A0247
Title: Spatial disaggregation using a Bayesian low-rank geoadditive model
Authors: Christel Faes - Hasselt University (Belgium) [presenting]
Abstract: A novel Bayesian spatial disaggregation model is presented for count data that integrates smooth covariate effects and spatial dependencies within a unified framework. The model leverages penalized splines to flexibly capture nonlinear relationships with covariates. For modeling spatial correlation, a spline-based low-rank kriging approximation is adopted, which improves computational tractability. Additionally, Laplace approximation is used for posterior inference, offering substantial efficiency gains over traditional Markov chain Monte Carlo (MCMC) techniques. Two estimation strategies are explored: One based on the exact likelihood and another using a spatially discrete approximation to further enhance computational performance. Through simulation studies, it is shown that both approaches yield accurate estimates, with the approximate method delivering notable computational savings. The model is applied to disaggregate disease incidence data, generating high-resolution risk maps. Overall, the approach provides a flexible, scalable, and practical framework for spatial disaggregation in geostatistical applications involving count data.
