicators 112:106131. 524 Ali, S., and S. D. Ripley. 1997. Handbook of the Birds of india and Pakistan (2nd ed.). 525 Oxford University Press, New Delhi. 526 Alström, P., P. G. P. Ericson, U. Olsson, and P. Sundberg. 2006. Phylogeny and 527 classification of the avian superfamily Sylvioidea. Molecular Phylogenetics and Evolution 528 38:381–397. 529 Avolio, M. L., I. T. Carroll, S. L. Collins, G. R. Houseman, L. M. Hallett, F. Isbell, S. E. 530 Koerner, et al. 2019. A comprehensive approach to analyzing community dynamics using 531 rank abundance curves. Ecosphere 10:e02881. 532 Blumstein, D. T., D. J. Mennill, P. Clemins, L. Girod, K. Yao, G. Patricelli, J. L. Deppe, et al. 533 2011. Acoustic monitoring in terrestrial environments using microphone arrays: Applications, 534 technological considerations and prospectus. Journal of Applied Ecology 48:758–767. 535 Campos-Cerqueira, M., and T. M. Aide. 2016. Improving distribution data of threatened 536 species by combining acoustic monitoring and occupancy modelling. Methods in Ecology 537 and Evolution 7:1340–1348. 538 Cardoso, G. C., and T. D. Price. 2010. Community convergence in bird song. Evolutionary 539 Ecology 24:447–461. 540 Cavender-Bares, J., D. D. Ackerly, D. A. Baum, and F. A. Bazzaz. 2004. Phylogenetic 541 overdispersion in Floridian oak communities. The American Naturalist 163:823–43. 542 Chek, A. A., J. P. Bogart, and S. C. Lougheed. 2003. Mating signal partitioning in multi- 543 species assemblages: A null model test using frogs. Ecology Letters 6:235–247. 544 Chitnis, S. S., S. Rajan, and A. Krishnan. 2020. Sympatric wren-warblers partition acoustic 545 signal space and song perch height. Behavioral Ecology 31:559–567. 546 Correll, M. D., E. H. Strasser, A. W. Green, and A. O. Panjabi. 2019. Quantifying specialist 547 avifaunal decline in grassland birds of the Northern Great Plains. Ecosphere 10. 548 Dabadghao, P. M., and K. A. Shankarnarayan. 1973. The grass cover of India. Indian 549 Council of Agricultural Research, New Delhi. bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

550 del Hoyo, J., N. J. Collar, D. A. Christie, A. Elliott, and L. D. C. Fishpool. 2014. HBW and 551 BirdLife International Illustrated Checklist of the Birds of the World. Lynx Edicions BirdLife 552 International. 553 Dorazio, R. M., M. Kéry, J. A. Royle, and M. Plattner. 2010. Models for inference in dynamic 554 metacommunity systems. Ecology 91:2466–2475. 555 Dutta, S., A. R. Rahmani, and Y. V. Jhala. 2011. Running out of time? The great Indian 556 bustard Ardeotis nigriceps-status, viability, and conservation strategies. European Journal of 557 Wildlife Research 57:615–625. 558 Furnas, B. J., and R. L. Callas. 2015. Using automated recorders and occupancy models to 559 monitor common forest birds across a large geographic region. Journal of Wildlife 560 Management 79:325–337. 561 Grimmett, R., C. Inskipp, and T. Inskipp. 1998. Birds of the Indian Subcontinent (1st Editio.). 562 A&C Black. 563 Hallett, L., M. Avolio, I. Carroll, S. Jones, A. MacDonald, D. Flynn, P. Slaughter, et al. 2020. 564 codyn: Community Dynamics Metrics. 565 Hamer, T. L., C. H. Flather, and B. R. Noon. 2006. Factors associated with grassland bird 566 species richness: The relative roles of grassland area, landscape structure, and prey. 567 Landscape Ecology 21:569–583. 568 Hill, J. M., J. F. Egan, G. E. Stauffer, and D. R. Diefenbach. 2014. Habitat availability is a 569 more plausible explanation than insecticide acute toxicity for U.S. grassland bird species 570 declines. PLoS ONE 9. 571 Hovick, T. J., R. Dwayne Elmore, and S. D. Fuhlendorf. 2014. Structural heterogeneity 572 increases diversity of non-breeding grassland birds. Ecosphere 5. 573 Hovick, T. J., R. D. Elmore, S. D. Fuhlendorf, D. M. Engle, and R. G. Hamilton. 2015. Spatial 574 Heterogeneity Increases Diversity and Stability in Grassland Bird Communities. Ecological 575 Applications 25:662–72. 576 Iknayan, K. J., and S. R. Beissinger. 2018. Collapse of a desert bird community over the 577 past century driven by climate change. Proceedings of the National Academy of Sciences of 578 the United States of America 115:8597–8602. 579 Ives, A. R., and M. R. Helmus. 2010. Phylogenetic metrics of community similarity. American 580 Naturalist 176:E128–E142. 581 Jacoboski, L. I., R. K. Paulsen, and S. M. Hartz. 2017. Bird-grassland associations in 582 protected and non-protected areas in southern Brazil. Perspectives in Ecology and 583 Conservation 15:109–114. 584 Jetz, W., G. H. Thomas, J. B. Joy, K. Hartmann, and a. O. Mooers. 2012. The global 585 diversity of birds in space and time. Nature 491:444–448. 586 Jha, R. R. S., J. J. Thakuri, A. R. Rahmani, M. Dhakal, N. Khongsai, N. M. B. Pradhan, N. 587 Shinde, et al. 2018. Distribution, movements, and survival of the critically endangered 588 Bengal Florican Houbaropsis bengalensis in India and Nepal. Journal of Ornithology 589 159:851–866. 590 Kaur, M., P. Joshi, K. Sarma, and S. K. Das. 2020. Assessment of plant community structure 591 in Tal Chhapar Wildlife Sanctuary, Rajasthan, India. Species 21:126–139. 592 Kembel, S. W., P. D. Cowan, M. R. Helmus, W. K. Cornwell, H. Morlon, D. D. Ackerly, S. P. 593 Blomberg, et al. 2010. Picante: R tools for integrating phylogenies and ecology. 594 Bioinformatics 26:1463–1464. 595 Krishnan, A. 2019. Acoustic community structure and seasonal turnover in tropical South 596 Asian birds. Behavioral Ecology 30:1364–1374. 597 Krishnan, A., and K. Tamma. 2016. Divergent morphological and acoustic traits in sympatric 598 communities of Asian barbets. Royal Society Open Science 3:160117. 599 Londe, D. W., S. D. Fuhlendorf, R. D. Elmore, and C. A. Davis. 2019. Landscape 600 heterogeneity influences the response of grassland birds to energy development. Wildlife 601 Biology 2019:1–11. 602 Luther, D. 2009. The influence of the acoustic community on songs of birds in a neotropical 603 rain forest. Behavioral Ecology 20:864–871. 604 Nelson, D. A., and P. Marler. 1990. The perception of birdsong and an ecological concept of bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

605 signal space. Pages 443–478 in W. C. Stebbins and M. A. Berkley, eds. Comparative 606 Perception: Complex Signals, Vol 2. Wiley, New York. 607 Nerlekar, A. N., and J. W. Veldman. 2020. High plant diversity and slow assembly of old- 608 growth grasslands. Proceedings of the National Academy of Sciences of the United States 609 of America 1–7. 610 Price, T. D., D. M. Hooper, C. D. Buchanan, U. S. Johansson, D. T. Tietze, P. Alström, U. 611 Olsson, et al. 2014. Niche filling slows the diversification of Himalayan songbirds. Nature 612 509:222–5. 613 R Core Team, . 2013. R: A language and environment for statistical computing. R 614 Foundation for Statistical Computing, Vienna, Austria. 615 Rahmani, A. R. 2012. Threatened Birds of India: Their Conservation Requirements (Bombay 616 Natural History Society). Oxford University Press, New Delhi. 617 Rahmani, A. R. 2016a. D’Ering Memorial Wildlife Sanctuary: Report of Summer surveys in 618 2016. Mumbai. 619 ———. 2016b. Conservation of Threatened Grassland birds of the Brahmaputra floodplains. 620 Mumbai. 621 Rasmussen, P. C., and J. C. Anderton. 2005. Birds of South Asia: The Ripley Guide. Lynx 622 Edicions, Barcelona. 623 Ratnam, J., K. W. Tomlinson, D. N. Rasquinha, and M. Sankaran. 2016. Savannahs of Asia: 624 Antiquity, biogeography, and an uncertain future. Philosophical Transactions of the Royal 625 Society B: Biological Sciences 371. 626 Raynor, E. J., C. E. Whalen, M. B. Brown, and L. A. Powell. 2017. Grassland bird community 627 and acoustic complexity appear unaffected by proximity to a wind energy facility in the 628 Nebraska Sandhills. The Condor 119:484–496. 629 Ripley, S. D., and B. M. Beehler. 1990. Patterns of Speciation in Indian Birds. Journal of 630 Biogeography 17:639. 631 Schmidt, A. K. D., H. Römer, and K. Riede. 2013. Spectral niche segregation and 632 community organization in a tropical cricket assemblage. Behavioral Ecology 24:470–480. 633 Smith, B., and J. B. Wilson. 1996. A Consumer’s Guide to Evenness Indices. Oikos 76:70. 634 Srinivasan, U., K. Tamma, and U. Ramakrishnan. 2014. Past climate and species ecology 635 drive nested species richness patterns along an east-west axis in the Himalaya. Global 636 Ecology and Biogeography 23:52–60. 637 Sugai, L. S. M., T. S. F. Silva, J. W. Ribeiro, and D. Llusia. 2019. Terrestrial Passive 638 Acoustic Monitoring: Review and Perspectives. BioScience 69:5–11. 639 Tobias, J. A., R. Planque, D. L. Cram, and N. Seddon. 2014. Species interactions and the 640 structure of complex communication networks. Proceedings of the National Academy of 641 Sciences 111:1020–1025. 642 Tucker, C. M., M. W. Cadotte, S. B. Carvalho, T. Jonathan Davies, S. Ferrier, S. A. Fritz, R. 643 Grenyer, et al. 2017. A guide to phylogenetic metrics for conservation, community ecology 644 and macroecology. Biological Reviews 92:698–715. 645 Vickery, P. D., and J. R. Herkert. 1999. Ecology and Conservation of Grassland birds of the 646 Western Hemisphere. (P. D. Vickery & J. R. Herkert, eds.)Studies in Avian Biology (Vol. 19). 647 Cooper Ornithological Society, Camarillo, CA, USA. 648 Webb, C. O. 2000. Exploring the Phylogenetic Structure of Ecological Communities: An 649 Example for Rain Forest Trees. The American Naturalist 156:145–155. 650 Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue. 2002. Phylogenies and 651 Community Ecology. Annual Review of Ecology and Systematics 33:475–505. 652 Wood, C. M., V. D. Popescu, H. Klinck, J. J. Keane, R. J. Gutiérrez, S. C. Sawyer, and M. Z. 653 Peery. 2019. Detecting small changes in populations at landscape scales: a bioacoustic site- 654 occupancy framework. Ecological Indicators 98:492–507. 655 656 657 Figure legends 658 Figure 1: Locations of the two sanctuaries, showing the various habitats sampled. Photos 659 courtesy Taksh Sangwan (left) and Ram Mohan (right). bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

660 661 Figure 2: (A,B) Some of the birds comprising the acoustic community in wet (A) and dry (B) 662 grasslands. (A) top row, left to right, photographer’s name in brackets: Paradoxornis 663 flavirostris (Taksh Sangwan), Chrysomma altirostre (Roon Bhuyan), Prinia flaviventris (SL), 664 middle row: Turdoides earlei (SL), Pellorneum palustre (Siva R), Alaudala raytal (SL), 665 bottom row: Psilopogon asiaticus (SL), Graminicola bengalensis (Taksh Sangwan). (B) top 666 row: Pycnonotus leucotis (SL), Pterocles exustus (SL), Pastor roseus (AK), middle row: 667 Calandrella brachydactyla (SL), Turdoides caudata (AK), Francolinus pondicerianus (AK), 668 bottom row: Prinia buchanani (AK), Sylvia curruca (SL). (C,D) Spectrograms of sample 669 vocalizations recorded during data collection in wet (C) and dry (D) grasslands. 670 671 Figure 3: Ranked abundance distributions for both acoustic communities using the acoustic 672 abundance index. The box contains species names in order of abundance, and color-coded 673 by habitat preference. 674 675 Figure 4: (A,B) Ranked abundance distributions (black bars) with CV of site-wise 676 abundances for each species (grey bars). A higher CV indicates more heterogeneous 677 distributions. (C,D) Species-wise (rows) values of site-wise abundance across recording 678 sites (columns), with warmer colors indicating higher abundance. A checkerboard pattern is 679 indicative of more heterogeneity in species distributions across recording sites. Species are 680 color-coded according to their habitat preferences; note in particular the species in black 681 (grassland residents), for whom patterns are discussed in the text. Laticilla cinerascens is 682 included here solely for conservation interest on account of its rarity. 683 684 Figure 5: Biplot of the first two canonical variables from a MANOVA on the principal 685 components of acoustic parameters. These parameters were measured for each species in 686 the two acoustic communities to derive a signal space, plotted here as a minimum convex 687 polygon using MATLAB. Note the overlap in polygons both when considering all species (A) 688 and only grassland species (B). 689 690 Figure 6: (A) Majority-rule consensus trees representing the phylogenetic relationships of 691 species in both acoustic communities, color-coded by habitat preference as in earlier figures. 692 Thickness of terminal branches is in proportion to abundance index for each species. (B-G) 693 Phylogenetic alpha diversity (B,C,E,F) and beta diversity (D,G) metrics for both acoustic 694 communities, both for all species (B-D) and grassland species (E-G). In C,D,F and G, 695 indices are also calculated after weighting for abundance. In all scenarios, dry grasslands 696 possess higher phylogenetic diversity. 697 698 Figure 7: Consensus trees for each acoustic community with the major clades color-coded, 699 and their respective signal spaces represented by minimum convex polygons in the centre. 700 Wet grasslands are on the left (blue outline on polygons), dry grasslands on the right (red 701 outline), and the polygons are colored according to the clade they represent. The numbers 702 above each polygon represent its area divided by the area of the minimum convex polygon 703 containing all species (see Figure 5). Note the dominance of the Sylvioidea in wet 704 grasslands. 705 bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Abundance Wet grasslands- D'Ering 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

0 1 Dry grasslands- Tal Chhapar Grassland resident Grassland migrant Non-grassland resident/migrant

Species by rank 01 02 035 30 25 20 15 10 5 D'Ering Tal Chhapar

Corvus macrorhynchos Turdoides malcolmi Turdoides earlei Sylvia curruca Francolinus pondicerianus Timalia pileata Psittacula krameri Gallus gallus Corvus splendens Pycnonotus cafer Lanius excubitor Prinia flaviventris Anthus campestris Anthus hodgsoni Pycnonotus cafer Prinia gracilis Streptopelia decaocto Paradoxornis flavirostris Dicrurus macrocercus Cisticola exilis Pavo cristatus Orthotomus sutorius Prinia buchanani Tadorna ferruginea Vanellus indicus Streptopelia decaocto Ficedula parva

Rank Prinia inornata Pycnonotus leucotis Halcyon smyrnensis Psilopogon lineatus Turdoides caudata Acrocephalus agricola Calandrella brachydactyla Graminicola bengalensis Pastor roseus/Acridotheres tristis Tringa ochropus Motacilla flava Dendrocitta vagabunda Chrysomma altirostre Lanius vittatus Dendrocitta vagabunda Euodice malabarica Oriolus xanthornus Anthus similis Phylloscopus affinis Prinia inornata Lanius tephronotus Copsychus fulicatus Psilopogon asiaticus Pericrocotus cinnamomeus Anthus trivialis Saroglossa spilopterus/ Cinnyris asiaticus Acridotheres fuscus Oenanthe sp. Motacilla alba Gymnoris xanthocollis Francolinus gularis Passer domesticus Pellorneum palustre Grus grus Phylloscopus fuscatus Anser indicus Muscicapidae sp. (Phoenicurus auroreus) Pterocles exustus Alaudala raytal Motacilla cinerea Merops orientalis A bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this versionB posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. 2.5 Wet grasslands- D'Ering 2.5 Dry grasslands- Tal Chhapar Abundance Abundance Heterogeneity Heterogeneity

2 2

1.5 1.5 Index value index value 1 1

0.5 0.5

0 0 5 10 15 20 25 30 5 10 15 20 25 30 35 Rank Rank C D 1 Turdoides earlei Turdoides malcolmi 1 Timalia pileata Francolinus pondicerianus Pycnonotus cafer Lanius excubitor Prinia flaviventris 0.9 Pycnonotus cafer 0.9 Prinia gracilis Streptopelia decaocto Dicrurus macrocercus Paradoxornis flavirostris 0.8 Prinia buchanani Cisticola exilis Pycnonotus leucotis 0.8 Orthotomus sutorius Turdoides caudata Streptopelia decaocto Lanius vittatus Prinia inornata 0.7 Euodice malabarica Prinia inornata 0.7 Graminicola bengalensis Copsychus fulicatus Chrysomma altirostre Cinnyris asiaticus Francolinus gularis 0.6 Pterocles exustus 0.6 Pellorneum palustre Merops orientalis Alaudala raytal Sylvia curruca Laticilla cinerascens 0.5 Anthus campestris Anthus hodgsoni Ficedula parva 0.5 Acrocephalus agricola Calandrella brachydactyla Motacilla flava Anthus similis 0.4 Anthus trivialis 0.4 Phylloscopus affinis Oenanthe sp. Lanius tephronotus Psittacula krameri Motacilla alba Corvus splendens Phylloscopus fuscatus 0.3 Site-wise abundance Pavo cristatus 0.3 Corvus macrorhynchos Vanellus indicus Site-wise abundance Halcyon smyrnensis Gallus gallus Pastor roseus/Acridotheres tristis Tadorna ferruginea 0.2 Tringa ochropus 0.2 Psilopogon lineatus Dendrocitta vagabunda Dendrocitta vagabunda Pericrocotus cinnamomeus Oriolus xanthornus 0.1 Gymnoris xanthocollis 0.1 Psilopogon asiaticus Passer domesticus Saroglossa spilopterus/Acridotheres fuscus Grus grus Muscicapidae sp. (Phoenicurus auroreus) Anser indicus 0 Motacilla cinerea 0 Site 3 Site 1 Site 1 Site 2 Site 3 Site 2 Site 1 Site 2 Site 3 Anchalghat Borghuli Site 4/5 Tal Chhapar 'Gaushala' bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this version posted August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

A Overall Acoustic Communities 2.5

2

1.5 Dry grasslands- Tal Chhapar 1

0.5 Wet grasslands- D'Ering

0

-0.5

-1

-1.5 Second canonical variable (MANOVA) -2

-2.5 -3 -2 -1 0 1 2 3 First canonical variable (MANOVA)

Grassland Species B 2.5 )

A 2

NOV 1.5 A 1

0.5

l variable (M 0

-0.5

-1 d canonica -1.5

Secon -2

-2.5 -3 -2 -1 0 1 2 3 First canonical variable (MANOVA) A Pavo cristatus Gallus gallus Francolinus pondicerianus Francolinus gularis bioRxiv preprint doi: https://doi.org/10.1101/2020.08.07.241612; this versionAnser posted indicus August 7, 2020. The copyright holder for this preprint (which was not certified by peer review) isTadorna the ferrugineaauthor/funder. All rightsTringa ochropus reserved. No reuse allowed without permission. Psilopogon asiaticus Vanellus indicus Psilopogon lineatus Merops orientalis Dendrocitta vagabunda Halcyon smyrnensis Psittacula krameri Corvus macrorhynchos Pericrocotus cinnamomeus Lanius tephronotus Dendrocitta vagabunda Oriolus xanthornus Corvus splendens Saroglossa spilopterus Lanius vittatus Acridotheres fuscus Lanius excubitor Dicrurus macrocercus Phoenicurus auroreus Pastor roseus Anthus hodgsoni Acridotheres tristis Motacilla alba Oenanthe picata Motacilla flava Ficedula parva Turdoides earlei Copsychus fulicatus Anthus trivialis Graminicola bengalensis Anthus similis Pellorneum palustre Anthus campestris Timalia pileata Motacilla cinerea Paradoxornis flavirostris Gymnoris xanthocollis Chrysomma altirostre Passer domesticus Euodice malabarica Orthotomus sutorius Cinnyris asiaticus Prinia flaviventris Turdoides malcolmi Prinia gracilis Turdoides caudata Prinia inornata Sylvia curruca Cisticola exilis Prinia buchanani Prinia inornata Pycnonotus cafer Pycnonotus leucotis Acrocephalus agricola Pycnonotus cafer Phylloscopus fuscatus Calandrella brachydactyla Phylloscopus affinis Pterocles exustus Alaudala raytal Streptopelia decaocto Streptopelia decaocto Grus grus

10.0 10.0 Total acoustic community B C D 35 160 160

Wet Grassland MNTD- Wet Grassland CPD 30 Dry Grassland 140 MNTD- Dry Grassland 140 CNTD MPD- Wet Grassland MPD- Dry Grassland 25 120 120

20 100 100

Count 15 80 80

10 60 60

5 40 40

0 20 20

1000 1200 1400 1600 1800 20 40 60 80 100 120 140 160 Phylogenetic beta (abundance diversity weighted) 20 40 60 80 100 120 140 160

Phylogenetic Diversity Phylogenetic alpha diversity (abundance weighted) Phylogenetic alpha diversity Phylogenetic beta diversity E F Grassland species G 30 140 140 Wet Grassland MNTD- Wet Grassland Dry Grassland CPD MNTD- Dry Grassland CNTD 25 120 MPD- Wet Grassland 120 MPD- Dry Grassland

20 100 100

15 80 80 Count

10 60 60

5 40 40

0 20 20 600 700 800 900 1000 20 40 60 80 100 120 140 20 40 60 80 100 120 140 Phylogenetic alpha diversity (abundance weighted)

Phylogenetic Diversity Phylogenetic beta (abundance diversity weighted) Phylogenetic alpha diversity Phylogenetic beta diversity bioRxiv preprint doi: Wethttps://doi.org/10.1101/2020.08.07.241612 grasslands ; this version posted August 7, 2020.Dry The grasslands copyright holder for this preprint Non-passerine (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. Non-passerine 2 0.11 2 0.48

0 0

-2 -2 -2 0 2 -2 0 2 Corvoidea 2 0.16 2 0.07 Corvoidea Muscicapoidea 0 0

-2 -2 -2 0 2 -2 0 2 Muscicapoidea

2 0.02 2 0.26 Passeroidea 0 0 Passeroidea

-2 -2 -2 0 2 -2 0 2

2 0.03 2 0.35 MANOVA 2nd canonical variable MANOVA 0 0

-2 -2 -2 0 2 -2 0 2

2 2 0.56 0.36 Sylvioidea Sylvioidea 0 0

-2 -2 -2 0 2 -2 0 2

10.0 MANOVA 1st canonical variable 10.0