When accessibility distorts niche estimates: A virtual species framework across species prevalence and sample size gradients

Abstract

Species distribution models are widely used in ecology and conservation biology to predict species' habitat suitability, but their performance is often compromised by biases in occurrence data. One of the most pervasive sources of bias arises from geographically uneven sampling, particularly the overrepresentation of areas close to roads. In this work, we outline a reproducible workflow to assess whether such geographic biases affect the representation of species' ecological niches in environmental space. We simulated 250 virtual species within the Abruzzo region (Italy) and evaluated the effects of sampling bias by systematically varying sample size and species prevalence (i.e., geographic range size). For each species, we compared random sampling from within the species' range with sampling constrained to areas near roads.

Rather than revealing a uniform effect of road-biased sampling, our results show that its consequences depend strongly on species prevalence, sample size, and the spatial correspondence between species' ranges and the accessible areas near the roads. The extent to which niche estimates are distorted therefore varies among species, implying that the effects of road-based sampling are species-dependent: sampling near roads can either misrepresent or adequately capture environmental conditions depending on the species considered. These findings highlight the need to be aware of how sampling patterns interact with species' spatial distributions when interpreting model outputs.