Statistical Downscaling Modeling with Time Lag Components for Forecasting Rainfall in Wet and Dry Seasons

Abstract

Climate change in Indonesia often poses a serious threat to the agricultural sector. The impacts can include reduced agricultural productivity. In this context, rainfall variables are frequently used in research related to the impacts of climate change. In this study, precipitation data from the global circulation model (GCM) outputs are used as predictor variables and rainfall data from the Indramayu station are used as response variables in statistical downscaling modeling. The cross-correlation function between these variables plays an important role in statistical downscaling modeling. The cross-correlation function can enhance the correlation between predictor variables and response variables. Therefore, this research aims to compare the rainfall prediction results using initial GCM data (GCM) and GCM data with lag components (lagged GCM) determined based on the cross-correlation function. The methods used in statistical downscaling modeling are partial least squares regression (PLSR) and principal component regression (PCR). The modeling results using data from the period 1993-2020 show that the PLSR model on lagged GCM data is the best compared to other models (PLSR on GCM data, PCR on GCM data, and PCR on lagged GCM data). This model produces the highest coefficient of determination and the smallest RMSE value. Furthermore, the PLSR model on lagged GCM data can predict the 2008 rainfall data, following the actual rainfall pattern with the smallest RMSEP value. In general, modeling using lagged GCM data provides better rainfall prediction results compared to GCM data