A better understanding of the El Niño-Southern Oscillation (ENSO) mechanisms in the coupled ocean-atmosphere system is required before we can make a reliable estimate of its predictability. Previous studies have shown that the structure of the tropical Pacific mean climatology, i.e., mean state, has a significant potential impact on the characteristics of the ENSO cycle. In this study, such impact is investigated quantitatively through a set of numerical experiments using the Climate Community System Model version3 (CCSM3), one of the world's leading general circulation climate models developed by the National Center for Atmospheric Research (NCAR). In a sensitivity experiment, an empirical time-independent heat flux correction over the tropical ocean is applied to the oceanic component of CCSM3. In comparison with the fully coupled control run, the annual mean SST and precipitation of the sensitivity run are more realistic: the cold biases in the central equatorial Pacific and warm biases the South America coast are significantly reduced. The double ITCZ problem in the control run is also eased. A major benefit of the improved annual mean state is a more realistic annual cycle of the SST and precipitation over the equatorial eastern Pacific and Nino3.4 regions (5°S-5°N, 170°W-120°W), which is associated with a reduction of the unrealistic semi-annual signals in the control run. In addition, our results demonstrate that the model ENSO cycle is sensitive to these modifications of the mean state: Instead of a strong and regular biennial oscillation in the control run, ENSO variability is less regular and with longer period (but shorter than observed 4-year period) in the heat flux corrected simulation, which is more realistic in comparison with observations. However, the ENSO events in the flux corrected run has weaker amplitude and do not show a significant phase-locking with season as the observation and the control run do.

In order to examine the impact of the mean state on the ENSO prediction, a serial of seasonal hindcasts with 12-month integration is performed using the coupled system with and without the heat flux correction. The initial conditions of the hindcasts are for the 1st of January and July of each year from 1982 to 1998. An ensemble of 3 hindcasts is generated with the same ocean initial condition but perturbed initial conditions for other components (i.e., atmosphere, land and sea ice). Based on the measurement of the Anomaly correlation coefficient (ACC) and Root Mean Square Error (RMSE), the predictive skills of the SST anomaly area-averaged in Nino3.4 region from the control and flux corrected forecasts are comparable in the first 6 months lead time in the January hindcasts and the first 8 months lead time in the July hindcasts. The flux corrected forecast shows slightly higher prediction skill in 7-9 lead months for the January and 9-11 lead months in the July hindcasts. However, their difference is not statistically significant at the 90% level.