Generative Deep Learning to Reconstruct Tropical Cyclone Microwave Brightness Temperature from Geostationary Infrared Radiometers

Tropical cyclones (TCs) are cyclonic low-pressure circulation with warm-core structures that form over tropical oceans and are one of the most destructive weather systems in the world. They usually cause natural disasters such as strong winds, heavy precipitation, and storm surges, which bring many casualties and economic losses. Most life cycles of TCs are located in the ocean, which is a blind spot for conventional observations. Meteorological satellites have become an important means of all-weather monitoring of TCs since the 1960s. The infrared sensor carried by stationary satellites has the advantages of high temporal and spatial resolution and wide spatial coverage, but it can only get cloud-top information. The microwave sensor carried by low-orbit satellites can penetrate the cloud top and reveal the convection structure. Still, the temporal and spatial resolution are limited, and it is difficult to observe the moving TC continuously. We use generative deep learning models to convert from infrared to microwave brightness temperatures for application in the TC study. Experimental results reveal that increasing the types of input conditions and generation targets can improve the generation effect. This work can compensate for the lack of TC range microwave brightness temperature data, and provide support for real-time TC monitoring, data assimilation, and improvement capabilities of TC intensity and size estimation and prediction.