Generalized Linear Mixed Model and Lasso Regularization for Statistical Downscaling

Rainfall is one of the climatic elements in tropics which very influential agriculture, especially determining growing season. Thus, proper rainfall modeling needed to help determine best time start cultivating soil. can be done using Statistical Downscaling (SDS) method. SDS a statistical model field climatology analyze relationship between large-scale and small-scale climate data. This study uses response variables as data form explanatory General Circulation Model (GCM) output precipitation. However, application known cause several problems, including correlated not stationary variables, multi-dimensional multicollinearity, spatial correlation grids. Modeling with some these problems will violations assumptions independence multicollinearity. research aims Indramayu Regency, West Java Province combined regression Generalized linear mixed (GLMM) Least Absolute Selection Shrinkage Operator (LASSO) regulation (L1). GLMM was used deal problem Lasso Regulation (L1) multicollinearity or number that greater than variable. Several models were formed find for rainfall. GLMM-Lasso Normal spread compared Gamma (Gamma-GLMM). The results showed RMSE R-square smaller Gamma-GLMM models. it concluded downscaling solve previously mentioned constraints. Received February 10, 2021Revised March 29, 2021Accepted 2021