Carbon Flux Variability From a Relatively Simple Ecosystem Model With Assimilated Data Is Consistent With Terrestrial Biosphere Model Estimates

Abstract

Abstract Modeling the net carbon balance is challenging due to the knowledge gaps in the variability and processes controlling gross carbon fluxes. Terrestrial carbon cycle modeling is susceptible to several sources of bias, including meteorological uncertainty, model structural uncertainty, and model parametric uncertainty. To determine the impact of these uncertainties, we compare three model‐derived representations of the global terrestrial carbon balance across 1997–2009: (1) observation‐constrained model‐data fusion (CARBon Data Model FraMEwork, CARDAMOM), (2) the reanalysis‐driven Trends in Net Land‐Atmosphere Carbon Exchange (TRENDY) land biosphere model ensemble, and (3) the Coupled Model Intercomparison Project 5 (CMIP5) Earth System Model ensemble. We consider the spread in carbon cycle simulations attributable primarily to parametric uncertainty (CARDAMOM), structural uncertainty (TRENDY), and combined structural and simulated meteorological uncertainty (CMIP5). We find that the spread across the CARDAMOM ensemble long‐term mean—produced by parameter uncertainty—is larger than the spread of TRENDY and CMIP5 for net biosphere exchange (NBE) but similar for gross primary productivity (GPP). The carbon flux dynamics of CARDAMOM compares to models in TRENDY as well as models in TRENDY compare to each other in many regions for NBE seasonal (nine of 12), NBE interannual (11 of 12), and GPP seasonal variability (7 of 12), although not for GPP interannual variability (2 of 12). The simple model structure of CARDAMOM and systematic assimilation of observations is sufficient to produce carbon dynamics within the range of more complex models. These results are encouraging for the use of model‐data fusion products with empirically estimated uncertainty for global carbon cycle studies.

Keywords

• carbon cycle
• data assimilation
• carbon modeling