Understanding Hurricane Intensity Predictability Limits and Model Error Using two Different Ensemble Techniques

Operational centers are routinely forecasting TC intensity measured by the maximum surface wind speed out to 5 days and plans are made for increasing this lead time to 7 days in the near future. However, the question of what is the predictability limit of the peak wind speed that is controlled by both the hurricane mean vortex and small-scale features such as convectively generated wind gust has not been answered. This study aims to address the question of TC intensity predictability limits and forecast uncertainty due to model error by using ensemble forecasts generated from two different ensemble techniques. Ensemble A is a physical parameterization ensemble in which each member uses a different physical parameterization scheme, which represents the uncertainty due to parameterized physical processes. Ensemble B uses a stochastic kinetic energy backscatter algorithm. This algorithm injects small amounts of kinetic energy that were lost to diffusion back onto the resolvable scale by adding a small term to the u,v and T tendency equations in a stochastic pattern at each time step. Both five day forecast ensembles are comprised of 10 members, using the WRF model with triply-nested vortex-following domains and 12-4-1.33 km grid spacing. The ensembles are analyzed in terms of the maximum wind speed, azimuthally averaged wind speed and evolution of mean error kinetic energy. The difference in maximum wind speed between the weakest and strongest member in ensemble A reaches 65 kt four days into the forecast, proving the serious difficulty of accurately forecasting maximum peak winds. While the spread is not as pronounced in ensemble B, it still reaches 20 kt after 36 h forecast time. However, ensemble B also shows that the forecast uncertainty of storm size measured by azimuthally averaged wind speed remains small until 60 h, but grows significantly thereafter. The ensemble mean error kinetic energy growth characteristics were used to infer scale-dependent predictability limits of features within the hurricane. For convective scale features O(1km), the predictability limit is ~ 3 hours. For larger scale features such as inner-core rainbands, the preliminary results suggest a predictability limit of 12-24 h. The size of the hurricane vortex (wavenumber zero) and wavenumber one asymmetry have a predictability limit larger than five days, as the analysis shows that the mean error kinetic energy is not saturated for this scale at the end of the forecast. However, the forecast uncertainty increases with longer forecasts.