The level of variability present in operational model simulations of marine cyclogenesis was examined. Successive forecasts valid for the same 12-h time period of analysed maximum cyclone central pressure fall from the National Meteorological Center (NMC) nested grid model (NGM) and the Canadian Meteorological Centre (CMC) regional finite element model (RFE) were examined for one cold season (November 1988–March 1989). All cyclones with an analysed 12-h maximum central pressure fall occurring within the Western Atlantic region were included in the study, comprising a total of 52 sets of 0–12-,12–24-, 24–36-, and 36–48-h forecasts.
Analyzed cyclone development spanned a wide range of intensities from no development (one case) to extraordinarily rapid development (greater than 24 mb in 12 h, 14 cases). The primary storm track was located offshore, resulting in reduced precipitation along the coastal regions. Both models tended to under-forecast cyclone development, and under-represent the clustering of cyclone activity within a narrow band offshore. This tendency became progressively stronger in the NGM with forecast range, while the behavior of the RFE was complicated by an apparent short-range (0–12-h) spin-up problem.
The degree of variability in a sequence of forecasts was not well related to the overall prediction error for either model, belying the view that consistency in successive model forecasts indicates reliability. Each model exhibited forecast of low variance with high error (consistently poor forecasts of development) and sets of high variance with low error (inconsistent, but within range of the analyzed development). The models performed more similarly with regard to their mean error characteristics for a given set of forecasts than in terms of the dispersion of that error within the forecast sequences.