Three decades of simulated global terrestrial carbon fluxes from a data assimilation system confronted with different periods of observations

Abstract

Abstract. During the last decade, carbon cycle data assimilation systems (CCDAS) have focused on improving the simulation of seasonal and mean global carbon fluxes over a few years by simultaneous assimilation of multiple data streams. However, the ability of a CCDAS to predict longer-term trends and variability of the global carbon cycle and the constraint provided by the observations have not yet been assessed. Here, we evaluate two near-decade-long assimilation experiments of the Max Planck Institute – Carbon Cycle Data Assimilation System (MPI-CCDAS v1) using spaceborne estimates of the fraction of absorbed photosynthetic active radiation (FAPAR) and atmospheric CO2 concentrations from the global network of flask measurement sites from either 1982 to 1990 or 1990 to 2000. We contrast these simulations with independent observations from the period 1982–2010, as well as a third MPI-CCDAS assimilation run using data from the full 1982–2010 period, and an atmospheric inversion covering the same data and time. With 30 years of data, MPI-CCDAS is capable of representing land uptake to a sufficient degree to make it compatible with the atmospheric CO2 record. The long-term trend and seasonal amplitude of atmospheric CO2 concentrations at station level over the period 1982 to 2010 is considerably improved after assimilating only the first decade (1982–1990) of observations. After 15–19 years of prognostic simulation, the simulated CO2 mixing ratio in 2007–2010 diverges by only 2±1.3 ppm from the observations, the atmospheric inversion, and the MPI-CCDAS assimilation run using observations from the full period. The long-term trend, phenological seasonality, and interannual variability (IAV) of FAPAR in the Northern Hemisphere over the last 1 to 2 decades after the assimilation were also improved. Despite imperfections in the representation of the IAV in atmospheric CO2, model–data fusion for a decade of data can already contribute to the prognostic capacity of land carbon cycle models at relevant timescales.