Geostationary satellites, unequipped with microwave instrumentation, cannot see through the high clouds in the atmosphere, obscuring vital information on the formation and evolution of tropical cyclones. However, passive microwave imagery can penetrate through the upper level clouds, providing valuable insights to tropical cyclone storm systems. Even so, hemispheric passes of these polar-orbiter microwave satellites are limited, providing only narrow swaths over the globe. Our team has developed a method of generating synthetic passive microwave (SPM) imagery of tropical cyclone systems from traditional geostationary satellite imagery, revealing the previously unseen internal structure of tropical cyclones at high temporal resolution. For example, the eye of a storm maybe obscured in the geostationary image. By extracting the corresponding microwave imagery through our approach, forecasters and emergency managers could identify the eye of the storm much sooner. This work would provide (1) gain deeper understanding of tropical cyclone evolution (see below) and (2) improve TC forecasting abilities.

Accurately estimating uncertainty in tropical cyclone forecasts is vital for effective disaster preparedness and response. Neural networks are increasingly used for these predictions, but providing reliable uncertainty estimates has been a challenge. Our team is currently working on novel ways to quantify and incorporate uncertainty into Earth system science applications, including tropical cyclone forecasting. We have devised a novel method using sinh-arcsinh-normal distributions to estimate heteroscedastic and asymmetric uncertainties, improving upon traditional techniques.

Select projects: Robust AI