Systematic errors in the behaviour of cyclones in the ECMWF operational models

Systematic errors in the forecasts (up to 10 days) of cyclones over the north Atlantic and Europe by the ECMWF operational grid point model (N48) for the 1981–82 winter seasons are described, and some corresponding results for the ECMWF operational spectral model (T63) for the 1983–84 winter season are shown. Statistics for the cyclone tracks, central pressure, deepening rates, etc., are given.

The forecasts from the grid point model tend to displace the cyclone tracks towards the south, especially after day 4, except near Newfoundland where a northward displacement is found. The corresponding errors are reduced in the spectral model. A systematic phase error is evident in the grid point model with the waves being too slow, usually in the developing stages of cyclones. This error is also apparent for the fast moving cyclones in the spectral model. For the grid point model, the forecast deepening and filling rates are less than those observed. However, the deepening rate is more realistic in the spectral model, though the filling rate is still underestimated. For both models the horizontal scale of the cyclones is too small during the first 2 or 3 days, especially in the forecasts started from several days prior to the day on which the cyclone was generated; during the later stages of cyclone development, the scale is too large. In comparing these errors to similar errors documented for other numerical models, some hypotheses are discussed about their likely causes.