RainGAN: A Conditional Rain Fields Generator

Abstract

Rain fields’ simulation is an important tool for several research fields and applications. However, most simulations are based on a naive model that cannot capture complex spatial distribution. In this work, we present RainGAN, a generative model that enables a generation of a realistic, complex rain field that is conditioned on user parameters such as max peak, number of peaks, etc. In addition, we construct a dataset of typical rain fields that are based on radar measurement and have been utilized in the training process. We conducted several experiments and demonstrate the generator quality using both numerical and visual results.