Probabilistic Forecasts of Tropical Cyclone Intensity for the North Atlantic

Although tropical cyclone track forecasts continue to show steady improvement, intensity forecasting remains extremely difficult. In fact, since 1990 the accuracy of tropical cyclone intensity forecasts has shown only small improvement in forecast skill relative to climatology (Rappaport et al. 2012). The minor increase in skill is almost entirely attributed to statistically-based forecasts utilizing the Statistical Hurricane Intensity Prediction Scheme (SHIPS), Decay-SHIPS, and the Logistic Growth Equation Model (LGEM). In fact, the LGEM is now regarded as one of the most skillful intensity forecast models (dynamical or statistical) at extended forecast lead-times in the North Atlantic (Cangialosi and Franklin 2011). Currently, the operational version of LGEM utilizes a deterministic track forecast to derive a growth term, which is a function of vertical wind shear and conditional instability, along with an upper-bound intensity term based on maximum potential intensity theory. Here we evaluate whether tropical cyclone intensity forecasting may be made more skillful using an ensemble of track and environment forecasts as input parameters for LGEM to generate a large set of ensemble intensity forecasts. Previously, Musgrave et al. (2012) has explored the utility of an LGEM-ensemble approach, but with respect to generating a consensus intensity forecast based on a small set of deterministic forecast models. Instead, our goal is to better characterize the intensity uncertainty for a tropical cyclone especially at extended lead-times based on a large set of dynamical forecasts of the tropical cyclone's large-scale environment. Using the ECMWF high-resolution atmospheric model in conjunction with its ensemble counterparts, we construct statistical-dynamical intensity forecasts for North Atlantic tropical cyclones during the period 2010-2012. Preliminary results show that the LGEM-ensemble approach provides lower ranked skill scores than either the raw or bias-corrected ECMWF ensemble forecasts. The new approach also leads to an improved spread-skill relationship especially at extended forecast lead-times precluding the need for additional post-processing.