Statistical modelling of North Atlantic tropical cyclone tracks

We present a statistical model of North Atlantic tropical cyclone tracks from genesis site through lysis. To propagate tracks we use the means and variances of latitudinal and longitudinal displacements and model the remaining anomalies as autoregressive. Coefficients are determined by averaging near-neighbour historical track data, with ‘near’ determined optimally by using jackknife out-of-sample validation to maximize the likelihood of the observations. The number of cyclones in a simulated year is sampled randomly from the historical record, and the cyclone genesis sites are simulated with a spatial probability density function using kernels with optimized bandwidths. Simulated cyclones suffer lysis with a probability again determined from optimal averaging of historical lysis rates. We evaluate the track model by comparing an ensemble of 1950–2003 simulations to the historical record using several diagnostics, including landfall rates. In most regions, but not all, the observations fall within the variability across the ensemble members, indicating that the simulations and observations are statistically indistinguishable. An intensity component to the TC model, necessary for risk assessment applications, is currently under development