A Bayesian Geostatistical Framework for Malaria Risk Prediction Using Multi-Source Remote Sensing in Nigeria

Spatial prediction of malaria risk across Nigeria's diverse ecological zones requires modelling frameworks that accommodate the complex non-linear relationships between environmental drivers and disease transmission while rigorously quantifying spatial uncertainty. This study develops a Bayesian geostatistical framework for malaria risk prediction using multi-source remote sensing and field survey data across all six geopolitical zones. Remote sensing covariates including land surface temperature, precipitation, NDVI, normalised difference water index, and land use are extracted from Landsat, MODIS, Sentinel-2, and CHIRPS products for 2020 to 2023. A spatially structured Bayesian model implementing integrated nested Laplace approximation is fitted to malaria parasite rate survey data from 847 communities. The model explicitly estimates spatially correlated random effects and produces probabilistic risk maps with credible interval uncertainty bounds. Model cross-validation using held-out geographic clusters demonstrates superior predictive performance over conventional regression models. Results identify high-risk zones straddling the boundary of the Guinea and Sudan savannas and in the riverine communities of South South Nigeria. Keywords: Bayesian geostatistics, malaria risk prediction, remote sensing, Nigeria, spatial uncertainty