|
Estimating the probability of tropical cyclone rapid intensification (RI)Collaborating scientist: Mark DeMaria (NESDIS) The TPC/NHC has identified predicting the onset, duration and magnitude of tropical cyclone rapid intensification (RI) events as their number one forecast priority. Since current operational tropical cyclone intensity forecast models have not yet demonstrated the ability to predict rapid intensification events adequately, Kaplan and DeMaria (2003) developed a simple index for estimating the probability of RI (defined as the 95th percentile of all over-water 24-h intensity changes or 30 kt) for systems in the Atlantic basin using SHIPS model output (DeMaria et al. 2005). The threshold version of the RI index compares the magnitude of seven RI predictors determined from operational SHIPS model output to previously determined RI thresholds. It was provided in real-time to the Tropical Prediction Center/National Hurricane Center (TPC/NHC) from 2001-2003 with financial support from the NOAA Joint Hurricane Testbed (JHT) and was made operational by TPC/NHC starting with the 2004 Hurricane season. Recently, an RI index analogous to the Atlantic version was also developed for use in the E. Pacific basin with financial support from the JHT. An RI threshold of 35 kt in 24-h was employed in the E. Pacific basin since this corresponded to the 95% percentile of all over-water 24 hr intensity changes in that basin. The E. Pacific RI index was run experimentally during the 2004 hurricane season. A weakness of the threshold version of the RI index is that the estimates of the probability of RI are determined by comparing the magnitudes of each of the RI predictors to a single (fixed) RI threshold. Consequently, an RI predictor that barely meets an RI threshold (and thus is viewed as marginally conducive for RI) is given the same weight as one that is much more conducive to the likelihood that a system may undergo RI. To ameliorate this problem, the RI index was reformulated so that the degree to which an RI predictor was determined to be conducive to the occurrence of RI was used quantitatively to assess the probability of RI. Specifically, the maximum and minimum values at which RI was observed to have occurred were employed to determine the scaled contribution of each of the 7 RI predictors. The scaled index value of each predictor was 0 when it was the least conducive for RI and 1 when it was the most conducive so that the total scaled magnitude for any given case could vary from 0 to 7. Figure 1 shows a comparison of the Brier skill (Wilks 1995) of the threshold and scaled versions of the RI index relative to the climatological probability of RI for the 1995-2003 developmental samples. The figure shows that the scaled version had slightly more skill in the Atlantic basin and significantly more skill in the E. Pacific basin. Figure 2 shows an independent evaluation of the skill of the threshold and scaled versions of the RI index obtained by re-running the 2004 season cases using the operational data collected in real-time. The figure shows that the scaled version of the RI index had 7 to 10% more skill than the threshold version in the Atlantic and E. Pacific basins, respectively. However, neither version had skill in the Atlantic basin for the 2004 season. Sensitivity tests were conducted to evaluate how the skill of the RI index varied for different intensity thresholds. For these tests, the scaled RI index was re-derived for 3 additional intensity thresholds in each basin (20,25, and 35 kt in the Atlantic and 20,25, and 30 in the E. Pacific) by employing the same 1995-2003 dependent database that was previously employed to derive the RI index. Figure 3 shows that the level of skill increases as the intensity threshold decreases and that the skill is substantially greater in the E. Pacific than it is for the Atlantic basin. An independent assessment of the level of skill of the RI index for the 4 different intensity thresholds (20,25,30,35 kt) was conducted by re-running the 2004 cases using the operational data collected in real-time during the 2004 hurricane season. Figure 4 shows that the trends observed for the dependent results depicted in Fig. 3 were also visible for the independent 2004 sample. Specifically, the RI index was more skillful in the E. Pacific basin than the Atlantic basin and the index was more skillful for lower intensity thresholds. Based upon the results shown in Fig. 4 the TPC has decided to run the version of the RI index derived using an intensity threshold of 25 kt for a 24-h period during the 2005 Atlantic and E. Pacific hurricane season. This equates to about the 90% percentile of all 24-h over-water intensity changes. FY04-05 Achievements:
Key references: Kaplan, J., and M. DeMaria, 2003: Large-scale characteristics of rapidly intensifying tropical cyclones in the North Atlantic basin. Wea. Forecasting, 18, 1093-1108. DeMaria, M., M. Manelli, L.K. Shay, J.A. Knaff, and J. Kaplan, 2005: Further improvements to the Statistical Hurricane Intensity Prediction Scheme (SHIPS), Wea. Forecasting, in Press. Wilks, D.S., 1995: Statistical Methods in the Atmospheric Sciences, Academic Press, 467pp.
Last Modified : May 20, 2005
|
Links of Interest
AOML Tools & Resources
Employee Tools
Stay Connected
|
| ||