Stochastic Modeling of Tropical Cyclone Wind Radii

Predictions of tropical cyclone wind radii have seen limited improvement over the past decade. Here we describe new methods and parameters for predicting tropical cyclone wind radii. By combining the SHIPS input predictors with the HURDAT2 and Extended Best Track wind radii we create stochastic models that employ traditional linear regression, random forests, and gradient boosted forest to produce a model to predict 34, 50, and 64 kt tropical cyclone wind radii. We also use several parametric cyclone wind profiles to calculate cyclone inertial stability and angular momentum as additional predictors. The combination of these techniques yields a final model that produces a marginal improvement in forecast errors over existing tools. The results of these models also suggest a difference in the factors that play a role in determining the 64 and 34 kt wind radii, suggesting that different processes are governing the inner and outer core dynamics within cyclones.