Abstract
This study analyses 18 years of weekly reported dengue cases (January 2002-December 2020; 988 weeks) from Costa Rica's Central Valley to examine seasonal and multi-year patterns. To model the spatio-temporal dynamics of dengue, we employ three statistical approaches for case counts: the spatial hurdle integer-valued generalized autoregressive conditional heteroskedasticity (INGARCH) model, the spatial zero-inflated generalized Poisson (ZIGP)-INGARCH model, and the endemic-epidemic (EE) model. Covariates include rainfall and maximum temperature or alternatively seasonal Fourier terms to represent annual seasonality. Using a Bayesian framework, we fit the spatial INGARCH-family models to weekly dengue cases. The EE model and the ZIGP-INGARCH model, both with Fourier seasonal terms, show the best predictive accuracy and provide estimates of seasonal intensity and peak timing relevant for dengue surveillance. Incorporating annual seasonality improves modelling of multivariate weekly dengue cases in Costa Rica's Central Valley, underscoring the importance of cyclical patterns for strengthening early warning systems and guiding targeted vector control.