Climate variables are not the dominant predictor of Arctic shorebird distributions

throughout their annual cycle. Here, we saw that within closely related shorebird taxa in the same Arctic ecosystem, the total effect of climate explained anywhere from 10% to 80% of the explained deviance in shorebird occupancy. In the debate about whether climate-only models are adequate for describing habitat suit- ability, one important factor that leads different papers to come to different conclusions is the spatial scale at which studies are conducted [44]. Even Pearson and Dawson [10], advocates of climate-only models, recommend that these analyses be undertaken with careful consider- ation: “. . .bioclimate envelope models can provide a useful starting point when applied to suit- able species and at appropriate scales.” However, it seems that whether or not a climate-only model is appropriate for the given species and scale is often assumed rather than tested. For example, Wauchope et al. [24] write “Data were interpolated to a spatial grain size of 10 x 10 km to reflect the approximate resolution of most of the distributional records and a scale at which climate, rather than microhabitat factors, is more likely to be limiting [45]”. Identifying which predictors are relevant at a particular spatial scale requires consideration of both the grain and the extent [44]. Spatial grain has been demonstrated to influence the rela- tive importance of climate predictors. For example, Luoto et al. [47] found that climate-only models of bird distributions in Finland were improved by including land-cover at 10km and

20km resolutions, but not at 40km and 80km resolutions. The spatial extent of a study also plays a significant role in defining how much variability we find in each of the predictor vari- ables. The extent of our study covers only a part of the breeding range of some species, and just beyond the range of others. Studies with much larger extents, covering the full range of climate conditions across the continent would have a stronger signal for climate, identifying a wider range of conditions species do not use. Habitat selection is a hierarchical process [46]. Identify- ing the appropriate spatial scale for the type of habitat selection being studied will help identify when climate-only models are sufficient, and when other predictors such as land cover are important. Optimizing the scale of effect for land cover and other predictors that can be measured at a range of spatial scales is an important step that is often neglected [36]. Here, we found that Arctic-breeding shorebirds were responding to land cover at a radius of up to 10km, but that the scale of effect differed across species and land cover categories. A single-scale study of ver- tebrates in Florida [47], contrary to our results, found that including additional predictors had a relatively minor effect on the accuracy of climate-only habitat suitability models. They sug- gested that the 4km grain of their study may have been too coarse to resolve the land cover associations for their study species, following the hypothesis that biotic interactions are most

important at smaller spatial scales. Another important difference between our study and that of Bucklin et al. [48] may be that the spatial extent of our study area was approximately 20x larger; therefore, there may be more available variation in land cover across our sites, increas- ing its importance. Single-scale studies may therefore miss important predictors, because they are not measuring the predictors at the scale at which species are responding to the predictor [44]. Other reasons that climate-only models may be preferred are purely practical. Although important, additional predictors are often neglected because they are unavailable, or harder to obtain at the extent and resolution needed [20, 49]. Particularly for future predictions of habi- tat suitability, future climate projections are widely available, while other future projections such as land cover are much more limited in their availability, their spatial resolution, their accuracy, and the level of detail that they provide in terms of the number of classification cate- gories [50, 51]. We conclude that environmental predictors beyond climate are important for improving the accuracy of habitat suitability models explaining species distributions. We found that a PLOS ONE Climate variables are not the dominant predictor of Arctic shorebird distributions PLOS ONE | https://doi.org/10.1371/journal.pone.0285115 May 17, 2023 10 / 14