Medium-range forecasts

The design of ensemble methods for medium-range weather forecasting ten years ago was aiming mainly at addressing the problem of limited predictability of supra-synoptic weather regimes within the 6–10 days range. From this point of view, it has been shown since then that the ensembles have delivered improved forecasts compared with a purely deterministic approach, both improving single-value estimates by removing stochastic errors (ensemble mean) and providing reliable and still sharp estimates of the probability distributions of large scale flow patterns such as blocking (Chessa and Lalaurette, 2001; Pelly and Hoskins, 2003).

Severe weather forecasts seem, however, to be clearly beyond reach of such medium-range, global and therefore relatively coarse grid models. Indeed the experience in trying to use medium-range forecasting systems to be alerted of the risk of severe weather more than one day in advance is very limited. On most occasions, civil security services are alerted not earlier than the day before the event, while public warnings are only issued on the same day – tropical cyclones being a notable exception to this common rule. The reasons usually given by forecasters as to why they do not use numerical forecasts in the early medium range are mainly twofold:

1. The global numerical models are generating nothing looking remotely like severe weather.

2. If one takes signatures from the global models that are associated with severe weather, there is no consistency in the forecast from one day to the next: the rate of false alarms would be far too high to be considered.