Predicting Tropical Cyclone Intensity Forecast Error

The lack of consistency between the current operational tropical cyclone intensity forecast models and large fluctuations in model performance decreases the value of tropical cyclone forecasts. One approach to creating more reliable tropical cyclone intensity forecasts with the resources currently available is to create real-time skill predictions that help forecasters and end users know when a particular model forecast will be more or less skillful than average. This a priori expectation of forecast performance combats the adverse effects of the substantial day-to-day, model-to-model, and storm-to-storm fluctuations in forecast quality.

As a first step towards providing real-time error predictions to accompany each tropical cyclone intensity forecast, Bhatia and Nolan (2013) studied the relationship between synoptic parameters, TC attributes, and forecast error. Their results indicate that certain storm environments are inherently more challenging for individual models to forecast. In this study, we build on previous results by using storm-specific characteristics as well as parameters representing initial condition error and the stability of the atmospheric flow to predict forecast error. Error predictions will be generated for 24-120 hour intensity forecasts for LGEM, DSHP, HWFI, and GHMI. The ability of different regression techniques will be compared to evaluate which method will provide the most utility in an operational setting.