A Novel Bayesian Additive Regression Trees Ensemble Model Based on Linear Regression and Nonlinear Regression for Torrential Rain Forecasting

Abstract

In order to improve the accuracy of precipitation forecasting with the linear regression of traditional statistical model and the nonlinear regression of Neural Network (NN) model, especially in torrential rain, a novel Bayesian Additive Regression Trees (BART) ensemble model is proposed in this paper. Firstly, three different linear regression model are used to extract the linear characteristic of rainfall system with the Partial Squares Least Regression, the Quantile Regression and the M-regression. Secondly, three different NNs model are used to extract the nonlinear characteristics of rainfall system with the General Regression Neural Network (GR--NN), the Radial Basis Function Neural Network (RBF--NN) and the Levenberg-Marquardt Algorithm Neural Network (LMA--NN). Finally, the BART is used for ensemble model based on linear and nonlinear regression. For illustration, a summer daily rainfall example is utilized to show the feasibility of the BART ensemble model in improving the accuracy of torrential rainfall with linear regression and nonlinear regression model. Empirical results obtained reveal that the torrential rainfall prediction by using the BART ensemble model is generally better than those obtained using other models presented in this paper in terms of the same evaluation measurements. Our findings reveal that the BRAT ensemble model proposed here can be used as an alternative forecasting tool for a Severe Weather application in achieving greater forecasting accuracy and improving prediction quality further.