Size Matters: Assessing the Predictive Skill of TC Size Forecasts with a new Hurricane Wind Climatology

The size of a hurricane plays a critical role in its impact potential, especially at landfall. Recent high-impact events such as Hurricanes Ike (2008) and Sandy (2012) have shown that even relatively weak storms (in terms of peak wind speed) with expansive spatial wind fields can cause significant storm surge and tremendous socioeconomic losses. Despite the need for accurate predictions of hurricane size and structure, explicit hurricane size forecasting is largely absent. Numerical models can produce realistic representations of the hurricane wind field, yet it has been difficult to evaluate size predictions, mainly because size estimates in the best-track database are uncertain. As a result, the predictive skill of current forecast models with respect to hurricane size is virtually unknown. Here, we present a new approach to investigate the predictive skill of ensemble forecasts in terms of hurricane size and structure. Using an information theoretical paradigm, predictions of hurricane size are skillful as long as the variance of the forecast size distribution of size is less than that of a climatological distribution. In this study, a climatological frequency distribution of hurricane size is derived from a novel hurricane wind climatology, which is comprised of surface wind analyses from Stepped Frequency Microwave Radiometer measurements acquired during 72 aircraft reconnaissance missions into 21 western Atlantic hurricanes. A high-resolution, 100-member ensemble of Hurricane Earl (2010) provides the forecast distribution. Throughout the 7-day forecast, the distribution of the 34-kt wind radius features less variance than the climatology, implying that the ensemble is able to skillfully predict the overall size of Earl's wind field. This new approach provides an opportunity to evaluate hurricane forecasts beyond the simplistic maximum wind speed metric, and exemplifies a pathway to investigating the predictive skill of next-generation high-resolution prediction ensembles.