Backpropagation neural-network-based retrieval of atmospheric water vapor and cloud liquid water from IRS-P4 MSMR

Abstract

A new multiparameter retrieval algorithm based on a backpropagation neural network (BPNN) has been developed for deriving integrated water vapor (WV) and cloud liquid water (CLW) contents over oceans from brightness temperatures (BTs) measured by the Multi-frequency Scanning Microwave Radiometer (MSMR) launched onboard Indian Remote Sensing satellite IRS-P4. The MSMR measures brightness temperatures in vertical and horizontal polarizations at 6.0-, 10.65-, 18.0-, and 21.0-GHz frequencies. The data are available at three spatial grid resolutions of 150, 75, and 50 km. In this paper, a BPNN has been trained using brightness temperatures simulated through radiative transfer model and simulated surface and atmospheric parameters. The present algorithm has been compared with the operational MSMR retrieval algorithm based on statistical regression using the same dataset. The validation of WV with in situ data (Vaisala radiosonde) is presented. Moreover, comparison of WV and CLW derived from MSMR using BPNN with the finished products from the Special Sensor Microwave/Imager and the Tropical Rainfall Measuring Mission Microwave Imager has also been carried out. The complexity of the BPNN in retrieval of geophysical products, individually and simultaneously, has also been discussed. Simultaneous retrieval of WV and CLW improves the results.