Symposium on Observations, Data Assimilation, and Probabilistic Prediction

6.7

Spread-skill relationship in the Canadian ensemble prediction system

Richard Verret, MSC, Dorval, PQ, Canada; and F. Pithois, L. Lefaivre, P. Houtekamer, L. Wilson, G. Pellerin, and M. Klasa

The usefulness of an Ensemble Prediction System resides mostly in the variety of possible solutions that the system can offer to a given meteorological forecast problem. The differences between all weather scenarios presented by each member of the Ensemble prediction System, or the variance within the ensemble forecasts, lead toward the study of the spread-skill relationship. If such a relationship exists, it would then be possible to associate higher skill and better confidence in the forecasts when the ensemble variance is low, and vice versa, forecasts with lower skill and a lower confidence should be expected when the ensemble variance is large. The outcomes of such a spread-skill relationship, amongst others, include the possibility of forecasting the forecast skill on one hand and to use the spread of the ensemble as an indication of the confidence on the deterministic forecast on the other hand.

The Perfect Prog statistical adaptation system operational at the Canadian Meteorological Center has been run on each of the sixteen members of CMC Ensemble Prediction System. The ensemble variance of the statistical 12-h probability of precipitation forecasts is being evaluated as a proxi for a confidence index. The 12-h probability of precipitation forecasts generated from each member of the ensemble have been verified at all projection times to ten days, at 264 Canadian stations over the period extending from June 2000 to December 2000 inclusive. The skill of the ensemble average forecasts converges toward that of climatology by 180 hour projection time. This is an indication that skill with respect to climatology can be expected up to 7 days. Contingency tables of Brier scores of the 12-h probability of precipitation forecasts of the control model versus the ensemble variance have been constructed to study the spread-skill relationship. The results indicate that it is possible to use the ensemble variance as a proxi for a confidence index. Based on the chi-squared test, the spread-skill relationship is statistically significant to 240 hour projection time.

A brief description of the CMC Ensemble Prediction System will be presented. The spread-skill relationship applied to the Perfect Prog statistical 12-h probability of precipitation will be demonstrated and results of the cross-validation of the confidence index in forecast mode will be presented.

extended abstract  Extended Abstract (72K)

Session 6, Ensembles and data assimilation
Thursday, 17 January 2002, 8:45 AM-1:30 PM

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page