Investigating the Variability in Skill of Statistical RI Prediction Models

Despite some recent improvements in tropical cyclone (TC) intensity forecasting skill, predicting changes in TC intensity remains problematic. In particular, the forecasting of episodes of rapid intensification (RI) has proven to be especially difficult prompting the National Hurricane Center (NHC) to declare it as their highest operational forecasting priority. In recent years, a statistical rapid intensification index (SHIPS-RII) that employs environmental data from the Statistical Hurricane Intensity Prediction Scheme (SHIPS) to estimate the probability of RI has been developed based upon linear discriminant analysis. Although the original version of the SHIPS-RI that is currently employed as an operational forecasting tool by the NHC in both the Atlantic and eastern Pacific basins was derived for a lead time of 24 h, additional versions have also since been developed for lead times of 12-h, 36-h, and 48-h. While recent results indicate that the aforementioned RI models appear to exhibit some skill in identifying episodes of RI, the overall level of skill of those models can be somewhat limited (particularly in the Atlantic basin). Moreover, significant case-to-case variability in RI model skill is sometimes observed. Thus, in an effort to better understand when the RI models are likely to be the most (least) skillful an analysis of the surrounding environmental conditions and initial tropical cyclone structure is performed to investigate whether differences in either of these factors appear to contribute to the observed variability in predictive skill of the RI models.