Abstract: The torrential rainstorm that occurred in the city of Huludao,Liaoning Province,on 20-21 August,2024 (referred to as the "8·20" torrential rainstorm) was selected. The applicability of five merged precipitation datasets (C3SPA_1km,C3SPA_5km,C2SPA_5km,C3SPA_1km_10m,L2SPA_1km_10m) in the extreme rainstorm is compared and evaluated by using hierarchical and non-hierarchical assessment under the conditions of complete and missing observation. The results show that five merged precipitation datasets underestimate the actual cumulative precipitation to varying degrees,with the level of short-term heavy precipitation increases,the underestimation of merged precipitation data increases. Among the five datasets,the 3-source 1km data and 3-source 5 km data have the best reconstructing ability under the conditions of complete observation,but their reconstructing ability decreased more significantly under missing observation condition.The 3-source 1 km data and 3-source 5 km data are significantly better than the 2-source 5 km data,the reconstructing ability of Liaoning 1km data is worse than that of the 3-source datasets,it is significantly better than that of the 2-source 5 km data.The nearest grid point may not always accurately reconstruct precipitation at a station,and there may be a more accurate precipitation value within 2 to 3 grids distances from the station. It is suggested that hourly merged precipitation product should be developed to improve assessments and reduce underestimation in observation blind areas.