Global carbon budget variations in emission-driven earth system model predictions

Predictions of the variations in anthropogenic global carbon budget (GCB), i.e., CO2 emissions and their redistribution among the atmosphere, ocean, and land reservoirs, is crucial to constrain the global carbon cycle and climate change of the past and facilitate their prediction and projection into the future. Global carbon project assesses the GCB every year by taking into account available datasets and stand-alone model component simulations. The utilization of different data sources leads to an unclosed budget, i.e., budget imbalance. We propose a novel approach to assess the GCB in decadal prediction systems based on emission-driven earth system models (ESMs). Such a fully coupled prediction system enables a closed carbon budgeting and therefore provides an additional line of evidence for the ongoing assessments of the GCB.

As ESMs have their own mean state and internal variability, we assimilate ocean and atmospheric observational and reanalysis data into Max Planck Institute Earth system model (MPI-ESM) to reconstruct the actual evolution of climate and carbon cycle towards to the real world. In the emission-driven model configuration, the carbon cycle changes in response to the physical state changes, in the meanwhile, the feedback of atmospheric CO₂ changes to physics are also considered via interactive carbon cycle. Our reconstructions capture the observed GCB variations in the past decades. They show high correlations relative to the assessments from the global carbon project of 0.75, 0.75 and 0.97 for atmospheric CO₂ growth, air-land CO₂ fluxes and air-sea CO₂ fluxes, respectively. Retrospective predictions starting from the reconstruction show promising predictive skill for the global carbon cycle up to 5 years for the air-sea CO₂ fluxes and up to 2 years for the air-land CO₂ fluxes and atmospheric carbon growth rate. Furthermore, evolution in atmospheric CO₂ concentration in comparing to satellite and in-situ observations show robust skill in reconstruction and next-year prediction.