Abstract:Combining the gauged precipitation and the predicted precipitation in 24 h and 48 h of regional numerical model named GRAPES during 2016 to 2017, GRAPES was quantitatively evaluated in the four provinces （regions） of Northwest China, using the indices including mean error, root mean square error, correlation coefficient and TS score. The results showed that the forecast accuracy of rain probability was higher than 0.7, daily spatial correlation coefficient was 0.2-0.4. The highest bias appeared in summer, the mean errors of 24 h and 48 h forecast were 4 mm·d-1 and 6 mm·d-1, the root mean square errorr were 6 mm·d-1 and 8 mm·d-1, respectively. The TS of heavy rain and above was less than 0.1, TSs of light rain and moderate rain were 0.2-0.5 and 0.1-0.2, respectively. Spatially, the 24 h and 48 h forecast accuracies of rain probability were higher than 0.6 in most regions, the correlation coefficients in eastern Gansu, middle and southern Shaanxi were higher than 0.6. The highest mean error of 24 h forecast appeared in the southern part of Qinghai, Gansu and Shaanxi, which reached to 2-4 mm·d-1. The highest mean error of 48 h forecast reached to 5~8 mm·d-1 and appeared in the southern Shaanxi. The mean error of 48 h forecast was 1~2 mm·d-1 higher than 24 h forecast in other regions, The TS score of 24 h forecast for each precipitation level was obviously better than that of 48 h, 24 h forecast could predict heavy rain and rainstorm, while 48 h forecast showed poor ability for moderate rain and above.