Tropical cyclone modeling in a probabilistic framework

It can be argued that the most important aspects of the tropical cyclone life cycle (genesis and rapid intensity change) are also the most poorly understood and the least susceptible to prediction. Added to the problem are the poorly understood error characteristics of the observations intended to improve model initializations via data assimilation as well as the model errors themselves. These inherent uncertainties suggest that the modeling problem ought naturally to be cast in a probabilistic frameowrk. Among the data assimilation methods currently available, the ensemble Kalman filter (EnKF) is best suited to this task.

Despite the degree of its simplifying sumptions, namely that the state probability density function (PDF) is Gaussian, the EnKF enjoys the advantage of flow-dependent error statistics. In the highly nonlinear environment of a developing or intensifying tropical cyclone, where scale interaction processes are of paramount importance, this is a decided advantage. Indeed, it is shown that an EnKF with relatively small ensemble size is not only capable of capturing key aspects of tropical cyclone evolution (i.e. accurately estimating the PDF mean), it is also capable of returning accurate forecast error estimates (i.e. PDF covariance). While these results are encouraging, it should be stressed that, to be widely applicable, much work needs to be done on key aspects of filter design such as ensemble initialization and both model and observation error modeling.