Assimilation of satellite carbon dioxide retrievals and calibration of their error statistics

Satellite CO2 observations, like those from GOSAT and OCO-2, provide new opportunities to estimate carbon flux globally. However, these flux estimates are strongly dependent on the bias corrections and uncertainties applied in the combination of observed and simulated CO2. We present the results of the assimilation of GOSAT/ACOS Version 3.4 average column dry-air mixing ratios of CO2 into the GEOS-5 model and use those results to calibrate the a priori and retrieval error statistics. In particular, we compare the mean difference of the predictions against hourly data from the Total Carbon Column Observing Network (TCCON) and 9 sites in the Japan-Russia Siberian Tall Tower Inland Observation Network (JR-STATION). While model-free bias correction methods must focus on time windows where the satellite passes near a given site, the model-based approach can compare against all data from the sites during a given time span. The difference is most notable when low light or cloudy conditions prevent satellite retrievals over long periods of time, e.g., during the winter over the Eurasian boreal forests. This type of evaluation and improvement of error statistics is essential to maximizing the impact of space-based CO2 observations and improving flux attribution.