Measurement Network Design within the CarbonTracker-Lagrange Inverse Modelling Framework

The CarbonTracker-Lagrange Inverse Modelling framework was used to investigate the impact of additional observations on estimates of carbon dioxide emissions and removals by plants over North America. The emission and uptake fluxes are estimated along with their uncertainties a using Bayesian inverse modeling approach whereby “first guess” fluxes are adjusted to improve agreement with available measurement of atmospheric carbon dioxide. The current measurement network consists primarily of sites operated by NOAA’s Global Monitoring Division and by Environment and Climate Change Canada (ECCC). Twenty-three new candidate measurement locations were considered, corresponding to sites operated by Earth Networks (www.earthnetworks.com). Data records from each site were assessed by four criteria: data coverage over time, the quality of the data, geographic location (with regards to both what was expected to be useful for the inversion when the location of other sites was accounted for and the proximity of the site to potent localized fluxes, such as cities), and their impact on the optimized flux estimates. Synthetic data were created to simulate the atmospheric response to specified “true” fluxes from a biospheric model. The true fluxes were multiplied by footprint functions derived from the Stochastic Time-Inverted Lagrangian Transport – Weather Research and Forecasting Model and random noise was added to simulate measurement errors. These synthetic data were then used in a set of inversions, each of which included the existing network plus one additional candidate site along with a baseline case that was performed using synthetic measurements corresponding to only NOAA and ECCC sites. The inversions were done for a test case over the June, July, and August months of 2015, and the performance of each site was assessed mainly over the central month of July, corresponding to the period with maximum plant uptake. Several criteria were used to determine the impact of the inclusion of each site on the inversion, including the difference in optimized fluxes, the difference in the uncertainty of the results, and the difference in the distance from the optimized fluxes to the “true” fluxes.