Rainfall Estimation from Tempest-D Cubesat Observations

This paper presents a machine learning model to estimate surface rainfall from TEMPEST-D observations. An artificial neural network (ANN) was chosen to build the rainfall estimation model from TEMPEST-D measurements. TEMPEST-D brightness temperature (TB) observations performed at five frequencies (i.e. 87, 164, 174, 178 and 181 GHz) were used as inputs, and the Multi-Radar/Multi-Sensor System (MRMS) radar-only rain rate product at the surface was used as ground truth and target to train the ANN model. A spatial alignment algorithm was developed to align the TEMPEST-D observed storm with the storm measurement from ground radar. The training data set was generated from 14 storm events observed simultaneously by the ground radar network and TEMPEST-D over the continental U.S. Two storm events were used for independent testing. The testing showed that estimated rainfall matched well with the MRMS surface rainfall product in terms of rainfall intensity, area, and precipitation system pattern. The structural similarity index measure scores for the two independent test cases are 0.72 and 0.81.