Abstract: A numerical forecast product explanation technology by the model output statistics (MOS) method was a good measure of element forecast. Wind direction was general explained by a scalar method, while it is not perfect. A vector method of wind direction forecast was set up by the MOS method. Wind direction models of u wind and v wind in the different stations for different time validity were established by methods of a correlation analysis and a regression analysis. The composite wind direction of u wind and v wind was predicted by the models. Three schemes were compared, including the vector method, scalar method and model output method. Wind direction in the 20 weather stations and at the four times was predicted based on daily observational fields of many physical elements from T639 model. The data from May 15 to September 15 during 2008-2012 were used to build the model, and those on June 1 to 30, 2012 were used to test the sample. The results show that the accuracy rates of wind direction for 12 hours, 24 hours, 36 hours and 48 hours is higher from the vector method than from the scalar method, and they are increased by 85.0%, 26.3%, 11.30% and 19.9% respectively. The accuracy rates for these hours forecasts are increased by 35.3%, 20.2%, 43.6% and 21.0% compared with the results from model output. In general, the vector method is the best among three methods. It is difficult to forecast for a certain station because unreasonable estimation for the model or frequent wind direction change.

Key words: font-family: "Times New Roman", mso-font-kerning: 1.0pt, mso-ansi-language: EN-US, mso-fareast-language: ZH-CN, mso-bidi-language: AR-SA,  Wind direction forecast')">mso-fareast-font-family: 宋体"> Wind direction forecast, Multiple regression, Vector prediction method