Exploring Tropical Cyclone Structure and Evolution with AI-based Synthetic Passive Microwave Data

Satellite observations are an important part of systems for monitoring tropical cyclone formation and evolution, for gathering information about tropical cyclone structure and behavior, and for analyzing tropical cyclone position and intensity. Geostationary satellites provide a wealth of data, such as visible and infrared images of tropical cyclones. Geostationary satellites have high temporal frequency and broad spatial coverage; however, geostationary satellite channels are typically limited to observing features at cloud-top height. Passive microwave imagery, unlike infrared and visible satellite imagery, can exhibit sensitivity to an array of surface and atmospheric variables, depending on the microwave frequency. At frequencies between 85 and 91 GHz, passive microwave imagers can see through cirrus clouds, providing critical insights regarding tropical cyclones’ convective structure and evolution. However, to have imagery at a useful horizontal resolution at these frequencies, the satellites use low-Earth orbits, meaning they are limited to at most a handful of overpasses per day over a given location. This poor temporal resolution of passive microwave imagery is a limiting factor in using this data to study the evolution of tropical cyclone convective structures. Recently, researchers at CIRA and NESDIS have used machine learning algorithms to build a synthetic passive microwave (SPM) dataset in order to address some of these issues. The SPM dataset ingests the full suite of visible and infrared imagery (that is, all channels) from geostationary satellites into an ensemble of fully-connected artificial neural network (ANN) and convolutional neural network (CNN) configurations that output a synthetic 89-GHz passive microwave imagery at temporal frequencies that match that of the geostationary satellite data. The SPM dataset consists of thousands of images of global tropical cyclones observed over the past 5 years. This dataset forms the basis of a study of various aspects of tropical cyclone structure and evolution at higher temporal frequencies not previously available. This work is focused on analyzing the SPM dataset using traditional statistical models as well as artificial intelligence (AI) methods, with a focus on using AI to gain new physical insights into convective structure evolution inside tropical cyclones.

Disclaimer: The scientific results and conclusions, as well as any views or opinions expressed herein, are those of the author(s) and do not necessarily reflect those of NOAA or the Department of Commerce.