UAV-derived absence data improve species distribution model performance for the plateau pika

Reliable absence data remain a major limitation in the development and application of species distribution models (SDMs). Both field-sampled absences and algorithmically generated pseudo-absences are fundamental to model parameterization, yet efficient approaches for obtaining robust absence information and how absence data shape SDM outcomes are still lacking. Here, using Biodiversity Modelling (BIOMOD2), we systematically assessed SDM performance across multiple pseudo-absence generation strategies and sample sizes, leveraging a presence/absence dataset ( n = 2261; 1743 presences and 518 absences) for plateau pika ( Ochotona curzoniae ) derived from unmanned aerial vehicle (UAV) surveys on the Qinghai–Tibetan Plateau. UAV-derived absence data consistently produced the highest model accuracy, particularly for Random Forest models (Kappa = 0.566; TSS = 0.611; AUC = 0.863). Among pseudo-absence strategies, the surface range envelope (SRE) strategy performed the best, whereas sample size showed relatively minor influence. Importantly, the spatial configuration of absence samples altered the weighting of environmental predictors, resulting in pronounced differences in model performance. Our findings demonstrate that absence data are pivotal to SDM accuracy and that UAV technology fundamentally advances the acquisition of presence/absence data by enabling efficient, spatially extensive, and reliable sampling. These methodological innovations are essential for improving biodiversity forecasting and informing management strategies under accelerating climate change and intensifying anthropogenic pressures. • Absence data from unmanned aerial vehicles improve model accuracy for plateau pika. • Spatially stratified pseudo-absence outperforms random absence sampling. • Absence sample design affects predictor importance and model performance. • Aerial surveys enable large-scale, field-validated species distribution modelling. • Improved distribution models support biodiversity forecasting and conservation.