Reevaluating the effect of the Tropical Cyclone Environment on Intensity

Tropical cyclone (TC) intensity forecasts from the National Hurricane Center (NHC) have gained skill over the past two decades due to improvements in the guidance suite, which includes statistical-dynamical models such as the Statistical Hurricane Intensity Prediction Scheme (SHIPS) and Logistic Growth Equation Model (LGEM). Both machine learning techniques leverage information consisting of environmental predictors from geostationary satellites and numerical weather prediction models to forecast TC intensity change. While TC forecasts have become more skillful, the NHC is tasked with continuing to make improves to the prediction of rapid intensification, track, and storm surges as well as to create more effective warning systems.

For statistical machine-learning methods to continue to provide more useful information to forecasters, it is vital to continue research into how a TC interacts and reacts to its environment. The release of the fifth generation of ECMWF’s atmospheric reanalysis product (ERA5) presents an opportunity to reevaluate the role of the TC environment on intensification, understand global TC variability, and improve the statistical model predictors. Keeping the diagnostic methods consistent with SHIPS, ERA5 is used to quantify how well the reanalysis represents TC structure, dynamics, and lifecycle. Results show that ERA5 represents TC structure and thus performs best with larger, more organized systems and has some difficulties addressing small-scale features. Nonetheless, ERA5, with its greater resolution, appears to have more realistic TC structures than other reanalyses. Using diagnostics derived rom ERA5, connections between various environmental parameters with storm intensity and intensity change are accessed using Granger causality—Granger causality provides insight into whether a parameter is correlated to an event or provides predictive information. We find that many of the traditional environmental parameters used in intensity forecasting exhibit a causal relationship. But, Granger causality also suggests that several parameters do not have as strong a relationship as previously believed.