Predictability Horizons in the Global Carbon Cycle Inferred From a Perfect‐Model Framework

On interannual timescales the growth rate of atmospheric CO2 is largely controlled by the response of the land and ocean carbon sinks to climate variability. Yet, it is unknown to what extent this variability limits the predictability of atmospheric CO2 variations. Using perfect‐model Earth System Model simulations, we show that variations in atmospheric CO2 are potentially predictable for 3 years. We find a 2‐year predictability horizon for global oceanic CO2 flux with longer regional predictability of up to 7 years. The 2‐year predictability horizon of terrestrial CO2 flux originates in the tropics and midlatitudes. With the predictability of the isolated effects of land and ocean carbon sink on atmospheric CO2 of 5 and 12 years respectively, land dampens the overall predictability of atmospheric CO2 variations. Our research shows the potential of Earth System Model‐based predictions to forecast multiyear variations in atmospheric CO2. Plain Language Summary The amount of anthropogenic carbon emissions absorbed by land and ocean from the atmosphere varies annually due to their sensitivity to climate. Therefore, the atmospheric CO2 growth rate does not strictly follow the emissions signal. Whether decadal prediction systems can also predict variations of atmospheric CO2 has not been shown yet but is crucial to inform policy makers about the efficiency of the implementation of the Paris Agreement. Using numerical Earth System simulations in an idealized prediction framework, we show that global atmospheric CO2 is predictable up to 3 years in advance. The global ocean and land CO2 fluxes are predictable for 2 years. The isolated effects of the land and ocean carbon sink on atmospheric CO2 are predictable for 5 and 12 years, respectively. Therefore, the land carbon cycle limits atmospheric CO2 predictability. Our study demonstrates that simulation‐based multiyear forecasts have the potential to predict natural atmospheric CO2 variations. Key Points Global annual atmospheric CO2 variations are potentially predictable up to 3 years in advance The global oceanic CO2 flux is predictable for 2 years with regional predictability up to a decade The global land CO2 flux predictability of 2 years is dominated by the tropical forests