An Improved Method for Generating Ensembles of Tropical Cyclone Tracks

Risk management requires probabilistic prediction, and hence ensemble methods are of particular importance for high impact weather. For tropical cyclones, The Bureau of Meteorology currently uses the DeMaria et al.'s (2009) method for generating an ensemble of cyclone tracks, together with intensity and size data, which is then used to calculate wind exceedance probabilities and other useful statistics for the forecasts.

It would be desirable to use these tracks for other purposes, such as driving an ensemble of storm surge or ocean wave models. However, the current system has some limitations that make such an application difficult: (i) the tracks are not particularly smooth; (ii) the statistical model is discrete rather than continuous in time, so an arbitrary time-interpolation is needed; and (iii) a large number of empirical parameters are needed to define the model, so a correspondingly large training data set is needed for statistical stability.

In this poster, we present a prototype improved method that overcomes these issues, while not altering those statistical properties of the existing method that are used operationally. In addition, the (much fewer) parameters in the new model are readily estimated from NWP ensemble prediction systems, facilitating the coupling of these to the Monte Carlo track generator.