Comparison of ensemble models for drought prediction based on climate indexes

Current drought studies have already used the concept of equal ensemble streamflow prediction, but lack other kinds of ensemble prediction models for comparison of drought probabilistic prediction. Therefore, in this paper, an equal ensemble drought prediction (EEDP) model, a weighted ensemble drought prediction (WEDP) model, and a conditional ensemble drought prediction (CEDP) model were established and elaborately compared for drought prediction. The EEDP model directly uses the concept of ensemble streamflow prediction and assigns an equal weight to each ensemble member. The WEDP model assigns weights to ensemble members given the similarity of climate indexes between the historical and forecast year. The CEDP model considers the climate information as predictors and generates conditional ensembles of drought given the climate indexes. The verification of the proposed models was carried out using 26 meteorological stations in Jiangxi province (China), using the standard precipitation index to depict meteorological drought conditions in October, November, and December. The results show that compared to the EEDP model, the WEDP and CEDP models remarkably improved the accuracy and reduced the uncertainty of the drought prediction, indicating that a model considering climate information is much better than a purely statistical model. Meanwhile, the CEDP model outperformed the WEDP model in parameter estimation and accuracy. The prediction in southern Jiangxi province gets the best results compared with other regions. Our results provide a basis for drought planning and management in Jiangxi province.