The first task was to check whether the properties of El Niño and the Southern Oscillation in the models are close enough to reality to trust the projections under climate change. In general the current generation of climate models is better at simulating ENSO than the ones used for the Third Assessment Report (TAR) in 2001. However, there are large differences between the models. We downloaded the data from the PCMDI archive and checked some properties.

Most models simulate El Niño too far to the west, with a period that is too short. In these models ENSO is driven more by surface processes - direct coupling of sea surface temprature and wind - than by the subsurface processes that we know to dominate the observed ENSO.

Another group of models has a very regular oscillation, rather than the very irregular oscillations that are observed on earth. In these models La Niña also tends to be much stronger than El Niño, the opposite of reality. These models can also not be trusted to give a realistic forecast.

Finally, there is group of models that in the current climate simulate more or less the same irregular sequence of El Niño and La Niña events as has been seen over the last century. In these models the coupling between the ocean and atmophere is also in reasonable agreement with observations from the TAO array of buoys. These models were selected for further study.

The most realistic models do not show large shifts in the mean state. In some the sea surface temperature and sea-level pressure change in the same way as El Niño in the current climate, but much weaker (at most one quarter of the standard deviation in 2051-2100). In others the change is the other way. Other shifts in the climate are much more important, so based on the current models we do not expect the changes in the mean state to resemble El Niño or La Niña.

The standard deviation in the most reliable models increases or decreases up to 20% in 2051-2100, with more models showing an increase than a decrease. This change is about as large as the natural variability from the second half of the 19th century to the middle of the 20th century, so again the shift due to climate change is likely to be smaller than the shifts due to natural variability.

It is impossible to make forecasts for the relative strengths of El Niño and La Niña in the future climate as the current models do not reproduce these non-linear effects in the present climate.

Both the small effect of global warming on ENSO and the relatively large spread between the models can be explained by examining the sub-processes of ENSO. It turns out that these each change by large amounts, often a factor two, but these changes all tend to cancel. The outcome therefore depends strongly on the details of the models, which is why these disagree on the sign of the changes.

Due to the higher temperature of the equatorial Pacific Ocean there can be more clouds in the central Pacific. During El Niño these block the sun and decrease the warming, so that ENSO is damped more strongly. However, the strength of the trade winds decreases, due to a slowing down of the tropical circulation. This causes the cold water to come closer to the surface, and affect sea surface temperature more strongly. The direct effect of the wind is also increased, partly due to a shallower mixed surface layer that heats and cools more quickly. These two positive changes just compensate the increased damping.

Standard physics dictates that the same amount of temperature change gives more heating at a higher background temperature, and hence a larger wind effect. However, this is also cancelled on average by the increased stability of the tropical atmosphere over the oceans.

The observation that ENSO does not change by much in the most realistic climate models is a cancellation between large changes in different sub-processes of the ENSO cycle. We are still investigating whether this is accidental, or whether there is an underlying physical reason for this cancellation.