Abstract: To study the performance of different numerical models in forecasting rainstorms in Shandong Peninsula, precipitation data from 302 national and regional meteorological stations, the 5 km real-time data (CMPAS-5 km) from the CMA Multi-source Precipitation Analysis System, and products of high-resolution numerical models ECMWF, CMA-GFS, CMA-MESO, and CMA-SH9 (hereinafter referred to as EC, GFS, MESO, and SH, respectively) were used. For the rainstorm cases of the shear line pattern at the edge of western Pacific subtropical high (shear line rainstorms, hereinafter) and the cyclone pattern (cyclonic rainstorms, hereinafter) in Shandong Peninsula from 2021 to 2023, the traditional verification and two spatial verification methods (MODE and SAL) were employed to analyze the forecast ability of the models. The results are as follows. (1) According to traditional verification methods, MESO proved to be optimal for the shear line rainstorms, GFS exhibited the highest missing alarm rate but a lower false alarm rate, while EC had a higher false alarm rate. For cyclonic rainstorms, SH performed the best, and MESO had the lowest missing alarm rate. Overall, meso-scale models outperformed large-scale models. (2) The MODE-derived overall object-based similarity for the cyclonic rainstorms exceeded that for the shear line rainstorms. For the shear line and the cyclonic rainstorms, the mean overlapping area ratio was ranked as MESO > SH > EC > GFS and SH > EC > MESO > GFS, respectively. EC and GFS tended to underestimate the area of shear line rainstorms, whereas SH tended to overestimate it. Almost all models exhibited a northward or westward deviation in rainstorm object centroids for both patterns. Overall, for the shear line rainstorms at the edge of the western Pacific subtropical high, meso-scale models (SH and MESO) were preferred. For the cyclonic rainstorms, priority was given to SH and EC. (3) Based on the SAL spatial verification results, for the shear line rainstorms, MESO was recommended in average precipitation intensity forecasting, whereas large-scale models were more appropriate for the cyclonic rainstorms.