A Bayesian Approach to Spatio-Temporal Extreme Rainfall Modeling: Insights from West Java

The increasing phenomenon of extreme rainfall due to global climate change poses a serious challenge to hydrometeorological disaster risk management. Spatio-temporal modeling has proven to be a practical approach to understanding the distribution and intensity of extreme rainfall; however, its implementation still faces methodological obstacles, primarily due to the complexity of spatial and temporal data structures and limitations in the flexibility of grid-based spatial zoning. This study proposes the development of a Bayesian-based spatio-temporal model specifically designed to estimate extreme rainfall in West Java Province, a region with complex topographical conditions and high vulnerability to disasters. Three types of models were built, namely linear models with time trends, additive models without interaction, and additive models with spatial-temporal interaction. Spatial effects were modeled through the Conditional Autoregressive (CAR) approach, while parameter estimation was performed using the Integrated Nested Laplace Approximation (INLA) method. Evaluation results using RMSEP and DIC metrics show that the additive model with spatial-temporal interaction has the most optimal performance in predicting extreme rainfall. Longitude and latitude factors are identified as the dominant determining variables, which reinforces the critical role of geographical aspects in spatial estimation. This research not only expands the application scope of Bayesian methods in climate modeling but also provides a strong scientific basis for the development of early warning systems and data-driven disaster mitigation strategies. This approach can potentially be adapted for use in other regions and combined with Internet of Things (IoT) technologies to produce more precise and adaptive real-time extreme rainfall predictions.