ENSO simulated by intermediate coupled models and evaluated with observations over 1970-98. Part I: Role of the off-equatorial variability

ENSO simulations are investigated in 30-yr integrations of various intermed iate coupled models and compared with observed SST, wind, and thermocline d epth anomalies over the tropical Pacific. The Cane and Zebiak model simulat es warm events with a period close to the observations, but with westerlies that are located 30 degrees east of them and thermocline anomalies in the western Pacific that are much shallower Between two warm events, the model simulates a series of three weak and short cold SST peaks and hardly ever s imulates easterlies. The SST in the eastern equatorial Pacific is not sensi tive to thermocline depth anomalies, but to the anomalous downwelling of su rface currents induced by Ekman shear. The model simulates a pair of very s trong cyclonic wind stress curl anomalies on both sides of the equator in t he eastern off-equatorial domain between 7 degrees and 15 degrees lat. Thes e are necessary to maintain the oscillatory regime-so are the ocean meridio nal Rossby modes higher than 5. The thermocline zonal slopes required to ba lance the off-equatorial curl anomalies are about three times steeper than the ones required to balance the zonal stress along the equator. Thus the o ff-equator exerts an excess of zonal pressure, which by continuity affects the equatorial ocean and plays a crucial role in reversing and triggering t he growing events. Six months after the warm peaks, the whole ocean between 15 degrees S and 15 degrees N is significantly upwelled. The equatorial oc eanic heat content is recharged from the south prior to a warm event. Contrary to simulations when the model is driven by observed wind anomalies , increasing the friction in the baroclinic ocean does not decrease the off -equatorial variability but significantly alters the low-frequency oscillat ions that are no longer ENSO-like. Introducing the parameterization of subs urface temperature derived from hydrographic profiles in the ocean componen t does not allow the coupled model to recover cold events as in a forced co ntext, introducing the parameterization of convection derived from high-clo ud temperature measurements is the most effective improvement, but results still poorly agree with observations and are in contrast with the simulatio ns driven by observed SST, biased toward westerlies in the central Pacific, upwelled thermocline in the west, and warm SST in the east. Thus modifying the ocean component only or the atmosphere only does not have the same imp act on simulations as in a forced context. The coupling allows new mechanis ms to grow and govern the model behavior. One of them is the slow meridiona l oceanic mass adjusment in quasi-Sverdrup balance with the winds.