On 24 August 2004, The National Centers for Environmental Prediction implemented a new Coupled Forecast System (CFS) into operation. The system is comprised of an atmospheric data assimilation system, an oceanic data assimilation system and an ocean-atmosphere coupled model utilizing direct coupling. The CFS is run daily with a seven-day time lag to allow for adequate oceanic observations to initialize the model. This will provide an ensemble of about 30 forecasts for the NCEP Climate Prediction Center (CPC) each month when CPC issues its monthly and seasonal predictions.

The CFS model uses a version of the NCEP Global Forecast System model that was operational during 2003 for the atmosphere and the GFDL Modular Ocean Model version III (MOM3) for the ocean model. The coupling strategy used is a direct coupling with no corrections. An ensemble of 15 retrospective forecasts has been completed for the past 23 years (1981 - 2003) to provide for calibration of the real-time forecasts. Skill masks have also been generated using the retrospective forecasts.

In extended CMIP runs of over 30 years, the CFS model showed very little bias in the SST simulation and practically no drift. It also generated realistic ENSO signals in the equatorial tropics as well as strong MJO signals. While the ENSO simulations are very realistic, the MJO simulations showed room for improvement. The ability of the model to predict the equatorial SST in the seasonal time frame was compared with the older version of the NCEP coupled model (CMP14) using the retrospective forecasts and the results clearly indicated that the new CFS model is superior in its ability to predict the equatorial SST anomalies in the 3-9 month prediction range. Details of the retrospective forecast behaviors and verification statistics will be shown.