Learning-Based Rainfall Estimation via Communication Satellite Links

We present a method for estimating rainfall by opportunistic use of Ka-band satellite communication network. Our approach is based on the attenuation of the satellite link signal in the rain medium and exploits the nearly linear relation between the rain rate and the specific attenuation at Ka-band frequencies. Although our experimental setup is not intended to achieve high resolutions as millimeter wavelength weather radars, it is instructive because of easy availability of millions of satellite ground terminals throughout the world. The received signal is obtained over a passive link. Therefore, traditional weather radar signal processing to derive parameters for rainfall estimation algorithms is not feasible here. We overcome this disadvantage by employing neural network learning algorithms to extract relevant information. Initial results reveal that rainfall accumulations obtained through our method are 85% closer to the in situ rain gauge estimates than the nearest C-band German weather service Deutscher Wetterdienst (DWD) radar.