Deep Rainrate Estimation From Highly Attenuated Downlink Signals Of Ground-Based Communications Satellite Terminals

While the use of weather radars to continuously monitor the spatio-temporal dynamics of precipitation has grown in recent years, these systems are expensive and sparsely deployed across the world. In this context, densely located ground-based terminals for interactive satellite services have the potential for dual-use as weather sensors because they measure rain-attenuated power of the downlink signal. Although in the millimeter-wave regime, the rain rate has almost a linear relationship with specific attenuation, lack of other weather radar observables at satellite terminals imposes a daunting task of extracting rainfall rate from these highly attenuated signals. We address this problem by designing a deep convolutional neural network (CNN) that learns the relationship between the signal attenuation and rainfall rate observed by weather radars and rain gauges at a given location. During the prediction stage, the CNN accepts downlink attenuation as input and classifies the rain intensity which is then used to apply an appropriate rainfall estimator. Our experiments with real data show that, despite severe attenuation, CNN-based downlink rainfall accumulations closely follow the nearest C-band German weather service Deutscher Wetterdienst (DWD) radar.