Projection pursuit regression and principal component regression on statistical downscaling using artificial neural network for rainfall prediction in Jember

Abstract

Rainfall forecasting is essential for Indonesia, which is an agricultural country. Forecasting to see the rainfall needed to anticipate the danger of drought that will harm farmers. However, due to the complexity of topography and the interactions between the oceans, land, and atmosphere in Indonesia, it is difficult to predict rainfall. Therefore, Statistical Downscaling (SD) is needed to provide accurate rainfall predictions by considering the information about global atmospheric circulation obtained from the General Circulation Model (GCM). Statistics Downscaling (SD) modeling is a basic regression model based on the functional relationship between local scales, which is the response variable with the Global Circulation Model (GCM) global scale as a predictor variable. The Statistics Downscaling (SD) method used is Principal Component Regression (PCR) and Projection Pursuit Regression (PPR). The prediction of both methods was conducted by an Artificial Neural Network (ANN). The results showed that the prediction of rainfall in Jember using the PPR + ANN method (with the RMSE value of 79.58723) had better accuracy than the PPR, PCR, and PCR + ANN methods, which had RMSE values of 103.7539, 112.337 and 83.62029, respectively.