PrecipGAN: Merging Microwave and Infrared Data for Satellite Precipitation Estimation Using Generative Adversarial Network.

Global satellite precipitation estimation at high spatiotemporal resolutions is crucial for hydrological and meteorological applications but is still a challenging task. One major challenge is that the microwave data are discontinuous in space and time. We present a novel approach to merge incomplete passive microwave (PMW) precipitation estimates using the conditional information provided by complete infrared (IR) precipitation estimates based on the generative adversarial network (GAN), and name the algorithm PrecipGAN. PrecipGAN decomposes the precipitation system into content and evolution subspaces to propagate PMW estimates to regions outside the orbit coverage of PMW sensors. PrecipGAN can skillfully simulate the spatiotemporal changes of precipitation events, and produce precipitation estimates with overall better statistical performance than the baseline product Integrated MultisatellitE Retrievals for GPM (IMERG) Uncalibrated over the Continental US. PrecipGAN provides an alternative of accurate and computationally efficient algorithm that can be implemented globally to produce satellite‐based precipitation estimates. Plain Language Summary: Producing accurate precipitation estimates globally using satellite data at high spatial and temporal resolutions remains a major challenge for the remote sensing community. Uncertain precipitation products limit our capacity in various applications, such as hydrology, meteorology, or climatology. One difficult part in satellite precipitation estimation lies in combining infrared with passive microwave data effectively, as infrared data suffers from low detection accuracy but is superior in terms of spatiotemporal coverage while passive microwave data suffers from low spatiotemporal coverage but does excellent jobs at detecting precipitation accurately. Here, we present a novel deep learning model based on generative adversarial network to merge infrared and passive microwave data with a perspective based on the physical process of precipitation. The performance of the new algorithm is verified through comprehensive comparison with a state‐of‐the‐art precipitation product. The results show that this new algorithm has the potential to be a powerful alternative for generating global precipitation product merging infrared and microwave data. Key Points: A deep learning model called PrecipGAN is developed to merge passive microwave with infrared signals for seamless precipitation estimationPrecipGAN is designed based on the physical process of precipitation by decomposing it into content and evolution subspacesPrecipGAN can produce precipitation estimates with overall better statistical performance compared to an operational satellite product [ABSTRACT FROM AUTHOR]