Presence-only species distribution models are sensitive to sample prevalence: Evaluating models using spatial prediction stability and accuracy metrics.

• Spatial predictions from SDMs are sensitive to sample prevalence. • Accuracy metrics like AUC are unreliable indicators of spatial prediction accuracy. • SDMs should also be assessed on stability and consistency of geographic prediction. Species distribution modelling (SDM) is an important tool for ecologists, but different algorithms and different sampling strategies produce different results. Using virtual species with differing characteristics, this study investigated the effect of sampling strategy choices on SDM predictions across multiple algorithms and species, including the impacts of different sample size and prevalence choices, and the effects of validating models using presence and background data as opposed to true absences. We also assessed the consistency of predictions between algorithms, and investigated the effectiveness of using stability assessment of spatial predictions in geographic space to evaluate SDM predictions. Maxent performed most consistently under all scenarios both in regards to performance metrics and spatial prediction stability, and should be considered for most scenarios either on its own or as part of a model ensemble, in particular when true absences are not available. A key recommendation of this study is the use of metrics to assess agreement between replicate predictions as a measure of spatial stability, rather than relying solely on performance metrics such as area under the curve (AUC). Image, graphical abstract [ABSTRACT FROM AUTHOR]