Of the three reforecast ensembles, RFC_COR has smaller RMS values than does RFC_RAW for all lead times, even for week 2.
A comparison between the operational and reforecast ensembles shows that the operational ensemble mean (OPR_RAW) has a much lower RMS error than the ESRL/PSD hindcast (RFC_RAW).
The RFC_RAW short-range error is around 50% larger than that of OPR_RAW. Though the reforecast runs start from relatively poorer quality initial data than are used in the operational ensemble, RFC_COR works for short-range forecasts and its curve with reduced RMS error comes close to the OPR_RAW curve after day 10.
Both Figs. 7 and 8 show that the decaying averaging with a 2% weight and 45 days of training data works very well in the short range.
All measures are improved until day 5.