The role of stochastic forcing (SF) in the El Niño – Southern Oscillation (ENSO) variability in a coupled general circulation model (CGCM) is examined. The CGCM used is being employed for ENSO prediction by Bureau of Meteorology Research Center (BMRC), Australia. Using CGCM data, empirical orthogonal functions (EOFs) of tropical oceanic and atmospheric surface variables are compared to the singular vectors of covariance matrix between these variables. The analysis suggests that while a large percentage of anomalous ocean surface variability is linearly coupled to the atmosphere, a significant amount of variance in the atmosphere is uncorrelated to any contemporaneous changes in sea surface temperature anomalies (SSTA). This atmospheric component may act as SF on the coupled ocean-atmosphere system. A statistical model of surface zonal wind anomalies regressed with the SSTA is constructed. Two versions are constructed – one using data from 163 years run of the CGCM, and the other using last 29 years of data from NCEP-Department of Energy (DOE) - II reanalysis (NCEP-2). The residual surface zonal wind anomalies unpredicted by the statistical model are diagnosed for spatial and temporal characteristics, in order to be validated as SF.

A coupled ocean-atmosphere model of intermediate complexity (variant of Zebiak-Cane model) is then driven by the derived SF. Experiments are conducted using various values of the ocean-atmosphere coupling strength in the model. The resulting model ENSO is diagnosed for a broad range of characteristics. The diagnoses include statistical and spectral analysis, evolution of warm events, and seasonal variance locking. The results are compared with ENSO variability in the parent dataset of SF. An attempt is made to quantify the role of stochastic forcing in ENSO variability. Analysis suggests that SF plays a lesser role in the ENSO variability in the CGCM, than in the reanalysis. Further, the seasonal locking of warm events in stochastically forced model is compared to that in the parent dataset of SF. It is found that the deviations between the two are more in CGCM than in reanalysis. Experiments involving synthetic SF – composed of both CGCM and reanalysis stochastic components – reveal that these deviations can be traced to the unrealistic seasonal variability of Madden Julian Oscillation (MJO) in the CGCM.