Biases in regional carbon budgets from covariation of surface fluxes and weather in transport model inversions

Recent advances in atmospheric transport model inversions could significantly reduce uncertainties in land carbon uptake through the assimilation of CO2 concentration measurements at weekly and shorter timescales. The potential of these measurements for reducing biases in estimated land carbon sinks depends on the strength of covariation between surface fluxes and atmospheric transport at these timescales and how well transport models represent this covariation. Daily to seasonal covariation of surface fluxes and atmospheric transport was estimated in observations at the US Southern Great Plains Atmospheric Radiation Measurement Climate Research Facility, and compared to an atmospheric transport model inversion (CarbonTracker). Covariation of transport and surface fluxes was stronger in CarbonTracker than in observations on synoptic (daily to weekly) timescales, with a wet year (2007) having significant covariation compared to a dry year (2006). Differences between observed and CarbonTracker synoptic covariation resulted in a 0.3 ppm CO2 enhancement in boundary layer concentrations during the growing season, and a corresponding enhancement in carbon uptake by 13% of the seasonal cycle amplitude in 2007, as estimated by an offline simplified transport model. This synoptic rectification of surface flux variability was of similar magnitude to the interannual variability in carbon sinks alone, and indicates that interannual variability in the inversions can be affected by biases in simulated synoptic rectifier effects. The most significant covariation of surface fluxes and transport had periodicities of 10 days and greater, suggesting that surface flux inversions would benefit from improved simulations of the effects of soil moisture on boundary layer heights and surface CO2 fluxes. Soil moisture remote sensing could be used along with CO2 concentration measurements to further constrain atmospheric transport model inversions.