A comprehensive carbon dioxide analysis system for estimating CO2 emissions

Greenhouse gas emissions due to combustion of fossil fuel can be estimated from observations of variations in atmospheric trace gases in time and space. Quantitative interpretation of these variations requires accounting for stronger changes due to other processes such as ecosystem metabolism, biomass burning, and air-sea gas exchange that operate on global scales. We have developed and tested an analysis system for assimilation of CO2 variations measured by a combination of sampling, in-situ, and remotely-sensed observations. The system combines existing models of CO2 exchanges due to hourly photosynthesis and respiration, daily air-sea gas exchange, biomass burning, fossil fuel emissions, and atmospheric transport. This comprehensive system allows direct comparison to the observed record of both in-situ and remotely sensed atmospheric CO2 at hourly timescales. By design, we decompose surface fluxes of CO2 into the atmosphere into “fast processes” that are well-understood and modeled using mechanistic algorithms, and more slowly-varying fluxes due to land use change, incorrect specification of decomposing carbon pools, and other persistent biases in the forward component models. These slowly varying components are then estimated from atmospheric obervations by the Maximum Likelihood Ensembe Filter, a data assimilation framework. The system is operated on a 0.5° x 0.67° grid, providing global mesoscale coverage, and has good skill at replicating diurnal, synoptic, and seasonal variations over vegetated land surfaces. It is driven by meteorological output from the NASA Goddard EOS Data Assimilation System. Surface weather from the system drives calculations of terrestrial ecosystem metabolism (radiation, precipitation, humidity, temperature) and air-sea gas exchange (wind), with other input data coming from satellite data products.