The Ensemble Of Arima And Gstar Models In Forecasting Rainfall Using Kalman Filter

Abstract

Several forecasting rainfalls with various models have been carried out in the same area. The results in each forecasting may be different from each other and to choose the best one is difficult. In this study we will discuss the Super-Ensemble Kalman Filter method which combines two or more forecasting results using the Kalman Filter method to get maximum results. The rainfall data used in this study has been divided into 4 clusters using K-Means. The ARIMA and GSTAR models from the 4 clusters were selected as the best model by looking at the smallest RMSE value from each model then the best of ARIMA and GSTAR models were ensembled using Kalman Filter. Based on the results obtained, the Super-Ensemble Kalman Filter method provides maximum results in forecasting rainfall data.