Tropical Cyclone Intensity Change Predictability Estimates Using a Statistical-dynamical Model

The recently established NOAA Hurricane Forecast Improvement Project (HFIP) has set a 5-year goal to reduce track and intensity forecast errors by 20%, and a 10-year goal for a 50% reduction. The 10-year goal corresponds to an average error reduction of 5% per year. Figure 1 shows the annual average 48 h Atlantic track and intensity errors of the National Hurricane Center (NHC) official forecasts from 1985 to 2009 and the corresponding linear trend lines. The track forecasts have been decreasing at an average rate of about 3.7% per year, so the 5% per year improvement proposed by HFIP represents an acceleration of the error reduction rate. However, the intensity errors have only decreased at about 0.6% per year. Thus, the HFIP goals represent an order of magnitude increase in the improvement rate. In the 1970’s and 80’s the most accurate numerical track forecast models were statistical-dynamical (Neumann 1987, here after N87). This guidance included the NHC73 and NHC83 series of models, where the track was predicted statistically using input from climatology, persistence and the storm environment determined from a global atmospheric model. N87 also indicated that the reductions in the track forecast errors at that time were only about 0.5% per year and appeared to be leveling off. N87 developed an interesting methodology to estimate how much improvement could be achieved in track forecasts from statistical models. For that purpose, the most accurate track forecast model at the time (NHC83) was run in a “perfect prog” mode, where verifying analyses replaced the forecast fields used to estimate the storm environment predictors and best track input replaced the operational estimates of the initial storm position and motion vector. Figure 2 shows the average NHC track errors from 1976-1985 (the baseline period used by N87) and the estimate of the how much improvement was possible. Also shown are the 5-year averages of the NHC track errors from 1986 through 2009. This figure shows that the roughly 50% improvements in track errors estimated by N87 were achieved at most forecast times by the late 1990s and surpassed in the 2000s. These improvements were not, however, achieved by better statistical-dynamical models, but were primarily due to better dynamical models starting with the operational implementation of the GFDL hurricane model in 1995 (Rappaport et al., 2009). _________________ * Corresponding author address: Mark DeMaria, NOAA/NESDIS/STAR, CIRA/Colorado State University, 1375 Campus Delivery, Fort Collins, CO 80523-1375; e-mail: Mark. [email protected] Figure 1. Time series of the annual average NHC official 48 h track and intensity forecast errors since 1985.The linear trend lines of each are also shown.