Retrieval of hydrometeors from microwave radiances with a polarized radiative transfer model

Abstract

This paper reports the results of a Bayesian-based algorithm for the retrieval of hydrometeors from microwave satellite radiances. The retrieval technique proposed makes use of an indigenously developed polarized radiative transfer (RT) model that drives a data driven optimization engine (Bayesian) to perform retrievals of rain and other hydrometeors in a multi-layer, plane parallel raining atmosphere. For the sake of completeness and for the purposes of comparison, retrievals with Artificial Neural Networks (ANN) have also been done. Retrievals have been done first with a simplified two-layer atmosphere, where assumed values of hydrometeors are given to the forward model and these are taken as ‘measured radiances’. The efficacy of the two retrieval strategies is then tested for this case in order to establish accuracy and speed. The highlight of the work is however, the case study wherein a tropical storm in the Bay of Bengal is taken up, to critically examine the performance of the retrieval algorithm for an extreme event wherein a 14-layer realistic, raining atmosphere has been considered and retrievals are done against Tropical Rainfall Measuring Mission (TRMM) measured radiances. The key novelties of the work are:

• inclusion of polarization from both hydrometeors and oceans in the RT model, and

• populating the database involving atmospheric profiles vs. simulated radiances by profiles of similar rain events in the past.

In this work, the database was populated with TRMM retrieved profiles for tropical cyclones that occurred earlier in the area of interest (Indian Ocean), rather than with the Goddard Cloud Ensemble profiles. The use of (i) polarization in the forward model and (ii) creation of an a priori database for the retrieval denote the significant departure from the current state-of-the-art in the area.