Pooling information across sites and years to estimate bat fatalities at wind farms

Abstract Precise estimation of bat mortality at wind farms is essential for conservation planning. Carcass detection probabilities, however, are often low and data sparse, leading to large uncertainties in fatality estimates. Precision can be improved by pooling information across multiple sites and years. Using post‐construction monitoring data from 20 wind farms in Iowa, USA, we evaluated complete pooling (CP), no pooling (NP), and partial pooling (PP) approaches for estimating searcher efficiency, carcass persistence probability, and the fatality estimates obtained by integrating these detection‐probability components. For searcher efficiency and persistence probability, PP produced estimates that balanced the extremes of CP and NP, yielding narrower credible intervals than NP, especially at data‐limited sites, while preserving meaningful site‐level differences. Fatality estimates for 3 bat species, differing in carcass abundance, were generally similar in the PP and NP models, but both were often different from CP. Partial pooling showed gains in precision, especially for abundant species and years with less monitoring effort. A simulation study of scenarios with different amounts of heterogeneity among sites showed that PP provided more accurate predictions than CP and NP in intermediate and high heterogeneity settings. Our findings highlight the value of PP as a flexible strategy for improving detection probability estimation under ecological uncertainty and support its broader application in wildlife impact assessment.