Application of Statistical Downscaling with Principal Component Regression for Local Rainfall Forecasting in Jember Regency

Abstract

Global climate change causes various changes and extreme fluctuations in weather circumstances, including extreme changes in rainfall. An accurate rainfall forecasting was indeed needed in various agricultural activities. The statistical downscaling (SD) was developed to model the global climate circumstance data from the satellite, called the General Circulation Model (GCM). Combine with data on the earth from the weather station; the GCM predict the future local weather. The functional relationship in the SD was modeling the GCM output data as the predictors and the local-scale rainfall data as the response. The GCM’s ability to display predictive data for decades to come was a technological leap in forecasting the rainfall to study long term on weather/climate change. Statistically, this modeling requires the twos below: (1) a dimensional reduction in GCM data and (2) accurate predictive models on the functional relationship. In this study, rainfall forecasting was conducted in Jember Regency using Principal Component Analysis (PCA) for dimensional reduction and a predictive model of Principal Component Regression (PCR). The accuracy was measured in each cluster in the 8×8, and 10×10 domains with the RMSE statistic was around 80.41-101.35.