A Neural Network Ensemble Prediction Model Based on MGF and PLS for Drought and Waterlogging Disasters in Short-Range Climate

Taking the mean precipitation from 16 stations spread around the south China during the pre-flood season as the prediction object treated by Empirical Orthogonal Function (EOF) method, previous physical predictors and factors that reflected the significant period of predictands by means of the Mean Generating Functions (MGF) technique, were extracted useful information for prediction by using Partial Least-square regression (PLS) approach, thereby establishing a Genetic Neural Network (GNN) ensemble prediction (GNNEP) model. In order to evaluate the rainfall forecast skill over the studied region, predictions with a stepwise regression method were compared to those of GNN. The results show that GNN forecasts are superior to the ones obtained by the traditional stepwise regression method thus revealing a great potential for an operational suite.