Preservation biases in the fossil record distort species ecological niche and distribution models

Ecological niche models (ENMs) increasingly leverage the fossil record to understand species' environmental associations and predict their geographic distributions. However, fossils do not occur uniformly through time and space, which can compromise the robustness of ENMs and thus affect ecological conclusions. Here, we assessed how preservation biases in the fossil record impact our ability to reconstruct ecological niches and distributions of North American small mammals during the late Quaternary. First, using small mammal fossil occurrences and associated depositional environment data, we quantified preservation potential (the likelihood that a given environment supports fossil preservation) and the preservation niche (environmental correlates of preservation) for three late Quaternary time periods (the Last Glacial Maximum, the deglacial period, and the Holocene). Second, we imposed the calculated preservation potential on simulated distributions of six virtual species to evaluate its impact on reconstructing species niches and geographic distributions through time. We found that preservation potential was highest in the Holocene and lowest during the deglacial period, with the differences driven by variations in climate and the prevalence of Holocene archaeological sites. In all intervals, warm, wet, and highly seasonal environments exhibited low preservation potential. These spatial and temporal differences in preservation potential significantly influenced niche reconstructions and geographic predictions, particularly impeding model quality when species niches extended beyond the preservation niche. We warn that such distortions can lead to erroneous ecological inferences, including inaccurate predictions of species responses to environmental changes and mischaracterizations of community assembly processes. We propose that our approach to modelling preservation potential can be applied across different regions, time periods, and taxonomic groups to help correct distortions caused by sampling biases through weighted background point selection that reflects these processes. Ultimately, this framework enhances the ability to disentangle true ecological patterns from preservation artifacts, improving the reliability of fossil‐based ecological and evolutionary inferences.