Abstract
In this work, a regression model is proposed for predicting the summer rainfall around Taiwan area in which the global sea surface temperature (SST) are incorporated as the predictive variables. In constructing such a regression model, we first identify the effective spatial area according to the Portmanteau test statistics which quantify the dependence between the rainfall series and the SST series for each location. Then, the SST data within the identified area are summarized by the most important principal components to reduce the dimension of the predictive variables. Finally, the regression model is built by regressing the rainfall on these principal components of past SST data. The empirical results show that the regression model incorporating the information of SST performs better than the conventionally time series model in terms of smaller prediction mean square errors.