New methods for retrieval of rainfall rate over ocean with SSM/I data

Spaceborne microwave (MW) remote sensing of rainfall distribution with multi-channel radiometers has been proved as a powerful tool in past decade, in particular with DMSP's SSM/I data. Similar instruments but with different channel combinations, such as ADEOS-II/AMSR are being developed. Although there have been several retrieval schemes used for research and operational application, improvement of retrieval accuracy is still an important subject. As part of a NASDA project for ADEOS-II/AMSR retrieval algorithm development and Chinese National High-Tech R and D Program for Space Development, we proposed two kinds of retrieval algorithms for rainfall rate over ocean with SSM/I data. The first kind of algorithm is based on the probability pairing method, in which three different retrieval rain indices composed with different combination of SSM/I's channels are used as pairing indices with surface rainfall rate data provided by NASDA, Japan. Three different empirical rainfall rate--rain index relationship are produced. The second kind of algorithm is based on the method of self organization feature mapping (SOM), a kind of artificial neural network. SSM/I data and co-located surface rainfall data are put into SOM and clustering procedure is self-trained. After training, 154 clustering centers are formed and for each cluster a regressive relationship between retrieved rainfall rate and SSM/I data is established. In this paper, these two kinds of algorithm are briefly reviewed with their developments and validation. Their respective advantages and limitations are discussed.