Modelagem de áreas críticas de atropelamento de cachorro-domato (Cerdocyon thous) no Brasil

Abstract

Road infrastructure are essential for the social and economic development of the country and are constantly being expanded, since they allow connecting remote areas and transporting cargo and passengers throughout the country. The implementation of these projects has an impact on local biodiversity, among which is the wildlife roadkill. This study aims to evaluate the effectiveness of spatial modeling in defining critical areas of roadkill of Cerdocyon thous in Brazil and compare the environmental variables that condition the deaths in different Brazilian biomes. We used 3835 occurrence data of the species, collected between the years 2000 and 2019. As most of these records come from road kills, the potential distribution models to be produced present a bias for areas of occurrence of these events, in this way we adopted the name "Potential Areas of Roadkill -AREPAs". The AREPAs were defined through models produced by the software Maximun Entropy algorithm (Maxent) and used environmental variables (n=21) available in different databases, such as IBGE, IPEF, INPE and WorldClim. Six potential distribution models were built, one for each Brazilian terrestrial biome. The models were validated in three ways: (i) internally by Maxent; (ii) externally, from a new dataset and; (iii) through Hotspot analysis in Siriema software. As expected, the variables associated with the probability of roadkill were distinct for the biomes, which may be related to the generalist and opportunistic habits of the species. The external validation process showed a satisfactory result (hit rate>80%) for all models, indicating that, statistically, even for the biomes where lower AUC values were obtained (Caatinga=0.7005 and Atlantic Forest=0.72), a correct prediction of the critical areas of roadkill for the species was made. In general, the hotspots were located in areas of intermediate probability of roadkill, suggesting that these areas are a priority in terms of conservation. The results obtained in this study, since this is a highly generalist species, demonstrate the potential of using modeling to define critical areas for road kills, favoring the reduction of costs for decision making.