Abstract:Based on the forecast data from ECMWF, GFS, GRAPES_GFS, SMSWARMS, ZJWARMS and ZJWARRS, this paper verifies the forecast skills of the six models on the heavy rainfall occurred before, during and after the landfall of Typhoon “Lekima”(1909). Results show as follows:(1) For the 72 h rainfall, ECMWF has the best performance on heavy rainfall location prediction, but shows less skill on the intensity forecast. Three regional models have their own advantages respectively. ZJWARRS and ZJWARMS provide fairly precise forecasts on the torrential rainfall over landfalling site and the northwest of Zhejiang Province. SMSWARMS has good performance on rainfall location prediction as well, but the rainfall intensity is much lower than that obtained from observation. (2) In the case of daily 24 h precipitation, ECMWF and GFS get higher skill scores before the landfall of “Lekima”, while regional models exceed in the subsequent period. SMSWARMS, ZJWARRS and ZJWARMS have great advantage in the forecast of heavy downpour and extraordinary heavy rainfall within the lead time of 12 hours, in particular, ZJWARRS shows outstanding skills in the 0—3 h heavy downpour prediction. (3) As to the precipitation forecast over the flooded area (Yongjia, Linhai and Lin'an), ZJWARMS shows the best skills on both rainfall location and amount, and ZJWARRS puts forward complementary forecasts to ZJWARMS. There are location and intensity biases in the prediction from SMSWARMS data, and global models forecasts are of no much value as a reference in operational services. (4) Diagnosis of multi-meteorological elements using forecast data from five models indicates that the main reason for the great discrepancy in the torrential rain forecast before, during and after the landfall of “Lekima” results from the differences of moisture flux distribution and its convergence. For the provincial area of Zhejiang, ZJWARRS and ZJWARMS have great advantages over other models in the short-term heavy rainfall forecast.