An ANN Approach to Determine the Radar Cross Section of Non-Rotationally Symmetric Rain Drops

Non-rotationally symmetric rain drops can often be observed in turbulent weather situations. The main reason is the occurrence of asymmetric drop oscillation modes that are induced due to winds and collisions of drops. In recent studies, scattering parameters of thousands of individual drops were determined for C- and S-Band weather radar frequencies, by fully reconstructing the drops that were observed during turbulent weather situations with two-dimensional video disdrometers (2DVD). The computational effort, however, was considerable. In this study, therefore, a feed forward neural network was trained to predict the radar cross section of rain drops only by using a few selected characteristic parameters of the drops as input, all of which can be extracted from 2DVD data. Based on the comprehensive dataset for test, training, and validation, it could be shown that the reported radar cross sections are in general accurate by a fraction of a dB, while the computational effort is negligible.