Abstract:  Through the interpretation of MM5 numerical prediction products, the MOS statistical method was applied to forecast short time meteorological elements. A forecast model of precipitation and temperature with 4 hours interval was established in terms of MM5 numerical prediction products, the observational data from automatic weather stations and the radar data. Precipitation was a discontinuous variable, but it could be translated into a continuous variable by establishing a possibility function of precipitation. Thus, quantitative forecasting could be realized. 4140 samples were tested in this paper. The results show that MOS method is better than the MM5 method, especially for the cases when errors with the MM5 method are higher. According to the test of precipitation prediction, TS score is 65 % and the accuracy reaches 91%. The average error is less than 8mm for heavy rain samples (three hours accumulated rainfall > 5 mm), while it is 1 mm or less for weak precipitation samples (three hours accumulated rainfall < 1 mm). The MOS method has the higher accuracy with 98% for the prediction the non-rain day, and it is better than the MM5 prediction. However, the convective precipitation is still hard to forecast. For the temperature model, the error is less than 1.0 ℃ from 20:00 to 08:00, while it is 1.5 ℃ from 11:00 to 17:00.  The error can be controlled about 1.0 ℃ after season corrections.

Key words: MOS statistical method, Short-term Elements forecasting, Possibility function of precipitation, TS score, MOS statistical method, Forecasting of short term meteorological elements, Possibility function of precipitation, TS score