Passive microwave humidity profile retrievals using neural networks

A multilayer feedforward neural network was developed for passive microwave relative humidity profile retrievals. Retrievals for radiosonde-based simulated radiances at fifteen frequencies between 23.8 and 183.3 GHz yielded rms retrieval errors in relative humidity of 6-14 percent both over ocean and over land at pressure levels ranging from 131 mb to 1013 mb. These radio frequencies approximated those of the temperature- and humidity-sounding NOAA AMSU and MHS microwave spectrometers planned for U.S. operational low-Earth-orbit meteorological satellites. These retrieval results were comparable or superior to those obtained earlier with an iterative combined statistical and physical retrieval scheme