Exploring El Niño mechanisms in climate models

Increasing our understanding of the equatorial Pacific climate phenomenon El Niño – Southern Oscillation (ENSO) is important for making progress in climate prediction and establishing the effects of Global Change on ENSO. The GCMs that are used for IPCC scenarios all show ENSO-like behavior, but the resemblance to the observed ENSO varies from model to model. We wonder whether we can analyse how these differences arise. Which mechanisms are important for ENSO? Are they modelled correctly in climate models? Are models good enough to describe the asymmetry between El Niño and La Niña? Can we say something about El Niño in a future climate?

We trace the mechanisms governing El Niño in observations and in a set of 19 coupled global climate models (GCMs). To this end, we use a conceptual model that describes the relations between subsystems that play a role in ENSO using local linear regressions. The first relation describes a wind response to SST variability. The second and third relation describe the response of SST to both thermocline variability and wind stress variability. finally, a temporal damping on SST is described. We categorize six GCMs as having the most realistic balance of the various feedback mechanisms compared to observations. In four of these models the interannual mode also resembles the observed ENSO both spatially and temporally. In the other 13 models at least one part in the feedback loop between the ocean and atmosphere behaves differently as in observations. We thus selected a subset of best models based on the mechanisms that are important to describe El Niño.

For a subset of models as defined above we make projections into a future stable, warmer climate. Although there are large changes in the mean state, the overall ENSO properties do not change much. This is due to the fact that the effects of the changes in the different ENSO relations tend to cancel. In all models, the signs of the changes in ENSO mechanisms are similar. However, the sign of the small net effect differs from model to model.

A description in terms of linear couplings leads to symmetry between El Niño and La Niña events. In general, however, El Niño is larger than La Niña. In other words, the distribution of SST anomalies in the East Pacific is positively skewed as a result of nonlinear interactions in the system. Can we use nonlinearities in our conceptual model for analysing what is the likely origin of the skewness of ENSO? What is the role of atmospheric noise in this respect?

For observations the linear couplings are extended with a new description of atmospheric noise properties and some nonlinearities in the atmospheric terms. The effect of these nonlinearities are studied with an Intermediate Complexity Model (ICM) in which the fitted couplings and noise properties are implemented, but no further tuning is carried out. The description of atmospheric noise properties in terms of standard deviation and spatial and temporal correlation is sufficient for the excitation of ENSO in this ICM. The ENSO period and pattern of the ICM agree reasonably well with that found in observations. The skewness of SST anomalies has been evaluated after adding a nonlinearity in the response of the wind stress to SST anomalies, the state-dependence of atmospheric noise, and the positively skewed nature of atmospheric noise. The SST skewness is most affected by a nonlinearity in the response of the wind stress to SST anomalies. This is followed by the state-dependence of atmospheric noise. The skewed nature of atmospheric noise has only a minor effect on SST skewness.

GCMs tend to simulate lower atmospheric noise amplitudes than observations. Some GCMs show a nonlinear response of wind stress to SST, although weaker than in observations. These models simulate the most realistic SST skewness. Overall, both a nonlinear atmospheric response to SST and the dependence of noise on the background SST influence the El Niño/La Niña asymmetry.

Finally, we investigate the sensitivity of ENSO to uncertainties in the description of physical processes in a GCM by examining a set of GCM experiments with perturbations to key atmospheric and oceanic GCM parameters. For analyzing the runs, we use the same method as for the multi-model ensemble. The advantage of a perturbed physics ensemble is that it is not principally controlled by variations in the mean climate state. Studying only changes related to perturbed GCM parameters we conclude that feedbacks involved in the ICM response of SST to variations in wind stress and damping of SST anomalies provide the leading-order control on ENSO amplitude and spatial structure.

The method described in this thesis provides the possibility of using observations for exploring model biases in individual ENSO feedback processes in a quick and transparent way. Using this method will give better insight in GCMs and seasonal forecast models, which will help modellers improving their models. This will facilitate better seasonal forecasts and climate projections.