Narrowing uncertainties in projected warming by constraining using the past global warming trend with the pattern effect removed

Observational constraint methods based on emergent relationships between observable predictors and future projected warming across multi-model ensembles enable us to constrain multi-model projections. Unforced internal variability in predictors can weaken such emergent relationships. Assessing the Sixth Coupled Model Intercomparison Project (CMIP6) with all accessible realizations, we find that there of sea surface temperature (SST) trend over the eastern tropical pacific (ETP) which is well correlated with the global warming trend. The strong cooling in the ETP in observations induces a global-scale cooling, yet most realizations in the CMIP6 multi-model ensemble cannot reproduce it. Using the observed raw historical global mean near surface air temperature (GSAT) trend as a constraint therefore results in a relatively lower projected 21^st century warming. However, by removing the unforced internal variability associated with variation in the ETP in observed and simulated GSAT trends, we find an enhanced correlation between GSAT trends and projected warming and improved results in an imperfect model test. This approach results in a relatively higher 21^st century warming than a constrained projection based on the raw GSAT trend, and brings constrained projections into much closer agreement with projections constrained using climatological cloud metrics.