Infrared geostationary satellite precipitation retrievals trained with AMSU MIT millimeter-wave precipitation retrieval products

Abstract

This paper develops a precipitation retrieval algorithm for the Japanese Advanced Meteorological Imager (JAMI) aboard the Multi-functional Transport Satellites (MTSAT). The JAMI PSU Precipitation retrieval algorithm version 1 (JPP-1) employs neural networks trained and evaluated separately for land and sea using the AMSU MIT Precipitation retrieval (AMP) products retrieved using observations from the passive millimeter-wave spectrometer Advanced Microwave Sounding Unit (AMSU) aboard the U.S. National Oceanic and Atmospheric Administration (NOAA)-18 satellite. Inputs to neural networks include all JAMI infrared channels. Results show that JPP-1 surface precipitation rate retrievals are useful for rates higher than 2 and 1 mm/h for land and sea, respectively, and have good accuracy for rates higher than 4 mm/h for both land and sea. Retrievals for both land and sea are overestimated for rates below 4 mm/h and are underestimated otherwise. Correlation coefficients between AMP surface precipitation rates and JPP-1 retrievals are 0.64 and 0.70 for land and sea, respectively. Main retrieval errors include underestimation for high surface precipitation rates and false alarms due to the inability of JAMI to penetrate clouds.