Consistency of global carbon budget between concentration- and emission-driven historical experiments simulated by CMIP6 Earth system models and suggestion for improved simulation of CO₂ concentration

Abstract. Anthropogenically emitted CO₂ from fossil fuel use and land use change is partly absorbed by terrestrial ecosystems and the ocean, while the remainder retained in the atmosphere adds to the ongoing increase in atmospheric CO₂ concentration. Earth system models (ESMs) can simulate such dynamics of the global carbon cycle and consider its interaction with the physical climate system. The ESMs that participated in the Coupled Model Intercomparison Project phase 6 (CMIP6) performed historical simulations to reproduce past climate–carbon cycle dynamics. This study investigated the cause of CO₂ concentration biases in ESMs and identified how they might be reduced. First, we compared simulated historical carbon budgets in two types of experiments: one with prescribed CO₂ emissions (the emission-driven experiment, “E-HIST”) and the other with prescribed CO₂ concentration (the concentration-driven experiment, “C-HIST”). As CMIP7 design is being considered it is important to explore any differences or implications in what these variations can tell us. The findings confirmed that the multi-model means of the carbon budgets simulated by one type of experiment generally showed good agreement with those simulated by the other. However, the multi-model average of cumulative compatible fossil fuel emission diagnosed from the C-HIST experiment was lower by 35 PgC than that used as the prescribed input data to drive the E-HIST experiment; the multi-model average of simulated CO₂ concentration for 2014 in E-HIST was higher by 7 ppmv than that used to drive C-HIST. Second, we investigated the potential linkages of two types of carbon cycle indices: simulated CO₂ concentration in E-HIST and compatible fossil fuel emission in C-HIST. It was confirmed quantitatively that the two indices are reasonable indicators of overall model performance in the context of carbon cycle feedbacks, although most models cannot accurately reproduce the cumulative compatible fossil fuel emission and thus cannot reproduce the CO₂ concentration precisely. Third, analysis of the atmospheric CO₂ concentration in five historical eras enabled identification of periods that caused the concentration bias in individual models. Further analysis based on a combination of four types of historical experiments suggested non-negligible impacts of non-CO₂ effects on the carbon cycle, implying their potential importance for future projections. It is suggested that this non-CO₂ effect is the reason why the magnitude of the natural land carbon sink in historical simulations is difficult to explain based on analysis of idealized experiments. Finally, accurate reproduction of land use change emission is critical for better reproduction of the global carbon budget and CO₂ concentration. The magnitude of simulated land use change emission not only affects the level of net land carbon uptake but also determines the magnitude of the ocean carbon sink in the emission-driven experiment.