Rain Rate Retrieval Algorithm for Conical-Scanning Microwave Imagers Aided by Random Forest, RReliefF, and Multivariate Adaptive Regression Splines (RAMARS)

This paper proposes a rain rate retrieval algorithm for conical-scanning microwave imagers (RAMARS), as an alternative to the NASA Goddard profiling (GPROF) algorithm, that does not rely on any a priori information. The fundamental basis of the RAMARS follows the concept of the GPROF algorithm, which means, being consistent with the Tropical Rainfall Measuring Mission (TRMM) precipitation radar rain rate observations, but independent of any auxiliary information. The RAMARS is built upon the combination of state-of-the-art machine learning and regression techniques, comprising of random forest algorithm, RReliefF, and multivariate adaptive regression splines. The RAMARS is applicable to both over ocean and land as well as coast surface terrains. It has been demonstrated that, when comparing with the TRMM Precipitation Radar observations, the performance of the RAMARS algorithm is comparable with the 2A12 GPROF algorithm. Furthermore, the RAMARS has been applied to two cyclonic cases, hurricane Sandy in 2012, and cyclone Mahasen in 2013, showing a very good capability to reproduce the structure and intensity of the cyclone fields. The RAMARS is highly flexible, because of its four processing components, making it extremely suitable for use to other passive microwave imagers in the global precipitation measurement (GPM) constellation.