Thursday, 13 January 2005
A priori identification of the inferior solutions in an ensemble of dynamical forecasts
The skill level of week 2 (day 8 to 14 mean) forecasts and physical understand and methods to raise their skill level are investigated using the NCEP's daily operational ensemble forecasts. The anomaly correlation (AC) skill scores over the North Pacific and North American sector are examined. Occasionally, an official week 2 forecast can turn out to be an outlier, with validated AC score less than -0.5 for instance, while the average skill score for a season or longer is above +0.5. Catching an outlier at forecast time, such as the one just mentioned, is a challenging task. Three objective methods to identify the inferior solutions among members of an ensemble run are introduced here. They are: 1. The internal persistence of an individual perturbation run. There is a similarity between AC skill scores and time series of persistency from Day (5-11) to Day (8-14). Therefore, a low persistency can be used to catch an inferior solution. 2. The persistence between current week 2 forecast and the one made one day earlier. A drastic change is found likely to be an inferior forecast. 3. Similar to the above method except for resorting now to persistence between current forecast and the one made two days earlier. A test of this method suggests also its effectiveness in identifying an inferior solution. The above methods were tested on the NCEP's operational ensemble runs for DJFM extended seasons from 2000/2001 to 2003/2004. Consistently, they show promising results. A disastrous week 2 forecast can likely be avoided by an application of the above objective methods.