Abstract:The precipitation forecasts made by three global models, ECMWF-IFS, NCEP-GFS, and GRAPES-GFS, are verified based on the meteorological and hydrological observations over Yalong River Basin in western Sichuan Province in 2018 flood season. Generally, the 3 d precipitation forecasts of the three models consist fairly well with observations, especially for light rains in which omissions are less but false alarms are more common. However, in moderately and heavily rainy days, precipitation forecasts appear less useful. These forecasts work the best when east-moving plateau trough coupled with southwest stream dominates, but are the worst when weather induced by low-level shears along with southwest stream is more prevailing. When the west-extending subtropical high dominates, their skills in predicting rainfall are unstable. Moreover, intercomparisons of these three precipitation forecasts show that in Yalong River Basin ECMWF-IFS performs best, GRAPES-GFS ranks next, but more care needs to be taken when NCEP-GFS is applied in daily operation. ECMWF-IFS usually forecasts more rainfall and its TS scores for moderate and heavy rains are usually higher than the TS scores of the other two models, but it more probably makes false alarms for light rains. Precipitation forecasted by GRAPES-GFS is less than others, TS scores for light rains are the highest among the three models, while its ability of predicting moderate and heavy rains is still weaker than ECMWF-IFS. NCEP-GFS can forecast light rain well, but more easily misses moderate and heavy rains.