Hindcasts of Integrated Kinetic Energy in Atlantic Tropical Cyclones: A Neural Network Prediction Scheme

Abstract

Abstract A new statistical–dynamical scheme is presented for predicting integrated kinetic energy (IKE) in North Atlantic tropical cyclones from a series of environmental input parameters. Predicting IKE is desirable because the metric quantifies the energy across a storm’s entire wind field, allowing it to respond to changes in storm structure and size. As such, IKE is especially useful for quantifying risks in large, low-intensity, high-impact storms such as Sandy in 2012. The prediction scheme, named the Statistical Prediction of Integrated Kinetic Energy, version 2 (SPIKE2), builds upon a previous statistical IKE scheme, by using a series of artificial neural networks instead of more basic linear regression models. By using a more complex statistical scheme, SPIKE2 is able to distinguish nonlinear signals in the environment that could cause fluctuations in IKE. In an effort to evaluate SPIKE2’s performance in a future operational setting, the model is calibrated using archived input parameters from Global Ensemble Forecast System (GEFS) control analyses, and is run in a hindcast mode from 1990 to 2011 using archived GEFS reforecasts. The hindcast results indicate that SPIKE2 performs significantly better than both persistence and climatological benchmarks.