For the three reforecast ensembles, it is not surprising to note that the optimal bias-corrected ensemble (RFC_OPT) shows the best performance when compared with the raw and climatological bias-corrected ensembles (RFC_RAW and RFC_COR).
A comparison between RFC_RAW and RFC_COR shows that RFC_COR has a noticeable RPSS improvement versus RFC_RAW, especially for the week-2 forecasts.
Using the climatological mean bias estimate, it is possible to make probabilistic week-2 forecasts more skillful than the raw reforecast. The RFC_COR comes from a reforecast ensemble with an older version of the model, which has initial data of relatively poorer quality compared with the operational ensembles, including OPR_RAW and OPR_DAV2%.
However, RFC_COR has an even better level of performance than OPR_RAW FIG. 7. RPSS of Northern Hemisphere 500 hPa geopotential height from 1 Mar 2004 to 28 Feb 2005 comparing the NCEP operational forecast (OPR) and ESRL reforecast (RFC): OPR_RAW is the NCEP operational raw ensemble forecast, OPR_OPT is the NCEP operational forecast using optimal bias correction, OPR_DAV2% is the NCEP operational forecast using a 2% weight for bias correction, RFC_RAW is the raw reforecast, RFC_OPT is the reforecast after optimal bias correction, and RFC_COR is the reforecast after removing the climatological mean bias.
406 WEATHER AND FORECASTING VOLUME 27 and OPR_DAV2% after day 10 (Fig. 7), indicating the effectiveness of a large data sample in improving week-2 forecasts.