选择对气候变化和人类活动极为敏感的黄河中游为研究区，采用集成模型输出统计方法（EMOS）对多源遥感降水数据（CHIRPS、CMORPH、PERSIANN-CDR和TMPA 3B42）进行数据融合，并对其基本统计性能和降水等级性能进行评估。研究发现：采用EMOS数据融合方法获得的降水数据集可更好地捕捉观测降水的空间分布和年际均值，其基本统计性能（δBIAS=-3.4%；δRMSE=13.1 mm；γKGE=0.233）明显优越于4种独立的遥感降水产品、4种产品的均值（MME）以及采用分位数映射法（QM）获取的数据集，但各统计指标存在较大的空间差异性；EMOS方法获得降水数据集显著地改善了对中雨的检测性能（探测率DPOD=0.38）；相对MME，EMOS对大雨的综合探测能力（关键成功指数ECSI)提高了12%，表明该方法对高强度降水的正确检测能力更具优势。基于EMOS多源数据融合方法获得的高精度数据集，可为黄河中游典型生态恢复区的极端降水水沙关系研究提供科学支撑，具有较好的科学价值和应用前景。
The error in the precipitation estimation will bring great uncertainty to hydrological process simulation and climate change prediction research.A series of remote sensing precipitation products have provided the possibility to accurately estimate largescale precipitation.However,the remote sensing products from different sources are quite inconsistent.It is urgent to obtain highresolution,highprecision precipitation data sets through multisource data fusion technology,with the purpose of reducing the error of single remotesensing precipitation products and improve the accuracy of precipitation estimation.The middle reaches of the Yellow River is located in the semiarid and semihumid transition zone of China,which is a crucial area for the ecological environment restoration projects.Precipitation is the main water resources for this region,while an accurate precipitation estimation is of great singificance to the sustainable development of the regional ecological environment and the stability of economic life. Based on the background,the middle reaches of the Yellow River are chosed as the research area,which is extremely sensitive to the climate change and human activities.The methods of ensemble model output statistical (EMOS),quantile mapping (QM),and simple mean to the four productions (MME),while the indexes of four basic statistical indicators of correlation coefficient,relative deviation,root mean square error,and KlingGupta coefficient,as well as three category indexes of detection rate (POD),false alarm rate (FAR),and critical success index (CSI),to quantitatively evaluate the performance of four precipitation products (CHIRPS,CMORPH,PERSIANNCDR and TMPA 3B42),and then compare and analyze the performance of different data fusion data sets. The results show that:(1) The fusion data set of EMOS can significantly improve the index of bias,root mean square error,and KGE,with the values of -3.4%,13.1 mm,γKGE=0.233,respectively,compared with MME and individual product.The stations with better performance of QM were mainly located in the southern region,while the better performance for EMOS was observed in the eastern region.(2) Both the QM and EMOS methods have significantly improved the performance of POD,with the average values reaching 0.44 and 0.38,respectively,but the spatial distribution POD calucated from EMOS data set is more concentrated.The critical success index (CSI) of QM and EMOS methods reached 0.30 and 0.28,respectively,increasing approximately 12% compared with MME.The QM method did not significantly improve the POD,but the POD is doubled with a value of 0.33 by the EMOS method. In summary,the data fusion results obtained by the EMOS method have significant advantages over other methods,and the performance improvements are spatially different.The precipitation data set obtained by the EMOS method significantly improves the detection ability for moderate and heavy rain,indicating better performance for detecting the highintensity precipitation events.The precipitation data set fused with EMOS can well provide data support for the study of the relationship between extreme precipitation and sediment transport in typical ecological restoration areas in the middle reaches of the Yellow River,and provide a scientific basis for the rapid restoration of vegetation on precipitation and other climate feedback.
国家自然科学基金（41701019；41971035）；中国科学院地理科学与资源研究所“秉维”优秀青年人才计划项目（2017RC204）；南京 信息工程大学2020年“优秀本科毕业论文（设计）支持计划”项目；南京信息工程大学人才启动项目（2017r069）；中国科学院陆地 水循环及地表过程重点实验室开放基金项目（2017A004）