Inman et al. (2021) showed that sample bias, a well known problem in presence-background SDM (species distribution modeling) not only affects accuracy and spatial prediction, but also affects variable selection and fitted environmental response curves. Using virtual species we found that the FactorBiasOut bias correction method showed the greatest improvement in recreating known distributions, although it did no better at correctly identifying environmental covariates or recreating species–environment relationships, than Geographic-Filter or Environmental-Filter methods. Check it out!
Figure: A) Rare species (low prevalence) and B) specialists (narrow niche) showed greatest improvement in prediction accuracy (increase in ESP, Expected fractions of Shared Presences) with FactorBiasOut