A probabilistic habitat–suitability overlap framework (HMI) reveals spatial bias in MaxEnt models of West African forest butterflies

Species distribution models (SDMs) are widely used to infer potential species ranges, yet their ecological reliability in rapidly transformed landscapes remains poorly understood. We applied Maximum Entropy (MaxEnt) modelling to estimate the potential distribution of twelve forest-specialist butterfly species of high commercial value across West Africa, with a focus on Côte d’Ivoire, Ghana and Liberia, based on bioclimatic covariates. Models showed good predicted performance (mean jackknife AUC = 0.854, significantly better than random; null model test p < 0.01), driven by both precipitation-related variables (i.e. precipitation seasonality; 57.4% of explained variance in model predictions) and temperature-related variables (annual mean temperature, 33%). However, spatial predictions showed limited concordance with contemporary forest cover: 45.23% in Liberia but only 4.69% in Cote d’Ivoire and 6.42% in Ghana. Key refugia such as Taï Forest (Cote d’Ivoire) were partially underpredicted. To evaluate whether the observed agreement between predicted habitat suitability and forest cover reflects genuine ecological associations or is instead driven by spatial bias, we developed a probabilistic framework in which the expected overlap is defined as the product of forest availability and suitability probability. Building on this null expectation, we derived a Habitat Matching Index (HMI), expressed as the ratio between observed and expected overlap, allowing us to quantify deviations from random spatial association and to distinguish ecological signal from patterns generated by habitat availability alone. HMI values differed markedly among countries (Côte d’Ivoire: 0.319; Ghana: 0.570; Liberia: 1.407), demonstrating that apparent model performance is strongly influenced by forest extent rather than ecological fidelity alone. These findings reveal a fundamental limitation of climate-based SDMs: high statistical accuracy does not necessarily imply ecological realism. Integrating habitat availability into model evaluation provides a critical correction, improving inference in heterogeneous and rapidly changing tropical landscapes. We advocate combining climatic, land-cover, and species-specific data to enhance SDM reliability in conservation biogeography.