A Study from NCEP Global Ensemble Forecast System Extend Forecast

In search of a suitable ensemble prediction system at sub-seasonal time scales, a series of 35-day global ensemble forecast experiments with the most recent version of the GEFS (version 11.0.0) are carried out. Three identical configurations of the GEFS with distinct treatment of the SSTs are compared. The first configuration, referred to as the benchmark consists of keeping the default configuration, which damps the initial SST anomaly to climatology. In the second configuration, akin to AMIP runs, the global SST analysis fields are prescribed every 24-hour as the integration of the prediction system progresses. The third configuration uses predicted SST anomalies from the CFSv2, which is a coupled model with a low resolution GFS model as its atmospheric component.

One critical aspect sought in the comparison of configurations is to determine the level at which MJO signal is captured and how the signal is propagated through the extra-tropics via tele-connections. In the benchmark experiments, a strong relationship between the prediction skill of extra-tropical flow and the MJO signal is found, with high skills during periods of strong MJO signal. This suggests that the atmospheric model adequately responds to forcing in the tropical Pacific. Prediction skills of main variables near the surface tend to decay fast beyond one week but upper level variables have useful skills beyond two weeks. Results of the second and third configuration indicate small changes in the flow of the extra-tropics but the bias of low-level and surface tropical variables is reduced. This talk will present diagnostics and skill assessments of the comparison.