Are road-kills representative of wildlife community obtained from atlas data?

Systematic road-kill surveys are useful to study the impact of roads on wildlife. However, they are time-and budget-consuming, so the use of non-systematic data in road ecology is currently gaining popularity (for instance, by environmental consultants). Some data sources such as atlases (i.e., compilations of species records from a given region), which can include non-systematic and citizen-science data, can entail several intrinsic biases, mostly due to uneven sampling effort and uneven species detectability. Here, we tested this prediction by verifying if data from the Spanish Atlas of Terrestrial Mammals mirror the road-kill patterns obtained from our own systematic roadkill surveys. We focused on the Mediterranean mesocarnivore guild due to its easy identification by citizens involved in atlas-data collection. We tested if the relative abundance of each species, their richness and diversity obtained from Atlas and our systematic surveys were related, using linear models, while controlling for human population and road density (potentially confounding effects). We further compared the patterns of species abundance obtained from both sources. Our results highlight that road-kill patterns do not mirror the Atlas patterns for the three metrics evaluated. This is probably due to survey biases in typical data from wildlife atlases. When analysing species individually, we found that some species are road-killed more (or less) than expected in relation to their abundance in atlas records. These results are probably due to species-specific ecological or behavioural traits such as species morphology or species behaviour when facing the road. We suggest that abundance from atlas data should not be used as a proxy for road-kill rates.PQ and RB were supported by Comunidad de Madrid (2018T1/AMB10374), FA by Fundacão da Ciência e Tecnologia (SFRH/BPD/115968/ 2016), and MD benefited from a Fundacão da Ciência e Tecnologia postdoctoral contract (CEECIND/03798/2017).Peer reviewe