Abstract

AbstractLagged‐average forecast experiments, produced with the Meteorological Office 11‐level general circulation model, are used to assess the benefits arising from the use of ensemble forecasts in practical monthly prediction.As measured by anomaly correlation, forecast skill is consistently positive at days 6‐15, but very small on average beyond 20 days. For the monthly mean forecast, removal of an independent estimate of the model systematic error improves the average score significantly. Compared to the mean score amongst individual members, a modest increase in skill is achieved by ensemble averaging. Theoretically, however, there remains scope for a much greater impact in the future, through the use of models capable of more accurate forecasts.Local forecast skill is correlated with a measure of the agreement between ensemble members, although the overall relationship is weak beyond the medium range. Nevertheless, for areas of high agreement, the probability of obtaining useful skill is considerably higher than average at days 6‐15 and 11‐20.The experiments are also assessed as probability forecasts, using the Ranked Probability Score to measure skill. the ensemble probability forecast (EPF) is created by averaging the probability forecasts of the individual members. Relative to the average score for the latter, the EPF gives improved skill. For the medium‐range period of the forecast (days 1‐10), the use of an optimized EPF, calculated by weighting each ensemble member according to its age, leads to a small increase in skill compared to the individual forecast from the most recent analysis. This effect is not observed in the corresponding anomaly correlation results. the above results for local skill prediction are confirmed, and somewhat enhanced, in the probabilistic formulation, using the ensemble variance of forecast categories as the predictor. A simple method of issuing confidence‐based adjustments to the forecast probabilities is discussed.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call