Emergent Constraints on CMIP6 Climate Warming Projections: Contrasting Cloud- and Surface Temperature–Based Constraints

Abstract

AbstractThe latest Coupled Model Intercomparison Project (CMIP6) multimodel ensemble shows a broader range of projected warming than the previous-generation CMIP5 ensemble. We show that the projected warming is well correlated with tropical and subtropical low-level cloud properties. These physically meaningful relations enable us to use observed cloud properties to constrain future climate warming. We develop multivariate linear regression models with metrics selected from a set of potential constraints based on a stepwise selection approach. The resulting linear regression model using two low-cloud metrics shows better cross-validated results than regression models that use single metrics as constraints. Application of a regression model using the low-cloud metrics to climate projections results in similar estimates of the mean, but substantially narrower uncertainty ranges, of projected twenty-first-century warming when compared with unconstrained simulations. The resulting projected global-mean warming in 2081–2100 relative to 1995–2014 is 2.84–5.12 K (5%–95% range) for Shared Socioeconomic Pathway (SSP) 5–8.5 compared with a range of 2.34–5.81 K for unconstrained projections, and 0.60–1.70 K for SSP1–2.6 compared to an unconstrained range of 0.38–2.04 K. We provide evidence for a higher lower bound of the projected warming range than that obtained from constrained projections based on the past global-mean temperature trend. Consideration of the impact of the sea surface temperature pattern effect on the recent observed warming trend, which is not well captured in the CMIP6 ensemble, indicates that the relatively low projected warming resulting from the global-mean temperature trend constraint may not be reliable and provides further justification for the use of climatologically based cloud metrics to constrain projections.