SPECIAL FEATURE: DATA ASSIMILATION Improving the quantification of terrestrial ecosystem carbon dynamics over the United States using an adjoint method

We used an adjoint version of the process-based Terrestrial Ecosystem Model (TEM) to optimize the model parameters in a spatially explicit manner by assimilating satellite-based estimates of gross primary production (GPP) in the conterminous United States. Traditionally, terrestrial ecosystem model parameterization is conducted at site-level for various plant functional types (PFTs). The optimal parameters are then extrapolated to regions that have the same plant functional types as parameterized. However, site-level parameterization might be only valid within a few kilometers in the footprint of the individual site. Extrapolation of the optimal parameters to a region with an assumption that the parameters are the same for all pixels for the same ecosystem type in the region may introduce significant errors in quantification of regional carbon dynamics. This study used Moderate Resolution Imaging Spectroradiometer GPP to optimize parameters for each pixel using adjoint method in a spatially explicit manner. The spatially explicit parameters were then used to quantify the regional carbon fluxes from 2000 to 2005. The estimated net ecosystem production (NEP) was used to drive a global atmospheric transport model, GEOSChem, to estimate the near surface atmospheric CO2 concentrations, which were then compared with flask measurements. We found that (1) the site-level TEM parameterization method provided good estimates of carbon fluxes at site levels, but had a large uncertainty in the regional simulations; (2) the spatially explicit parameterization improved the estimates of the spatial distribution and seasonal variation of regional carbon dynamics; and (3) when driven with the NEP estimated with the spatially explicit model, the GEOSChem captured the seasonal trend of near-surface CO2 concentrations better than that was driven with the estimates based on the site-level parameterized model. This study suggested that future quantification of regional carbon dynamics should consider the spatial variation of parameters, which could be optimized using spatially explicit carbon flux data.