Quantifying Spatio-temporal Convective Structure in Tropical Cyclones

Tropical Cyclones (TCs) are powerful, organized storms which rank among the deadliest and costliest natural disasters in the United States and other coastal countries around the globe. While traditional spatiotemporal statistical methods such as Gaussian process models have been utilized to great effect in a variety of meteorological and oceanographic applications, TC analysis and forecasting pose a unique set of challenges which require carefully tailored statistical approaches. Foremost among these is explainability. Scientists’ goals are to learn about the underlying physical processes driving TC behavior, while TC forecasts are ultimately issued by human meteorologists. It is therefor not enough for a statistical method to achieve good estimates or forecast accuracy; we must also be able to explain in terms of the underlying physics why our models make a given prediction. In this thesis, we develop a set of methods for utilizing geostationary satellite imagery in principled and explainable statistical analyses. This work is divided into three main parts: 

We first present analytic tools that quantify convective structure patterns in infrared satellite imagery for over-ocean TCs, yielding lower-dimensional but rich representations that support analysis and visualization of how these patterns evolve during rapid intensity change. The proposed ORB feature suite targets the global Organization, Radial structure, and Bulk morphology of TCs. By combining ORB and functional PCA, we arrive at an interpretable and rich representation of convective structure patterns that serve as inputs to machine learning methods. We then demonstrate via a linear classification that ORB provides valuable information regarding intensity change which is complementary to fundamental environmental factors such as vertical wind shear and sea surface temperature. 

We then propose a new nonparametric test of association between a time series of images and a series of binary event labels. By rewriting the statistical test as a regression problem, vii we leverage neural networks to infer modes of structural evolution of TC convection that are representative of the lead-up to rapid intensity change events. Dependencies between nearby sequences are handled by a bootstrap procedure that estimates the marginal distribution of the label series. We prove that type I error control is guaranteed under the assumption of stationarity as long as the distribution of the label series is well-estimated. We show empirical evidence that our proposed method identifies archetypes of evolving structure associated with elevated rapid intensification risk, typically marked by deep or deepening core convection over time. Such results provide a foundation for improved forecasts of rapid intensification. 

Finally, we present a new method for probabilistic forecasting not only of TC intensity but also TC convective structure as captured by ORB. We describe a prototype model based solely on observed infrared imagery and past operational intensity estimates. These structural forecasts simulate the spatio-temporal evolution of the radial profiles of cloudtop temperatures over the subsequent 12 hours, then predict intensities based on simluated structure. Intensity guidance from our prototype model achieves a marginally higher error than the National Hurricane Center’s official forecasts. We demonstrate that it is possible to reasonably predict short-term evolution of TC convective structure as depicted by infrared imagery, producing interpretable structural forecasts that may be valuable for TC operational guidance.