Information theoretic approach using neural network for determining radiometer observations from radar and vice versa

Abstract

Even though both the rain measuring instruments, radar and radiometer onboard the TRMM observe the same rain scenes, they both are fundamentally different instruments. Radar is an active instrument and measures backscatter component from vertical rain structure; whereas radiometer is a passive instrument that obtains integrated observation of full depth of the cloud and rain structure. Further, their spatial resolutions on ground are different. Nevertheless, both the instruments are observing the same rain scene and retrieve three dimensional rainfall products. Hence it is only natural to seek answer to the question, what type of information about radiometric observations can be directly retrieved from radar observations. While there are several ways to answer this question, an informational theoretic approach using neural networks has been described in the present work to find if radiometer observations can be predicted from radar observations. A database of TMI brightness temperature and collocated TRMM vertical attenuation corrected reflectivity factor from the year 2012 was considered. The entire database is further classified according to surface type. Separate neural networks were trained for land and ocean and the results are presented.