Uncertainty in Future Warming is Reduced by Combining Emergent Constraints
Emergent constraints are currently-observable quantities with skill at predicting future climate change. Identifying potential emergent constraints has become extremely popular in recent years because their predictions are grounded in quantities we can actually measure and because they provide estimates of climate sensitivity which are somewhat independent of prediction existing methods. This study catalogs the predictions from previously-proposed constraints for climate sensitivity and provides a mathematical basis for combining them.
Emergent constraints consistently predict strong warming. As a result, their combined prediction is for a very warm future with less uncertainty than found in previous studies. This result is alarming but is predicated on the credibility of the emergent constraints sampled – which hasn’t been firmly established yet. Even in light of this uncertainty, the sheer number of constraints providing a consistent answer suggests that we should prepare for extreme warming.
Many emergent constraints (currently-observable quantities with skill at predicting future change) have been proposed recently for equilibrium climate sensitivity (ECS: global-average temperature change after doubling CO2 and letting the planet re-equilibrate). This study catalogs 11 of these constraints, finding that most of them predict larger climate sensitivity than traditional methodologies. It also explores methods for using information from multiple emergent constraints to provide sharper estimates of ECS. This method is based on fitting a multivariate Gaussian PDF for all of the constraints and ECS using model data from phase 3 and 5 of the Coupled Model Intercomparison Project, extended to account for uncertainties in sampling this multidimensional PDF with a small number of models, for observational uncertainties in the constraints, and for overconfidence about the correlation of the constraints with the climate sensitivity. The method accounts for correlations between constraints but can become unstable when constraints are too strongly inter-related. Several fixes for this problem are provided. The simplest fix is to ignore correlations between constraints entirely. Doing so is shown to cause only minor changes to the general solution in cases where both are stable. Ignoring correlations between constraints provides a convenient decomposition of the contribution of each constraint into its individual ECS prediction and its weight towards the combined prediction. The ±2σ range of ECS predicted by our combination method is 4.3 ± 0.7 K without overconfidence adjustment and 4.0 ± 1.3 K with the assumption that predictor confidence is overestimated by a factor of 3.