Neural Network Microwave Precipitation Retrievals and Modeling Results

We describe a simulation methodology used to develop and validate precipitation retrieval algorithms for current and future passive microwave sounders with emphasis on the NPOESS (National Polar-orbiting Operational Environmental Satellite System) sensors. Precipitation algorithms are currently being developed for ATMS, MIS, and NAST-M. ATMS, like AMSU, will have channels near the oxygen bands throughout 50-60 GHz, the water vapor resonance band at 183.31 GHz, as well as several window channels. ATMS will offer improvements in radiometric and spatial resolution over the AMSU-A/B and MHS sensors currently flying on NASA (Aqua), NOAA (POES) and EUMETSAT (MetOp) satellites. The similarity of ATMS to AMSU-A/B will allow the AMSU-A/B precipitation algorithm developed by Chen and Staelin to be adapted for ATMS, and the improvements of ATMS over AMSU-A/B suggest that a superior precipitation retrieval algorithm can be developed for ATMS. Like the Chen and Staelin algorithm for AMSU-A/B, the algorithm for ATMS to be presented will also be based a statisticsbased approach involving extensive signal processing and neural network estimation in contrast to traditional physics-based approaches. One potential advantage of a neural-network-based algorithm is computational speed. The main difference in applying the Chen-Staelin method to ATMS will consist of using the output of the most up-to-date simulation methodology instead of the ground-based weather radar and earlier versions of the simulation methodology. We also present recent progress on the millimeter-wave radiance simulation methodology that is used to derive simulated global ground-truth data sets for the development of precipitation retrieval algorithms suitable for use on a global scale by spaceborne millimeter-wave spectrometers. The methodology utilizes the MM5 Cloud Resolving Model (CRM), at 1-km resolution, to generate atmospheric thermodynamic quantities (for example, humidity and hydrometeor profiles). These data are then input into a Radiative Transfer Algorithm (RTA) to simulate at-sensor millimeter-wave radiances at a variety of viewing geometries. The simulated radiances are filtered and resampled to match the sensor resolution and orientation.