The mean state dependence of ENSO atmospheric feedbacks and ENSO dynamics in climate models

We investigate the dependence of ENSO non-linearity on the mean state in a perturbed atmospheric physics ensemble with the Kiel Climate Model (KCM) and in CMIP5 models. Additionally, uncoupled simulations are conducted with the atmospheric component of the KCM to obtain further insight into the mean state dependence. It is found that the spatial ENSO non-linearity, i.e. that El Nino is located further east than La Nina, is underestimated in many state-of-the-art climate models1⁠. This can be explained with the underestimated strengths of the positive zonal wind feedback µ and the negative heat flux feedback α, which are strongly linearly related through sea surface temperature (SST), while at least in the KCM differences in model physics seem to be less important2⁠. In observations, strong zonal wind and heat flux feedbacks are caused by a convective response in the western central equatorial Pacific (Nino4 region), resulting from an eastward (westward) shift of the rising branch of the Walker Circulation (WC) during El Nino (La Nina). Climate models with a La Nina-like mean state, i.e. an equatorial SST cold bias in the Nino4 region, which is a common problem in many state-of-the-art climate models, simulate a too westward located rising branch of the WC (by up to 30°) and only a weak convective response. Thus, the position of the WC determines the strength of both the wind and heat flux feedback, which also explains why biases in these two feedbacks partly compensate in many climate models. Furthermore, a too eastward position of the WC leads to a fundamental change in ENSO dynamics, as ocean-atmosphere coupling shifts from a predominantly wind-driven to a more solar radiation-driven mode (Fig. 1a). On the other hand enhanced atmospheric feedbacks lead to a substantial improvement of the non-linearity of ENSO (Fig. 1b). Differences in the mean state SST are suggested to be a major source of ENSO diversity in current climate models.