Support Vector Regression in Statistical Downscaling for Rainfall Forecasting

Abstract

In recent atmospheric global climate change, rainfall forecasting has an important role especially in the archipelagic-agricultural country like Indonesia. We use statistical downscaling to get rainfall forecasting in order to support plantation crops in Kabupaten Jember. Therefore, long-term forecasting in this case is really needed. Statistical downscaling methods seek to draw empirical relationships that transform large-scale feature of global atmospheric condition called General Circulation Model (GCM) to a local scale rainfall variable. In this study we use Support Vector Regression (SVR) to construct the empirical relationship in Statistical Downscaling approach. The three grid size of GCM (8x8,10x10,12x2) were used to develop models and the best identified model was used for simulations of future rainfall forecasting. The result show that all the models in each grid sizes are able to simulate rainfall, however, SVR model in 8x8 grid size slightly better than other grid sizes and we get the SVR-SD’s cross-validation accuracy of 61.77 of Root Means Square Error (RMSE) and 78% R-square. Then we obtain forecasting value or rainfall for 2019-2020 period.