Statistical Downscaling to Predict Monthly Rainfall Using Generalized Linear Model with Gamma Distribution

Statistical Downscaling (SDS) models might involve ill-conditioned covariates (large dimension and high correlation/multicollinear). This problem could be solved by a variable selection technique using L1 regularization/LASSO or a dimension reduction approach using principal component analysis (PCA). In this paper, both methods were applied to generalized linear modeling with gamma distribution and compared to predict rainfall models at 11 rain posts in Indramayu. More over, generalized linear model with gamma distribution was used to obtain non-negative rainfall prediction and compared with principal component regression (PCR). Two types of ill-conditioned data with different characteristics (CMIP5 and GPCP version 2.2) were used as covariates in SDS modeling. The results show that three methods (PCR, Gamma-PC, and Gamma-L1) did not demonstrate significant difference in term of Root Mean Square Error (RMSE) after addition of dummy variables (month) in the models. However, a generalized linear modeling with gamma distribution could be considered as the best methods since it provided non-negative rainfall predictions