Improving the Estimation of Human Climate Influence by Selecting Appropriate Forcing Simulations

The regression-based optimal fingerprinting is a key tool for quantifying human climate influence. Most studies over the past decade used Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations, limiting fingerprinting regression configuration options. The CMIP6 Detection and Attribution Model Intercomparison Project (DAMIP) provides several types of individual forcing simulations and thus greater configuration flexibility. To avoid overfitting the limited observational data, we suggest that a DAMIP-based perfect model study is first used to best configure the fingerprinting regression prior to its application to observations. We find that a regression using all-forcing, aerosol-only, and natural-only simulations is an overall best option for constraining human-induced global terrestrial warming, which differs from choices commonly made previously. Applying this configuration to observations, we estimate that of the observed terrestrial warming of ∼1.5°C between 1850–1900 and 2011–2020, anthropogenic greenhouse gases contributed 1.4 to 2.3°C, offset by aerosol cooling of 0.2 to 1.2°C.