Abstract: The development of numerical precipitation forecasting model in high resolution is an ideal way to carry out the refined precipitation forecast service.However,the bias evaluation in model localization is an important process of the current application service.In this study,the bias analysis,clear-rain forecast accuracy,and precipitation TS scoring method were used to evaluate the precipitation forecast at Ankang hydropower Station,Shaanxi Province,China in the flood season from May 1 to September 30,2016 which was provided by the Shaanxi refined numerical forecasting team.The results indicated that the forecast accuracy rate of precipitation is declining with the increase of precipitation forecast aging.The prediction of heavy rain has a high accuracy rate without omission,while the prediction values are less than the actual values.The forecast accuracy rate initialized in 20:00 is higher than that in 08:00,and the accuracy rate at night is higher than that of daytime.Months with more rainfall days have a higher forest accuracy rate according to TS scoring than that of months with less rainfall.The forecast accuracy rate of the Ankang is better than that of Shiquan.The main reason is that the precipitation days (precipitation amount) of Ankang is more than those of Shiquan.These hourly and 3-hourly (within 72 hours) precipitation forecasts can provide a reference for the water dispatching service of Ankang hydroelectric power station.

Key words: numerical forecast, test and assessment, precipitation forecast