Publication Date


Document Type



An edited version of this paper was published by AGU. Copyright (2021) American Geophysical Union.

Please refer to:
Gerken, Tobias, Sha Feng, Klaus Keller, Thomas Lauvaux, Joshua P. DiGangi, Yonghoon Choi, Bianca Baier, and Kenneth J. Davis. Examining CO2 Model Observation Residuals Using ACT-America Data. Journal of Geophysical Research: Atmospheres 126, no.18 (2021): e2020JD034481.

Atmospheric CO2 inversion typically relies on the specification of prior flux and atmospheric model transport errors, which have large uncertainties. Here, we used ACT-America airborne observations to compare CO2 model-observation mismatch in the eastern U.S. and during four climatological seasons for the mesoscale WRF(-Chem) and global scale CarbonTracker/TM5 (CT) models. Models used identical surface carbon fluxes, and CT was used as CO2 boundary condition for WRF. Both models showed reasonable agreement with observations, and CO2 residuals follow near symmetric peaked (i.e. non-Gaussian) distribution with near zero bias of both models (CT: -0.34+/-3.12 ppm; WRF: 0.82+/-4.37 ppm). We also found large magnitude residuals at the tails of the distribution that contribute considerably to overall bias. Atmospheric boundary-layer biases (1-10 ppm) were much larger than free tropospheric biases (0.5-1ppm) and were of same magnitude as model-model differences, whereas free tropospheric biases were mostly governed by CO2 background conditions. Results revealed systematic differences in atmospheric transport, most pronounced in the warm and cold sectors of synoptic systems, highlighting the importance of transport for CO2 residuals. While CT could reproduce the principal CO2 dynamics associated with synoptic systems, WRF showed a clearer distinction for CO2 differences across fronts. Variograms were used to quantify spatial correlation of residuals and showed characteristic residual length scales of approximately 100 km to 300 km. Our findings suggest that inclusion of synoptic weather-dependent and non-Gaussian error structure may benefit inversion systems.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.