Machine Learning Approach to Estimate 3-D Tropical Cyclone Kinematic Structure

Learn how machine learning can revolutionize tropical cyclone forecasting by estimating three-dimensional wind structures from satellite observations in this 48-minute conference presentation. Discover the challenges of accurately representing tropical cyclone structure for high-resolution numerical model initialization and hazard communication, particularly the limitations of current tail Doppler radar (TDR) data coverage due to aircraft availability and spatial gaps. Explore the development of TC-SWARM (Tropical Cyclone Synthetic Wind Analysis using Regression Models), a suite of machine learning models trained on historical TDR analyses to establish statistical relationships between 3D tropical cyclone wind structures and globally available geostationary satellite observations. Examine how this innovative approach combines satellite data with operationally available environmental and intensity diagnostics to produce 3D wind field estimates with skill comparable to current best-track intensity uncertainties. Understand the methodology behind the machine learning suite, evaluate performance results from independent testing datasets, and see practical examples of how TC-SWARM fills observational data gaps by reconstructing training cases, ultimately advancing the accuracy of tropical cyclone structure representation for improved forecasting capabilities.