The Frankenstein species distribution modelling: A composite Bayesian framework accounting for spatio-temporal heterogeneity in ecologically diverse regions

Species Distribution Models (SDMs) are used to explore and predict species-environment relationships across spatial and temporal scales. However, traditional SDMs often assume a single latent spatial structure throughout the study domain, limiting their ability to capture the complex environmental heterogeneity that governs ecological systems. In this study, we propose a novel composite modelling framework, the Frankenstein SDM, which integrates multiple spatio-temporal configurations within a single model. Built upon the INLA-SPDE Bayesian hierarchical approach, the Frankenstein SDM enables region-specific spatio-temporal structures to represent varying ecological processes across heterogeneous space. We evaluated our framework with two simulations: one with two sub-regions having distinct spatio-temporal dynamics and region-specific covariate effects, and another with four sub-regions with diverse spatio-temporal dependencies representing complex scenarios. We evaluated the model using a simulated dataset representing two regions with intentionally distinct spatio-temporal dynamics. Model performance was assessed via the Watanabe-Akaike Information Criterion, and the Root Mean Square Error derived from cross-validation. The Frankenstein SDM outperformed all standard configurations that rely on a single latent spatial structure across heterogeneous data. Real-world data were applied from fishery-independent surveys in the Strait of Sicily, modelling the standardized density of young-of-the-year European hake (Merluccius merluccius). The Frankenstein SDM improved model performance, capturing the contrasting spatio-temporal dynamics and region-specific covariate influences between two ecologically distinct sectors. Our results suggest that composite spatio-temporal models like Frankenstein SDM are essential when ecological processes vary across space, providing a more realistic and management-relevant tool. Our findings highlight that explicitly incorporating environmental heterogeneity into SDMs enhances predictive accuracy and ecological interpretability and advocates a shift from traditional “one-size-fits-all” SDMs toward modular spatial modelling frameworks.