Predicting how the strength and character of the El Niño/Southern Oscillation (ENSO) will change as the climate warms is crucial for a number of societal impacts, yet there are fundamental limitations to our understanding of ENSO dynamics. The major obstacles are related to sampling length, physical adjustments to climate changes, errors in model physics and uncertainties in forcing projections. This dissertation uses these issues to assess what we currently can and cannot say about future ENSO variability.

The temporal extent of modern observations is too short to properly measure natural ENSO variations: averaging together at least 200 years is required to obtain robust ENSO statistics in a stable climate. Using paleoclimate ‘proxies’ to extend the observational record is another option, but is complicated by the uncertainties involved in translating between model and proxy signals. Coral oxygen isotopes, the most commonly used ENSO proxy, are shown to be governed by nonlinear dynamics: a more accurate ‘forward model’ for coral δ¹⁸O is needed. Even using such a model, at least 4-5 contemporaneous records will be required for accurate ENSO amplitude estimation.

Simulations using several IPCC-class general circulation models (GCMs) are used to demonstrate that the adjustment to climate change itself takes place over decadal timescales, meaning that ENSO response is not statistically significant during the 21st century. This implies that current model intercomparison experiments are insufficient to measure the true range of ENSO climate sensitivity. However, significant changes to atmospheric teleconnections may take place within the 21st century: the NCAR Community Climate System Model version 4 (CCSM4), for example, predicts harsher winters in the Southwestern US during La Niña and weaker Australian teleconnections during both El Niño and La Niña.

Stabilized CCSM3.5 simulations are then performed, which show that once the climate has equilibrated, the ENSO response to a CO2 increase eventually does become significant. However, the details of that response are sensitive to small changes in model physics: the ENSO climate sensitivity in the CCSM3.5 and CCSM4 oppose one another, and the mechanisms for the difference are as yet unclear. Seasonal forcing, high-frequency wind stress variability, or other processes may be responsible, but a complete diagnosis requires longer CCSM4 simulations than are currently available.

Finally, an additional complication is discussed: future ENSO projections all rely on the standardized emissions scenarios from the IPCC. Projections of future emissions reductions may be overly optimistic, perhaps requiring attention to a wider range of CO₂ changes for accurate ENSO impacts studies.