2023年9月6日22:00（北京时， 下同）至7日04:00甘肃夏河县发生强对流天气， 局部地区出现短时强降雨， 引发夏河县果宁村山洪灾害， 造成人员伤亡。本研究基于气象站观测降水对比分析了雷达估测降水（Radar Quantitative Precipitation Estimation， Radar-QPE）、 FY4B估测降水（FengYun 4B Quantitative Precipitation Estimation， FY4B-QPE）以及CMPA（CMA Multi-source Precipitation Analysis）降水产品特性， 并利用这些降水数据驱动水动力水文模型， 评估不同降水数据在本次山洪模拟中的效果。结果表明： （1）12 h累积降水量中， CMPA在大值区域位置和局地降水量级差异性方面表现出较高的准确性； Radar-QPE在累积降水量级上与AWS（Automatic Weather Station）较为接近， 但空间分布上存在显著差异； FY4B-QPE累积降水量级高估了33.8%。（2）在逐小时分布上， CMPA在时间演变、 空间分布以及降水量级上与AWS最为接近； Radar-QPE峰值偏小， 且峰值时间有所滞后， 降水主要为负偏差； FY4B-QPE峰值及峰值时间与实际情况一致， 但在降水的开始和结束时间存在偏差， 降水量的偏差主要为正偏差。（3）水文模拟研究中， CMPA、 Radar-QPE和FY4B-QPE均高估了水位， 但水位峰值出现时间与AWS较为一致， CMPA在均方根误差（RMSE）、 纳什效率系数（NSE）和相对偏差（Bias）方面表现最优， Radar-QPE次之， FY4B-QPE表现相对较差。虽然现有站点观测降水无法完全满足对中小尺度山洪的研究和预警需求， 但CMPA数据的高精度在一定程度上能有效补充传统气象观测站点的不足， 同时， Radar-QPE和FY4B-QPE的算法和精度需要进一步改进和提升。

From 22:00 on September 6， 2023 to 04:00 （Beijing Time） on September 7， Xiahe County in Gansu Province experienced severe convective weather， with short-term heavy rainfall in some areas， causing flash floods in Guoning Village， Xiahe County， resulting in casualties.In this study， the characteristics of Radar Quantitative Precipitation Estimation （Radar-QPE）， FengYun 4B Quantitative Precipitation Estimation （FY4B-QPE）， and CMA Multi-source Precipitation Analysis （CMPA） precipitation products were contrastive analyzed based on meteorological station observations.These precipitation data were used to drive the hydrodynamic hydrological model and evaluate the effect of different precipitation data in the flash flood simulation.The results showed that： （1） Among the 12-hour cumulative precipitation amounts， CMPA demonstrated higher accuracy in terms of the position of large value areas and differences in local precipitation levels； Radar-QPE was closer to AWS （Automatic Weather Station） in terms of cumulative precipitation level but showed significant differences in spatial distribution； FY4B-QPE overestimated the cumulative precipitation level by 33.8%.（2） In terms of hourly distribution， CMPA was most similar to AWS in terms of temporal evolution， spatial distribution， and precipitation level； Radar-QPE's peak values were smaller， and the peak times were lagged， with negative deviations in precipitation being dominant； FY4B-QPE's peak values and peak times were consistent with reality， but there were deviations in the start and end times of precipitation， with positive deviations in precipitation being dominant.（3） In the hydrological simulation study， CMPA， Radar-QPE， and FY4B-QPE all overestimated water levels， but the timing of water level peaks was more consistent with AWS.CMPA performed best in terms of RMSE （Root Mean Square Error）， NSE （Nash Efficiency Coefficient）， and Bias （Relative Deviation）， followed by Radar-QPE， and FY4B-QPE performed relatively poorly.Although existing site-observed precipitation cannot fully meet the needs of research and early warning for small and medium scale mountain floods， the high precision of CMPA data could effectively supplement the deficiencies of traditional meteorological observation stations to some extent.Meanwhile， the algorithms and accuracy of Radar-QPE and FY4B-QPE needed to be further improved and enhanced.