A Markov environment-dependent hurricane intensity model (MeHiM) is developed to simulate the climatology of hurricane intensity given the surrounding large-scale environment. The model considers three unobserved discrete states representing respectively storm’s slow, moderate, and rapid intensity change. Each state is associated with a probability distribution of intensity change. The movement of intensity change from one state to another, regarded as a Markov chain, is described by a transition probability matrix. The initial state is estimated with a Bayesian approach. All three model components (initial intensity, state transition, and intensity change) are dependent on environmental variables, including potential intensity, vertical wind shear, and midlevel relative humidity. This dependent Markov model shows a significant improvement over previous statistical models (e.g., linear, nonlinear, and finite mixture models) in capturing the probability distributions of 6-h and 24-h intensity change, lifetime maximum intensity, and landfall intensity, etc.
Although improved over previous models in estimating the climatology of hurricane intensity, MeHiM cannot always capture the state of extreme intensity change, due to the uncertainty in the (relatively simple) transition model. MeHiM is thus limited for real-time forecasting, as it captures only partially the storm’s significant strengthening, or rapid intensification (RI), Thus, we define the Extreme State Index (ESI) as an additional guidance, which suggests the transition to the extreme state of the intensity change. The ESI is firstly built on the same environmental parameters as MeHiM but with more sophisticated statistical methods including support vector regression and regression trees. Then a revised ESI is developed based on additional variables such as ocean heat content and predictors that are derived from satellite data. Our preliminary results show that, the combined MeHiM-ESI model is able to simulate the intensity evolution of most historical RI storms, indicating its great potential for real-time forecasting applications.