What Does Lightning Tell Us About Tropical Cyclone Structure Relevant to Forecasting Intensity Change?

Ultimately this project aims to utilize machine learning (ML) in the development of an automated real-time predictive tool to use lightning information from the Geostationary Lightning Mapper (GLM) in forecasting tropical cyclone (TC) intensity change.

Initial ML results have been mixed, with a tendency for overfitting, due to the stochastic nature of lightning, the limited electrification of oceanic convection, and the small sample size for rapidly intensifying TCs.

In this work, we take a step back to assess physical links between the spatial and temporal patterns in GLM lightning and TC structure and structure changes. The structure and structure changes can then in turn be physically linked to intensity changes.

Thus, rather than turning the problem over to ML, we seek to use a physically-based feature engineering approach to reduce dimensionality and overfitting, by transforming the lightning data into a TC structure-relative framework.

TC structure will be assessed using the Advanced Baseline Imager (ABI), and the structure-relative framework will use information about vertical wind shear and wind radii. We will distinguish between different information coming from rainband and inner core lightning and examine changes in storm size.

Since lightning typically increases as storms approach land, we seek clarification on the interpretation of lightning near land and will leverage previous results from this project on lightning climatology over the ocean to account for basin-specific differences.

This work seeks a physical understanding of the added value of lightning information for intensity forecasting when lightning is present in TCs.