Can we hope to improve prediction of tropical cyclone intensity?

During the past 15-20 years, 1-5 day tropical cyclone (TC) track forecasts have steadily improved. Nevertheless, during this same timeframe forecasts of TC intensity, defined by NHC as the peak 1-minute sustained 10-m wind anywhere in the storm, have shown little to no improvement. Annual U.S. hurricane losses average about $10 billion and a recent historical analysis of hurricane damages suggests a doubling of economic losses from landfalling hurricanes every ten years. The need for substantial improvements in TC track and intensity forecast capabilities have never been greater. In 2008 NOAA established the NOAA Hurricane Forecast Improvement Project (HFIP), a 10-year effort to accelerate improvements in one to five day forecasts for TC track, intensity, storm surge and to reduce forecast uncertainty, with an emphasis on rapid intensity change, defined as a 30 kt change in intensity in 24 h.

The first three years of HFIP have led to improvements in global and regional TC models and development of data assimilation technique to improve initial analyses of the TC inner core observations from aircraft and Doppler radars. These developments show promise for achieving HFIP's goal to improve TC track and intensity forecast guidance by 20% in 5 years. However, the intensity forecast guidance improvement is manifest at forecast times >36 h, and suggest that while the analyses are improved through data assimilation, there is a bias in the short-term (<36 h) model spin-up. One of the key questions HFIP is trying to address is what causes this short-term model spin-up bias. However, a more important question that needs to be addressed is whether models will ever be able to provide guidance on a metric as volatile as the peak wind anywhere in the storm. Using a peak of a distribution as a metric is fraught with many pitfalls as it is not a representative measure of the distribution of the wind field. In this talk other metrics more representative of the wind field distribution will be presented and discussed as potential means to address the model intensity verification problem.