In this study, we assess the characteristic bias in the diurnal precipitation forecasts from two models—CMA-SH9 and CMA-MESO—for hourly precipitation forecasts across four subregions in the Hedong Region of Gansu Province (the Gannan Plateau and the Longnan, Longdong, and Longzhong regions). These forecasts were assessed based on rainfall amount and frequency, using observed hourly precipitation data from 20 storm rainy days between May and September 2019-2021 in the same region. The results show the following: (1) Both models have similar forecasting abilities for precipitation amounts of ≥2.5 mm·h^-1 and ≥5 mm·h^-1 in the Gannan Plateau and the Longnan and Longdong regions. However, the CMA-MESO model facilitates easier precipitation forecasts of ≥5 mm·h^-1 in the Longzhong region than the CMA-SH9 model. (2) For rainfall ≥2.5 mm·h^-1, both models overestimate intensity during the day and underestimate it at night compared to observations, with significant deviations mostly occurring around the peak time of observed rainfall at night. The CMA-SH9 model generally outperforms the CMA-MESO model in predicting rainfall intensity in the Longzhong and Longdong regions most times of the day. In the Longnan region, the CMA-SH9 model performs better than the CMA-MESO model in predicting rainfall intensity at night but performs worse during the day. For rainfall ≥5 mm·h^-1, the CMA-SH9 model consistently predicts rainfall intensity better than the CMA-MESO model in the Longnan and Longdong regions, but worse in the Gannan Plateau, most of the time. (3) For rainfall ≥2.5 mm·h^-1, both models predict higher rainfall frequency during the day and lower frequency at night compared to observations, with significant deviations mostly occurring around the peak time of observed rainfall at night. The CMA-SH9 model mostly outperforms the CMA-MESO model in predicting rainfall frequency. For rainfall ≥5 mm·h^-1, both models underestimate rainfall frequency compared to observations in the Longnan, Longzhong, and Longdong regions, with the negative forecast deviation more pronounced around the peak time of observed rainfall at night. (4) Phase error are predominant across the four subregions for both models, while amplitude error was minimal.