Bayesian Tweedie Compound Poisson Gamma (TCPG) modeling for statistical downscaling of rainfall in West Java, Indonesia

Global climate models (GCM) are effective in representing climate processes at the global scale; however, they often exhibit biases and limited accuracy at the local scale. This limitation is particularly critical in monsoon-dominated regions such as West Java, where statistical downscaling (SD) provides an appropriate approach. This research aims to predict monthly rainfall in West Java using the Bayesian Tweedie Compound Poisson Gamma (TCPG) model with combined scenarios of bias correction and dummy variables. Bias correction used empirical quantile mapping (EQM) with CHIRPS data. Monthly rainfall as the response variable was modelled using a Bayesian TCPG regression, with parameter estimation performed through Bayesian Markov chain Monte Carlo (MCMC) using the Metropolis Hastings algorithm. The best model scenario was achieved using dummy variables without bias correction, with CNRM-ESM2-1 identified as the most effective Decadal Climate Prediction Project (DCPP) model. These findings enhance rainfall prediction accuracy in tropical monsoon regions and support agricultural and water resource planning in West Java.