Skill of Operational Dynamical Models in Cyclone Prediction Out to Five-Days Range during ERICA

Abstract

Investigating the skill of prediction of surface cyclones by operational models to ranges of five days, we studied the central and western North Atlantic region for the December 1988 through February 1989 period of the Experiment on Rapidly Intensifying Cyclones over the Atlantic (ERICA). Output was evaluated from the medium-range forecast (MRF) model of the National Meteorological Center (NMC) and (as available) from the models of the United Kingdom Meteorological Office and the European Centre for Medium-Range Weather Forecasting. Operational manual NMC analyses were used for verification. For the MRF model, the correlation coefficient between predicted and analyzed 24-h deepening or filling was 0.91 for day 0?1, decreasing to 0.57 for day 4?5. There was little bias. The other models yielded lower values at a given range, with an underestimate of deepening. Explosive deepening was identified by all three models at all ranges with at least small skill, on the basis of the Conditional Success Index. Scores for the MRF model were higher for a given range than for the other models. Position error for the MRF model was about 150 km in the initial analysis. In the forecast it grew linearly at the rate of about 125 km per day. Errors in the other models were slightly larger at a given range. The MRF model performed decidedly worse in the eastern Pacific region than in the Atlantic area during the same period. At 500 mb at 40°N, 65°W, the day-to-day height changes were somewhat better predicted by the MRF model than by the others, but different results were found at other selected points. The MRF guidance was used with considerable success by the ERICA forecasters in making operational decisions.