Investigation of the Dynamics of the Forecast Errors with the THORPEX Interactive Grand Global Ensemble (TIGGE)

We employ locally linear diagnostics to investigate the ability of ensemble forecast systems to capture the forecast uncertainty. For this investigation, we use two months worth of data for all ensembles included in TIGGE. We find that the ensembles from the different forecast centers fall into two categories with respect to our diagnostics: some ensembles are tuned to capture the total variance of the local forecast uncertainty, while others are tuned to capture only that part of the variance that is associated with error patterns correctly captured by the ensemble. The best example for the first approach is the ensemble of the Canadian Meteorological Center (CMC), while the best example of the second approach is the ensemble of the European Centre for Medium-Range Weather Forecasts (ECMWF). What makes our finding particularly interesting is that the centers do not use our diagnostics when they tune their ensemble forecast systems.