Least square support vector machine ensemble for daily rainfall forecasting based on linear and nonlinear regression

Abstract

Accurate forecasting of rainfall has been one of the most important issues in hydrological research. In this paper, a novel nonlinear regression ensemble model is proposed for rainfall forecasting. The model employs Least Square Support Vector Machine (LS-SVM) based on linear regression and nonlinear regression. Firstly, Projection Pursuit (PP) technology and Particle Swarm Optimization (PSO) algorithm are used to obtain the main factors of the rainfall, which optimize projection index from high dimensionality to a lower dimensional subspace. Secondly, using different linear regressions extract linear characteristics of the rainfall system, and using different Neural Network (NN) algorithms and different network architectures extract nonlinear characteristics of the rainfall system. Finally, LS-SVM regression is used for nonlinear ensemble model. This technique is implemented to forecast daily rainfall in Guangxi, China. Empirical results show that the prediction by using the LS-SVM ensemble model is generally better than those obtained using other models presented in this study in terms of the same evaluation measurements. The results suggest that our nonlinear ensemble model can be extended to meteorological applications in achieving greater forecasting accuracy and improving prediction quality. © 2010 Springer-Verlag Berlin Heidelberg.