Three statistical forecasting methods, i.e. ARIMA (Auto-Regressive Integrated Moving Average) model prediction, EMD (Empirical Mode Decomposition) decomposition prediction, and isolated prediction of frequency and amplitude based on Hilbert transformation, are designed and employed to make extra-seasonal prediction tests on the precipitation over Nanjing in 1998. Results show that the ARIMA model exhibits severe system errors and is hard to reproduce the abrupt variation of precipitation. Although the EMD decomposition prediction makes an obvious improvement in the evolution trend of precipitation, it still fails in the depiction of precipitation catastrophes in the summer due to its incapability of predicting high frequency modes. The isolated prediction method improves the prediction of high frequency modes since it can separately predict the frequency and amplitude of each mode and their interactions are avoided. Thereby the isolated prediction method gives a pretty good final prediction with the highest trend correlation and the smallest deviation. The two precipitation catastrophes in the summer of 1998 are realistically predicted. Additionally, a further verification of the precipitation prediction for 2003 also indicates that the isolated prediction method performs best among the three methods proposed in this study. The above results suggest that the isolated prediction method may provide a new idea for the technological improvement on extra-seasonal short-term climate prediction.