Analysis and Prediction of Integrated Kinetic Energy in Atlantic Tropical Cyclones

Integrated kinetic energy (IKE) is a recently developed metric that approximates the destructive potential of a tropical cyclone by assessing the size and strength of its wind field. Despite the potential usefulness of the IKE metric, there are few, if any, operational tools that are specifically designed to forecast IKE in real-time. Therefore, IKE and tropical cyclone structure are analyzed within historical Atlantic tropical cyclones from the past two decades in order to develop an understanding of the environmental and internal storm-driven processes that govern IKE variability. This analysis concurs with past research that IKE growth and decay is influenced by both traditional tropical cyclone development mechanisms and by other features such as extratropical transition and trough interactions. Using this framework, a series of statistical prediction tools are created in an effort to project IKE in Atlantic tropical cyclones from a series of relevant normalized input parameters. The resulting IKE prediction schemes are titled the "Statistical Prediction of Integrated Kinetic Energy (SPIKE)". The first version of SPIKE utilizes simple linear regression to project historical IKE quantities in a perfect prognostic mode for all storms between 1990 and 2011. This primitive model acts as a proof of concept, revealing that IKE can be skillfully forecasted relative to persistence out to 72 hours by even the simplest of statistical models if given accurate estimates of various metrics measured throughout the storm and its environment. The proof-of-concept version of SPIKE is improved upon in its second version, SPIKE2, by incorporating a more sophisticated system of adaptive statistical models. A system of artificial neural networks replaces the linear regression model to better capture the nonlinear relationships in the TC-environment system. In a perfect prognostic approach with analyzed input parameters, the neural networks outperform the linear models in nearly every measurable way. The system of neural networks is also more versatile, as it is capable of producing both deterministic and probabilistic tools. Overall, the results from these perfect prognostic exercises suggest that SPIKE2 has a high potential skill level relative to persistence and several other benchmarks. Finally, in an effort to assess its real-time performance, the SPIKE2 forecasting system is run in a mock-operational hindcast mode for the 1990 to 2011 North Atlantic hurricane seasons. Hindcasts of IKE are produced in this manner by running the neural networks with hindcasted input parameters from NOAA's second generation Global Ensemble Forecast System reforecast dataset. Ultimately, the results of the hindcast exercises indicate that the neural network system is capable of skillfully forecasting IKE in an operational setting at a level significantly higher than climatology and persistence. Ultimately, forecasts of IKE from these neural networks could potentially be an asset for operational meteorologists that would complement existing forecast tools in an effort to better assess the damage potential of landfalling tropical cyclones, particularly with regards to storm surge damage.