Spatial prediction of environmental suitability for dengue fever transmission based on geographical environmental similarity with spatial connectivity

Spatial prediction of environmental suitability for vector-borne disease transmission is crucial for public health, with spatial connectivity being a critical factor. However, existing models often face a trade-off between model interpretability and the accurate representation of spatial connectivity. Simpler, interpretable models tend to oversimplify connectivity, while more advanced models can lack interpretability and often rely on difficult-to-obtain fine-grained mobility data. To address this challenge, this paper proposes an integrated framework (GES-SC) for the spatial prediction of environmental suitability for vector-borne disease transmission, which combines Geographical Environmental Similarity (GES) with quantified Spatial Connectivity (SC). The method operates on the principle that environmentally similar and highly spatially connected locations have similar transmission potential. It quantifies connectivity using available road network data and distance decay, providing an effective alternative to direct mobility data. A case study in Guangzhou, China, demonstrates that integrating this quantified spatial connectivity enhances prediction accuracy. Compared to Geographically Weighted Regression, Bayesian Conditional Autoregressive, XGBoost, and a Simple Neural Network, the GES-SC method performed better in cross-validation, achieving lower RMSE and MAE, and a higher R². The proposed framework effectively addresses the accuracy-interpretability trade-off, improving spatial epidemiological prediction accuracy and providing an uncertainty measure for public health decisions.