It was found, as expected,  that the change in resolution had a substantial impact on the quality of the ensemble forecasts. The rms error of the ensemble control and mean forecasts is reduced (Figs. 3 and 4).As a result, the ensemble mean forecast now has scores that is equal to or better than those for the T170 high resolution MRF control forecast at all lead times. The gap between the ensemble mean error and ensemble spread, which should ideally be equal, is also reduced (Fig. 5). Moreover, the cloud of ensemble forecasts misses the verification only 6% of the time, compared to a 14% excessive missing rate (above that expected due to the finite size of the ensemble) at 60 hrs lead time (Fig. 6). In addition, the T170 high resolution control MRF forecast is the best member of the 23-member ensemble only 5.1% of the time, only 19% more often than expected if all members were equally likely (which correspondes to a 4.3% best verification score, Fig. 7). This is to be compared to 6.5% rate before the ensemble resolution increase, which is 51% above the 4.3% expected best verification rate (Fig. 8). 

With the implemented changes the ensemble forecasts provide improved forecast guidance twice a day. Regarding the possibility of the combined use of the latest set of ensemble forecasts and the 12-hour older set of ensemble, ensemble mean pattern anomaly correlation, rms (Fig. 9), and probabilistic verification scores (Fig. 10) indicate that the inclusion of the 12-hour older members degrades the quality of the ensemble until 132-168 hours lead time is reached. Apparently the disadvantages resulting from the inclusion of less skilful older members outweighs the advantages of having more members until 6-7 days lead time or so. Therefore it is recommended that for the first week only the latest set of ensemble members are used in forecast applications.