Estimating rainfall from satellite infrared imagery: Cloud patch analysis

Abstract

Most infrared-based techniques of satellite rainfall estimation contain substantial uncertainties due to the indirect relationship between precipitation particles and space-borne infrared observations of clouds. Generally, these uncertainties include (1) IR temperature threshold defining cold clouds; (2) inclusion of no-rain clouds; (3) exclusion of warm rain clouds; and (4) the coefficients between rain rate and cloud-top properties. To address these uncertainties, a methodology, Cloud Patch Analysis, was developed to estimate rainfall by removing large portion of no-rain clouds from IR cloud imagery. Seven cloud features, including physical, geometric and textural, were defined, and ID3, an inductive decision tree, was used to identify no-rain clouds. Particularly, textural characteristics were extended from square images to irregular cloud patches to extract cloud features related to rainfall. In addition, the method adopted a mechanism to adjust IR temperature threshold according to locations and seasons, and this adjustment can be made by the combination of microwave observations by polar-orbiting satellites with infrared observations by geostationary satellites. The application of the adjusted IR threshold to GPI algorithm showed significant improvement for monthly rainfall estimation. The method was applied to the Japanese Islands and surrounding oceanic regions in June and July/August 1989 and to the Florida region in June and August 1996. The monthly rainfall estimates by the proposed method showed significant and consistent improvements over those by GPI.