Role of stochastic forcing in ENSO in observations and a coupled GCM

A procedure is presented to estimate the role of atmospheric stochastic forcing (SF) in El Nino–Southern Oscillation (ENSO) simulated by a coupled ocean–atmosphere general circulation model (CGCM), in direct comparison to observations represented by a global reanalysis product. SF is extracted from the CGCM and reanalysis as surface wind anomalies linearly independent of the sea-surface temperature anomalies. Madden–Julian Oscillation (MJO) is isolated from SF to quantify its role in ENSO. A coupled ocean–atmosphere model of intermediate complexity is forced with SF, as well as its MJO and non-MJO components, from the reanalysis and CGCM. The role of SF is estimated by comparing the original ENSO in observations and the CGCM with that reproduced by the intermediate model. ENSO statistics in both reanalysis and CGCM are better reproduced when the intermediate model is tuned to be weakly stable than unstable. The intermediate model driven by SF from the reanalysis reproduces most characteristics of observed ENSO, such as its spectrum, seasonal phase-locking, fast decorrelation of ENSO SST during boreal spring, and its lag-correlation with SF. In contrast, not all characteristics of ENSO in the CGCM are reproduced by the intermediate model when SF from the CGCM is used. The seasonal phase-locking of ENSO in the CGCM is not reproduced at all. ENSO, therefore, appears to be driven by SF to a lesser degree in the CGCM than in observations. Characteristics of observed ENSO reproduced by the intermediate model (driven by SF) can be largely attributed to the MJO; which, for instance, is responsible for the fast decorrelation of ENSO SST during boreal spring in both reanalysis and CGCM. The non-MJO component seems to be more responsible than the MJO for erroneous features of ENSO in the CGCM.