Abstract
Accurately estimating wind turbines' direct impact on biodiversity remains challenging, as many carcasses are scavenged or displaced before detection. This bias is often assessed by placing carcasses around turbines, then integrated into fatality estimators. Given the limited number of carcasses used in studies, identifying key factors affecting persistence is essential to select relevant variables to account for. The development of robust predictive models could ultimately replace this approach. To this end, we aggregated over 10,000 carcass persistence records from France and used parametric survival models to evaluate the effects of habitat, weather, wind facilities, and carcass characteristics. Our results showed that the exponential distribution poorly fits carcass persistence over time, despite being assumed by most fatality estimators. Median persistence time was only 1.91 days. Survival models revealed significant effects of habitat and wind facilities variables: persistence increased with hedgerows length and urban area extent, but decreased with facilities age and the extent of forests, aquatic, and open natural areas. Birds persisted significantly longer than rodents. Meteorological variables had no effect, but carcasses persisted significantly longer in summer, followed by spring, and shortest in autumn. Despite these significant effects, the low explanatory power of our model suggests that carcass detection probability and removal by scavengers is poorly predicted by these variables alone. These results highlight that modeling alone cannot reliably predict persistence and cannot replace field-based trials. Instead, local persistence estimates should be integrated into robust fatality estimators allowing the appropriate distribution function to be selected.