Evaluating Atlantic tropical cyclone track error distributions for use in probabilistic forecasts of wind distribution

Abstract

This thesis investigates whether the National Hurricane Center (NHC) operational product for producing probabilistic forecasts of tropical cyclone (TC) wind distributions could be further improved by examining the distributions of track errors it draws upon to calculate probabilities. The track spread/skill relationship for several global ensemble prediction system forecasts is examined as a condition for a description of a full probability distribution function. The 2007, 2008, and 2009 NHC official track forecasts are compared to the ensemble prediction system model along-, cross-, and forecast-track errors. Significant differences in statistical properties were then identified among the groups to determine whether conditioning based on geographic location was warranted. Examination of each regional distribution interval suggests that differences in distributions existed for along-track and cross-track errors. Because errors for ensemble mean and deterministic forecasts typically have larger mean errors and larger variance than official forecast errors, it is unlikely that independent error distributions based on these models would refine the PDFs used in the probabilistic model. However, this should be tested with a sensitivity analysis and verified with the probability swath. Overall, conditional formatting suggests that the NHC probability product may be improved if the Monte Carlo (MC) model would draw from refined distributions of track errors based on TC location.