An environment-dependent probabilistic tropical cyclone model

A new probabilistic tropical cyclone (TC) model is developed to simulate tropical cyclone climatology and assess TC risk for the North Atlantic Basin. The model consists of three components—a hierarchical genesis model, a data-driven track model, and a Markov intensity model—and describes storm activity under a given climate environment. To build the hierarchical genesis model, we first conduct k-means clustering according to local environmental conditions so that spatial-temporal grid cells (e.g., 2deg*2deg*5days) that have similar environment conditions are assigned to be in the same cluster. The cluster centroids are then related to the expected number of genesis events of each cluster through a Poisson regression. As TCs' movement is mostly driven by steering background winds, we develop a data-driven track model based on historical TC track records in the whole basin, acknowledging that TCs sharing a similar trajectory would likely continue to develop in a similar pattern. Different from most wind-driven models, here we determine the storm's movement by both properties of selected historical similar tracks (determining the main movement direction) and the background wind (added as a random correction to the main direction). For the intensity along the track, we adopt a newly developed Markov environment-dependent hurricane intensity model (MeHiM) and add to it a model for simulating the initial condition. All three model components (i.e., genesis, track and intensity) are dependent on environmental variables including the potential intensity, environmental wind, midlevel relative humidity, and ocean mixing characteristics.

The three components are firstly evaluated individually based on observed TC activities. Then, we integrate all three parts as a whole risk model and simulate 1979-2014 TC climatology. Various aspects of TC statistics are used to evaluate the risk model, such as the annual genesis rate, spatial genesis distribution, and their interannual variability; TC track density; TC intensity density; and the distribution of landfall location and landfall intensity. This TC risk model will be further coupled with TC hazard models to quantify wind, surge, and rainfall hazards and probabilities under various climate conditions.