Reducing Uncertainty in Terrestrial Biosphere Models with Satellite Observations of Atmospheric CO2: Comparing MsTMIP with GOSAT

Abstract:

Global-scale terrestrial biosphere models (TBMs) vary in their underlying driving assumptions, inputs, and parameterizations. As a result, TBM estimates of carbon fluxes and pools vary greatly and the lack of direct observations of land-atmosphere carbon exchange at the same spatio-temporal resolution (e.g., 0.5° x 0.5° degree and sub-daily to monthly) of model estimates makes it difficult to assess TBM performance in terms of their ability to represent the terrestrial carbon cycle. Atmospheric CO₂ measurements, however, may be a potentially powerful observational constraint for TBMs because they provide an integrated view of surface sources and sinks of carbon. The Greenhouse Gases Observing Satellite (GOSAT) measures atmospheric CO₂ from space at spatio-temporal scales relatively consistent with model estimates. Using TBM estimates from the North American Carbon Program Multi-scale synthesis and Terrestrial Model Intercomparison Project (MsTMIP), together with estimates of fossil fuel emissions and air-sea fluxes, we translate surfaces fluxes into atmospheric CO₂ concentrations using the GEOS-Chem atmospheric transport model. The performance of MsTMIP TBMs is evaluated by comparing the dry air column-averaged mole fractions of CO₂ (Χ_CO2) from transported surface fluxes to observations of Χ_CO2 from GOSAT. While MsTMIP Χ_CO2 signals are generally consistent with GOSAT Χ_CO2 in the southern hemisphere, MsTMIP and GOSAT X_CO2 show profound differences in the northern hemisphere (NH). In general, MsTMIP X_CO2 tends to be higher than GOSAT X_CO2 at northern latitudes, especially in the NH summer and fall. Looking specifically at regions in the NH, we compare each MsTMIP Χ_CO2 to GOSAT Χ_CO2 in terms of its ability to reproduce the spatial distribution, magnitude and timing of the GOSAT Χ_CO2 seasonal cycle. We use the information derived from the comparison to link model performance with how certain processes are represented within the models themselves, thus aiding in model development and improvement.