Machine Learning Improves Satellite Rainfall Estimates - Eos

Spaceborne precipitation observing systems can provide global coverage but estimates typically suffer from uncertainties and biases. Conversely, ground based systems such as rain gauges and precipitation radar have higher accuracy but only limited spatial coverage. Chen et al. [2019] have developed a novel deep learning algorithm designed to construct a hybrid rainfall estimation system, where the ground radar is used to bridge the scale gaps between (accurate) rain gauge measurements and (less accurate) satellite observations. Such a non-parametric deep learning technique shows the potential for regional and global rainfall mapping and can also be expanded as a data fusion platform through incorporation of additional precipitation estimates such as outputs of numerical weather prediction models.