This paper introduces a modeling method for analyzing and predicting non-stationary time series. The annual rainfall data are used as one of the most important observation factors that can reflect the dynamic change of drought. Thus, this paper gets the annual precipitation data of the Leizhou Peninsula from 1957 to 1996, and adopts a method of time series analyzing, which is building an auto-regression moving-average (ARMA) model after the trend terms and periodic terms are extracted and examined. And then the posterior results are forecasted by the model. The result shows that the annual precipitation has indistinctive trend and obvious periodical change. The model of predicting precipitation on Leizhou Peninsula has high degree of accuracy, the data predicted can be used in the dynamic drought analysis of Leizhou Peninsula.