When Good Fit Goes Bad: Identifying and Minimising Overfitting in Ecological Niche Models

ABSTRACT Aim To synthesise current understanding of overfitting as a pervasive and often underdiagnosed problem in correlative ecological niche models (ENMs), and to assess its consequences for model interpretation, generalisation and transferability under contemporary data and methodological practices. Location Worldwide. Methods We reviewed the ENM literature, synthesising theoretical and empirical studies that address the detection and treatment of overfitting. We examined diagnostic approaches, including spatially and environmentally structured cross‐validation, and evaluated modelling practices related to sampling bias, predictor choice, study area definition, model complexity and regularisation. Results Common warning signs of overfitting include large discrepancies between training and testing performance, fragmented or ecologically implausible prediction maps and unrealistic response curves. Spatially and environmentally structured cross‐validation approaches are effective for assessing data independence and diagnosing inflated performance estimates. Although multiple mitigation strategies can reduce overfitting, the persistent lack of generalisation often reflects limitations related to data quality, spatial and environmental scale, and violated ecological assumptions rather than modelling choices alone. Main Conclusions Predictive accuracy alone is insufficient for evaluating correlative ENMs. Ecological plausibility, interpretability and consistency with known species distributions are equally important for assessing model reliability. We advocate for iterative model evaluation, explicit consideration of ecological assumptions, and transparent reporting as best practices to improve the trustworthiness and transferability of ENMs in ecological research and conservation planning.