A continental measure of urbanness predicts avian response to local urbanization

1 Understanding species-specific relationships with their environment is essential for ecology, 2 biogeography, and conservation biology. Moreover, understanding how these relationships 3 change with spatial scale is critical to mitigating potential threats to biodiversity. But methods 4 which measure inter-specific variation in response to environmental parameters that are also 5 generalizable across multiple spatial scales are scarce. We used broad-scale avian citizen science 6 data, over continental Australia, integrated with remotely-sensed products, to produce a measure 7 of urban-tolerance for a given species at a continental-scale. We then compared these urban-8 tolerances to modelled responses to urbanization at a local-scale, based on systematic sampling 9 within four small cities. For 49 species which had sufficient data for modelling, we found a 10 significant relationship (R 2 = 0.51) between continental-scale urbanness and local-scale 11 urbanness. We also found that relatively few citizen science observations (~250) are necessary 12 for reliable estimates of continental-scale species-specific urban scores to predict local-scale 13 response to urbanization. Our approach demonstrates the applicability of broad-scale citizen 14 science data, contrasting both the spatial grain and extent of standard point-count surveys generally only conducted at small spatial scales. Continental-scale responses in Australia are representative of small-scale responses to urbanization among four small cities in Australia, suggesting that our method of producing species-specific urban scores is robust and may be 18 generalized to other locations lacking appropriate data.


Introduction 23
Understanding species-environment relationships (Mertes and Jetz 2018) is a critical and 24 unifying goal in ecology (Hutchinson 1953, Levin 1992, biogeography ( of urbanization (i.e., the least urban-tolerant species), then environmental planners can attempt to 60 mitigate the threats specific to these least-tolerant species ( birdwatchers who submit bird observations in the form of 'checklists' -defined as a list of birds 131 seen or heard in a specified area. An extensive network of regional volunteers (Gilfedder et al. 132 2018) use their local expertise to provide filters for the submissions, limiting observations based 133 on unexpected species or abundances of species. If an observation trips a 'filter' then it is 134 reviewed before inclusion in the database. More detailed information on eBird protocols are 135 provided in Sullivan et al. (2014). 136

Species-specific scores 138
We used continental eBird data to assign species-specific urban scores for each species in the 139 analysis. This approach borrows from the longstanding theory behind urban adapters, avoiders, 140 and exploiters (Blair 1996, McDonnell and Hahs 2015, Geschke et al. 2018, and works 141 theoretically by assessing how a species responds to a continuous level of urbanization (Fig. 1). 142 include species in an area that was recorded from a great distance away). This was done by 156 including only complete eBird checklists -where the observer recorded all birds heard and/or 157 seen -from mainland Australia, which followed the travelling, random, stationary, area, or 158 BirdLife Australia protocols. We also filtered these checklists to those which recorded birds 159 between 5-240 minutes and travelled less than 5 km or less than 500 Ha area searches (La Sorte scale assignment was 32,642 ± 32,846 (sd). All but three species (Spotted Quail-Thrush, 169 Pilotbird, Beautiful Firetail) in our analysis had > 1000 continental eBird observations (Table  170 S1), and these were removed from analyses because they did not meet the minimum local-scale 171 observation threshold (see below). Following filtering, each eBird checklist was assigned a 172 measure of urbanization -on a continuous scale. This was done by taking the average radiance 173 of night-time lights within a 5 km buffer of each checklist. A buffer was used to minimize any 174 bias in eBird sampling protocols (e.g., mis-placement of eBird checklists by participants, and to 175 account for travelling checklists throughout an area) and the size of the buffer has no discernible 176 influence on the relative urban-score differences among species (Callaghan et al. 2019a). We The percent impervious surface was chosen as it is a direct measure of urbanization, and 207 generally readily available at local-scales for urban planners, whereas VIIRS night-time lights is 208 at 500-m resolution, not generally applicable at a small-scale. Hence, our approach compared 209 different spatial grains, albeit measuring the same environmental response in urbanization. 210

211
We extracted species-specific responses to urbanization at a local scale, using a modelling 212 approach and generated parameter estimates for each species, that were treated as the 'local-scale 213 urbanness'. The response variable in our models was the total number of presences (i.e., if a 214 species occurred in a 5 min sampling event) for each point (N=24) -i.e., the number of 215 presences for a species at a given sampling point. The total number of presences possible was 24, 216 given each survey point was sampled 24 times. The response variable was 'zero-filled', 217 accounting for complete absences of a given species at a given point, and each species thus had a 218 total sample size of 24 observations which were modelled. This response variable was modelled 219 against the percent impervious area at each survey point (N=24). We fitted Generalized Linear 220 Mixed Models (GLMMs; Bolker et al. 2009) with a Poisson distribution, where the random 221 effect was transect (i.e., city). This model was separately fitted to each species, and the 222 regression coefficient for the impervious surface area predictor for a given species was taken as 223 the species-specific response to urbanization at a local scale. Only species with a minimum of 10 224 presences across all surveys (out of a possible 576) were considered for the GLMMs, ensuring 225 that models would converge. Although species in the study region can show some seasonal 226 movement, this was not included in our models to minimize over-fitting, given the sample size of 227 the number of points (N=24). Additionally, many of the seasonal species were excluded from 228 analyses based on our cut-off for minimum of nonzero observations (i.e., many of the possible 229 migrants were only recorded <10 times). Our initial exploration considered negative binomial 230 model distributions, but AIC was consistently lower for Poisson than negative binomial, and

Regression of continental and local-scale urban measures 242
We observed a total of 94 species on our local-scale bird surveys (Table S1). Fifty-one species 243 had > 10 presences across all surveys (Table S1) and were thus considered for GLMMs. After 244 initial modelling, two species were further eliminated from analyses as their estimates from the 245 GLMM were outliers when compared with the rest of the dataset (Pilotbird and White-eared 246 Honeyeater; Appendix S1), likely resulting from a small sample size. Thus, 49 species were used 247 in our regression of continental and local-scale urban tolerance measures, with their continental-248 scale species-specific urban scores being log-transformed. Models were fitted using the 'lm' 249 function in R. We fitted this model first without any weighting, and then re-fitted the model by

265
A total of 94 species were observed on our local-level transects (Table S1). The species that was 266 most likely to be associated with urbanization at the local-scale was Rock Pigeon (parameter 267 estimate: 0.14), while the species least likely to be associated with urbanization at a local-scale 268 was Rufous Whistler (parameter estimate: -0.08; Fig. S3; full model results, including 269 significance of GLMMs can be found in Table S2). Of the 94 potential species, Rock Pigeon had 270 the highest continental-scale species-specific urban score (12.49) while Red-capped Robin had 271 the lowest continental-scale species-specific urban score (0.047). Of the 49 species included in 272 analyses, the mean urban score was 2.37 ± 2.81 (Fig. S4). Thus, Rock Pigeon had both the 273 highest local-urban score and continental-urban species-specific score showing some qualitative 274 agreement between the two approaches. Similarly, Superb Lyrebird had the second lowest local-275 urban score and the lowest continental-urban species-specific score (cf. Fig. S5 and Fig. S6). 276 Some species (e.g., Crested Pigeon, Spotted Pardalote, New Holland Honeyeater) had relatively 277 high continental-scale urban scores (i.e., ranked in the top 50%) but were still negatively 278 associated with urbanization at the local-scale. Conversely, some species (e.g., Gray Butcherbird, 279 Satin Bowerbird) had relatively low continental-scale urban scores (i.e., ranked in the bottom 280 50%) but were positively associated with urbanization at the local scale (cf. Fig. S5 and Fig. S6). 281

282
Continental species-specific urban scores significantly predicted (t=6.95, df=47, p < 0.001) the 283 localized urban scores with an R 2 of 0.51, and the relationship was even stronger (t=8.93, df=47, 284 p < 0.001, R 2 = 0.63) when the model was weighted by the standard error of the local-scale urban 285 scores' parameter estimates, to reduce distortion by species with small sample sizes. Even 286 without this correction, the relationship appears to be robust to the number of underlying samples 287 per species used to calculate the continental urban score. Indeed, of 100 different models, based 288 on sample sizes for continental-scale urban scores from 10 to 1000 there was little differentiation 289 in the underlying relationship (Fig. 2a), and the R 2 for these models leveled off after ~ 250 290 observations (Fig. 2b). We demonstrated a novel empirical relationship between continental-scale urbanness of birds in 294 Australia and local-scale urbanness among four small cities, relying on > 3 million citizen 295 science bird observations combined with intensive local-scale bird surveys, highlighting the 296 potential applications of broad-scale citizen science data. We found that a relatively small 297 number of citizen science observations (~250) are needed to provide reasonable estimates of 298 local-scale responses to urbanization. This approach highlights that continental-scale data may be 299 a sufficient proxy throughout regional cities to help guide urban planning and development -300 even when these cities lack the appropriate citizen science data. For example, urban planners in 301 developing cities can look at the continental ranking of species' urban tolerance and sufficiently 302 design cities that provide habitat and resources for those species most at risk (i.e., providing 303 artificial hollows for hollow-nesting birds or ensuring urban grasslands for at-risk granivorous 304 species). Concomitantly, urban planners can mitigate risks from the most harmful species (i.e., 305 despotic species which likely have the highest urban-tolerance scores). continuous measures of inter-specific variation, although we note that species can indeed be 319 clustered into those which respond to urbanization positively, negatively, and show mixed 320 responses (e.g., Fig. 1). The difference, however, is that these characterizations are informed, 321 incorporating inter-specific variation. This methodological approach of assessing species-specific urbanness of birds based on 359 continental citizen science data is in its infancy, and we highlight here some potential 360 opportunities for future research. First, and foremost, this approach may be applicable across 361 other taxa (e.g., butterflies, dragonflies, mammals), reliant mainly on spatial coordinates of a 362 large number of sightings -increasingly available via broad scale citizen science data (Chandler 363 et al. 2017). Second, although our analysis is focused on species-specific responses to 364 urbanization, we highlight that the broad-scale assignment of a species-specific response to its 365 environment may be repeated with other environmental factors (e.g., tree-cover, water-cover), 366 albeit these responses will be inter-correlated. This approach could use remotely-sensed 367 landcover products -other than urbanization -to assign species-specific responses. But 368 species' responses to other environmental factors should also be tested across spatial scales. 369 Third, although we focused on measuring inter-specific variation, this approach may be able to 370 be used to measure intra-specific variation, informing how local populations are adapting to 371 anthropogenic change (e.g., González-Oreja 2011). For example, some species did not conform 372 to the general results (e.g., New Holland Honeyeater, Spotted Pardalote, Galah) which is likely 373 explained by intra-specific variation in their continental population with some populations being 374 more 'urban' than other populations, which may not necessarily manifest in a specific location 375 (i.e., our local-scale study site). Fourth, we currently use large amounts of data to provide a 376 should also be adopted to regions where the fauna has differing migration strategies, thereby 381 assessing species-specific responses to urbanization intra-annually. disproportionately skewed towards urban areas (Kelling et al. 2015). Detection probability also 389 varies among species and between habitats (e.g., urban versus rural habitats), potentially limiting 390 the ability to draw inferences to poorly sampled species and habitats. For example, in our study, 391 we predominantly looked at common species, and our results may be only applicable to common 392 species, with more research necessary to understand how our results translate to uncommon and 393 rare species. This study was conducted in Australia -an area with relatively large amounts of 394 citizen science data -and our results may not be generalizable or applicable to other parts of the 395 world with less data (La Sorte and Somveille 2019) -and this should be tested in the future. 396 But with the global increase in such data (Chandler et al. 2017), we are hopeful that our approach 397 will be applicable to historically poorly sampled parts of the world (e.g., tropics, developing 398 countries). Given these biases, we do not suggest that systematic sampling should be replaced 399 with citizen science data, but rather, that they can complement one another to provide a more 400 generalized understanding in biodiversity research (Bayraktarov et al. 2019). Nevertheless, 401 methods such as the one we introduce here will likely be essential to track biodiversity responses 402 to urbanization into the Anthropocene.  Table S1. A table of the 94 species observed in the Blue Mountains and the total number of observations for each species. Also included is the number of continental observations, from eBird, used to assign continental-scale urban scores. Only species with > 10 local records were considered for analysis, and 2 were removed as outliers (Appendix S2).    Figure S3. Histogram of the parameter estimates from Generalized Linear Models fitted for each species, representing the local-scale response to urbanization. Species with a parameter estimate > 0 are responding positively to urbanization, while species with a parameter estimate < 0 are responding negatively to urbanization. Model results, including p-values, can be found in Table S2.