Neural Network Rainfall Estimation Based on GPM Dual-frequency Precipitation Radar Measurements

We implement a novel hybrid machine learning-based hybrid system consisting of two deep neural networks (DNNs) for GPM applications. This architecture estimates rainfall by building a relation between GPM radar observation and rain gauge measurement by using ground radar to bridge the gap between the spaceborne radar and ground rain gauge. The first DNN model is trained from gauge measurements to ground radar rainfall estimations. The second DNN is trained from ground radar rainfall estimation to spaceborne radar rainfall estimation. Using the two DNN models, the entire system can generate a rainfall product by linking spaceborne radar observations to ground rain gauge measurements via ground radar observations.