Advanced Rain/No-Rain Classification Methods for Microwave Radiometer Observations over Land

Abstract

Abstract Seto et al. developed rain/no-rain classification (RNC) methods over land for the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI). In this study, the methods are modified for application to other microwave radiometers. The previous methods match TMI observations with TRMM precipitation radar (PR) observations, classify the TMI pixels into rain pixels and no-rain pixels, and then statistically summarize the observed brightness temperature at the no-rain pixels into a land surface brightness temperature database. In the modified methods, the probability distribution of brightness temperature under no-rain conditions is derived from unclassified TMI pixels without the use of PR. A test with the TMI shows that the modified (PR independent) methods are better than the RNC method developed for the Goddard profiling algorithm (GPROF; the standard algorithm for the TMI) while they are slightly poorer than corresponding previous (PR dependent) methods. M2d, one of the PR-independent methods, is applied to observations from the Advanced Microwave Scanning Radiometer for Earth Observing Satellite (AMSR-E), is evaluated for a matchup case with PR, and is evaluated for 1 yr with a rain gauge dataset in Japan. M2d is incorporated into a retrieval algorithm developed by the Global Satellite Mapping of Precipitation project to be applied for the AMSR-E. In latitudes above 30°N, the rain-rate retrieval is compared with a rain gauge dataset by the Global Precipitation Climatology Center. Without a snow mask, a large amount of false rainfall due to snow contamination occurs. Therefore, a simple snow mask using the 23.8-GHz channel is applied and the threshold of the mask is optimized. Between 30° and 60°N, the optimized snow mask forces the miss of an estimated 10% of the total rainfall.