An Ensemble Forecast Method of Rainstorm Based on mRMR and Random Forest algorithms

Abstract

This thesis presents an ensemble method of rainstorm based on maximum relevance minimum redundancy (mRMR) and random forest algorithms called mRMR-RFR. The proposed method is applied to the forecasting results of ensemble numbers from European Centre for Medium-Range Weather Forecasts (ECMWF). The method filtrates the 51 collected forecast members of ECMWF using the mRMR algorithm, and selects the members with maximum relevance and minimum redundancy to the forecasting objects. The selected members are regarded as the input factors of the random forest algorithm for forecast modeling. Experimental forecasting statistics show that, compared with the interpolation method of the collected forecasting members of ECMWF, mRMR-RFR can result in better forecast effect and use the numerical forecast products more effectively. The proposed method is thus suitable for forecasting.