在统计分析“降水”和“非降水”视场点的Himawari-8（H8）成像仪（Advanced Himawari Imager，AHI）红外不同光谱亮温梯度变化基础上，开展了H8/AHI资料反演降水初步应用研究。以安徽区域为例，当有降水发生时，AHI通道7至通道16亮温梯度均有变化。采用字典学习和正则化约束法开展降水反演，首先构建匹配的AHI光谱“亮温”和GPM“降水”字典，作为历史样本库；其次基于“字典”利用K-最近邻法进行待反演红外光谱亮温“降水”和“非降水”信号识别；最后在降水信号“子空间”基于正则项约束完成红外资料反演降水。初步试验结果表明基于Gamma概率分布贝叶斯模型平均反演的降水与GPM降水具有较好的结构相似性，误差较小，临界成功指数值较高。进一步将该方法推广应用到AHI光谱亮温反演台风“玛莉亚”降水，得到此方法能够反演出台风的螺旋云雨带。

In this paper, the application of an algorithm for precipitation retrieval is studied based on the statistical analysis of the changes of brightness temperature gradient in different infrared spectra of Advanced Himawari Imager(AHI) of H8 in the field of view of “precipitation” and “non-precipitation”. Taking Anhui region as an example, when precipitation occurs, there is some change in brightness temperature gradient of AHI channel 7-16. Furthermore, dictionary learning and regularization constraints are used on precipitation retrieval. Firstly, based on the H8/AHI spectral brightness temperature data and GPM precipitation, spectral “brightness temperature” and “precipitation” dictionary are matched as historical sample databases. Secondly, K-nearest neighbor (KNN) method is used to identify “precipitation” and “non-precipitation” signals on the brightness temperature of the infrared spectrum based on the “dictionary”. Finally, precipitation retrieval for infrared data is carried out in the precipitation signal “subspace” with regularization constraints. The preliminary experimental results show that precipitation structure based on brightness temperature for H8/AHI, which was retrieved by using the Bayesian model averaging-gamma probability distribution model, has a good similarity with GPM, as well as low relative error, and the critical success index is higher than others. Furthermore, the algorithm is extended and applied to the AHI brightness temperature retrieval of typhoon “Maria” precipitation, and the spiral rain belt can be obtained.