Global assessment of tropical cyclone intensity statistical-dynamical hindcasts
Résumé
This paper assesses the characteristics of linear statistical models developed for tropical cyclone (TC) intensity prediction at global scale. To that end, multilinear regression models are developed separately for each TC-prone basin to estimate the intensification of a TC given its initial characteristics and environmental parameters along its track. We use identical large-scale environmental parameters in all basins, derived from a 1979–2012 reanalysis product. The resulting models display comparable skill to previously described similar hindcast schemes. Although the resulting mean absolute errors are rather similar in all basins, the models beat persistence by 20–40% in most basins, except in the North Atlantic and northern Indian Ocean, where the skill gain is weaker (10–25%). A large fraction (60–80%) of the skill gain arises from the TC characteristics (intensity and its rate of change) at the beginning of the forecast. Vertical shear followed by the maximum potential intensity are the environmental parameters that yield most skill globally, but with individual contributions that strongly depend on the basin. Hindcast models built from environmental predictors calculated from their seasonal climatology perform almost as well as using real-time values. This has the potential to considerably simplify the implementation of operational forecasts in such models. Finally, these models perform poorly to predict intensity changes for Category 2 and weaker TCs, while they are 2–4 times more skilful for the strongest TCs (Category 3 and above). This suggests that these linear models do not properly capture the processes controlling the early stages of TC intensification.