The OCO-2 v10 MIP

The Orbiting Carbon Observatory-2 (OCO-2) model intercomparison project (MIP) is a collaboration among atmospheric CO2 modelers to study the impact of assimilating OCO-2 retrieval data into atmospheric inversion models. The results from the set of simulations performed by this project have been released as a level-4 flux product.

Introduction Measuring CO2 is critical for understanding how quickly the planet will warm up in response to fossil fuel burning and other human-caused CO2 inputs, such as deforestation. About half of the anthropogenic CO2 we emit is taken up by the oceans and the land biosphere. To know how quickly the planet will warm, we must understand the processes controlling this CO2 uptake. We also need accurate estimates of the anthropogenic input from fossil fuel buring and deforestation. The spatial and temporal gradients in CO2 mole fractions around the globe can tell us about the surface fluxes that caused these gradients and shed light on the processes causing the fluxes.

Retrieval process OCO-2 infers atmospheric CO2 mixing ratios from space by looking at solar radiation in the near-infrared that has reflected off of the Earth's surface and atmosphere. It measures absorption by CO2 in the weak and strong CO2 bands at 1.61 and 2.06 μm, as well as by O2 in the O2 A-band at 0.76 μm. From these measurements, the CO2 mixing ratio as a function of altitude (or pressure) is inferred using inverse methods. Since the photons measured are affected not only by CO2 and O2 absorption, but also by scattering from clouds and aerosols as well as other processes, the full physics of the radiative transfer problem must be simulated to obtain CO2 estimates. The estimation scheme requires the use of a prior CO2 profile defined on each of the 20 pressure levels on which it operates. Despite this vertical discretization, only the pressure-weighted column average (XCO2) is reported by the OCO-2 team, since it is considered the most trustworthy piece of CO2 information in the retrieval.

The raw XCO2 values contain significant biases, as revealed by comparison to TCCON-retrieved XCO2, to models, to profile data from in situ CO2 measurements, and to prior knowledge of the atmospheric carbon cycle. To produce XCO2 estimates that can be used by carbon cycle researchers, an empirical bias correction is applied to the OCO-2 retrievals (cf. Osterman et al., 2016; O'Dell et al., 2018). The bias correction scheme is intentionally parsimonious to avoid adding spurious variability to the retrievals as a result of overfitting to the sparse and potentially biased evaluation constraints. This procedure can change the XCO2 by 2-3 ppm, but cannot be expected to remove all systematic errors from the retrievals.

OCO-2 product levels The radiance measurements from OCO-2 are packaged into Level-1 products. The column-averaged CO2 mixing ratios retrieved from these measurements are packaged into Level-2 products. For OCO-2, these Level-2 products include a bias-corrected version of the retrievals, as well as helpful diagnostic parameters. The CO2 column-average mixing ratios may be interpolated to obtain Level-3 products on a regular time/space grid. When the CO2 retrievals are used to estimate surface fluxes (sources and sinks) of CO2, these fluxes may be packaged into Level-4 products.

Atmospheric inverse models Atmospheric tracer transport models are often used to relate fluxes of CO2 into the atmosphere at the surface to the perturbations in CO2 mixing ratio that they cause in the interior of the atmosphere. Once this link is made, the unknown surface fluxes may be solved for using the CO2 mixing ratios (known from the measurements) as an inverse problem. A Bayesian prior is generally used to allow fluxes to be solved for robustly when the data used are sparse. Since the inversion framework is not set up to deal with biases in the measurements or errors in the transport model used, such errors are often assessed with sensitivity studies.

The great promise of observing CO2 from space is its global coverage, which should allow a more complete view of the workings of the global carbon cycle compared to the relatively sparse in situ network. A disadvantage of space-based observations is that the information on CO2 mixing ratios is the vertical average throughout the entire column; ideally, one would like to measure CO2 at or near the surface (as the in situ measurements do), where the impact of the surface fluxes taking up or emitting CO2 is the largest. Another complication is that the satellite retrievals are less accurate than that of the in situ measurements, given the complexities of the satellite retrieval problem. The systematic errors in these retrievals are partially mitigated by an empirical bias correction procedure, described in Osterman et al. (2016) and O'Dell et al. (2018). In practice, satellite retrievals and ground-based CO2 measurements have complemented each other. The sparse in situ-measurement-based approach is forced to rely more on atmospheric transport models when inferring CO2 surface fluxes. Flux estimates using satellite retrievals rely less on the transport models because of the dense retrieval coverage, but are affected more by systematic errors in the retrievals. Comparing the two different perspectives has led to new insights into the workings of the global carbon cycle.

The MIP experiments To understand the impact of transport model error and inversion method on estimated fluxes, we have performed flux inversions of the in situ measurements and OCO-2 retrievals with a group of different transport models and inversion methods. To gauge the effect of retrieval errors on the OCO-2 estimates, we have performed inversions with different combinations of the OCO-2 viewing mode (nadir or glint viewing geometry, land or ocean surface), and compared these to inversions assimilating the in situ measurements. This comparison has been done in the context of a formal model intercomparison project (MIP), with a preformulated experiment protocol, common formats for the submitted results, and sophisticated visualization tools. We have summarized the flux results obtained from the OCO-2 MIP in a Level-4 flux product, as described below.

Protocol details availablle on the downloads page.


Modelers submitting global flux inversion results to the OCO-2 MIP are given in Table 1, along with their transport model, meteorology source, and inversion method.
Table 1. OCO-2 MIP participants and model details.
Model Contact Institution Transport Model Meteorology Inverse Method
Ames Matthew Johnson and Sajeev Philip NASA Ames Research Center GEOS-Chem MERRA-2 4D-Var
CAMS Frédéric Chevallier LSCE France LMDz ERA-interim 4D-Var
COLA Zhiqiang Liu
CMS-Flux Junjie Liu NASA JPL GEOS-Chem GEOS-FP 4D-Var
CSU Andrew Schuh Colorado State University GEOS-Chem MERRA-2 Bayesian synthesis
CT Andy Jacobson University of Colorado and NOAA GML TM5 ERA-interim EnKF
JHU Scot Miller
LoFI Brad Weir
NIES Shamil Maksyuotov
OU Sean Crowell University of Oklahoma TM5 ERA-interim 4D-Var
PCTM David Baker Colorado State University PCTM MERRA-2 4D-Var
TM5-4DVAR Sourish Basu University of Maryland and NASA GMAO TM5 ERA-interim 4D-Var
UT Feng Deng University of Toronto GEOS-Chem GEOS-FP 4D-Var
WOMBAT Michael Bertolacci, Andrew Zammit Mangion, Noel Cressie University of Wollongong GEOS-Chem MERRA-2 MCMC

Inversion experiments

Modelers performed inversion experiments assimilating various types of observational constraints. These experiments extend from the beginnning of the OCO-2 retrievals in September 2014 through the end of 2020. Results are reported for the six-year interval 2015-2020.

  • IS: Assimilation of in situ CO2 measurements
  • LNLG: Assimilation of OCO-2 Land Nadir and Land Glint retrievals
  • LNLGIS: Assimilation of OCO-2 Land Nadir and Land Glint retrievals, and in situ CO2 measurements
  • OG: Assimilation of OCO-2 Ocean Glint retrievals
  • LNLGOGIS: Assimilation of in situ CO2 measurements and all OCO-2 retrievals


  • Schuh A. E., A. R. Jacobson, S. Basu, B. Weir, D. Baker, K. Bowman, F. Chevallier, S. Crowell, K. J. Davis, F. Deng, et al. Quantifying the impact of atmospheric transport uncertainty on co2 surface flux estimates. Global Biogeochemical Cycles, 33(4):484–500, 2019.
  • Chatterjee et al., in prep.
  • Crowell S., D. Baker, A. Schuh, S. Basu, A. R. Jacobson, F. Chevallier, J. Liu, F. Deng, L. Feng, K. McKain, et al. The 2015–2016 carbon cycle as seen from oco-2 and the global in situ network. Atmospheric Chemistry and Physics, 19(15):9797–9831, 2019.
  • O'Dell, C. W., A. Eldering, P. O. Wennberg, D. Crisp, M. R. Gunson, B. Fisher, C. Frankenberg, M. Kiel, H. Lindqvist, L. Mandrake, A. Merrelli, V. Natraj, R. R. Nelson, G. B. Osterman, V. H. Payne, T. R. Taylor, D. Wunch, B. J. Drouin, F. Oyafuso, A. Chang, J. McDuffie, M. Smyth, D. F. Baker, S. Basu, F. Chevallier, S. M. R. Crowell, L. Feng, P. I. Palmer, M. Dubey, O. E. Garc ́ıa, D. W. T. Griffith, F. Hase, L. T. Iraci, R. Kivi, I. Morino, J. Notholt, H. Ohyama, C. Petri, C. M. Roehl, M. K. Sha, K. Strong, R. Sussmann, Y. Te, O. Uchino, and V. A. Velazco. Improved retrievals of carbon dioxide from the Orbiting Carbon observatory-2 with the version 8 ACOS algorithm. Atmospheric Measurement Techniques Discussions, 2018:1–57, 2018. doi: 10.5194/amt-2018-257
  • Osterman, G. B., A. Eldering, C. Avis, B. Chafin, C. W. O’Dell, C. Frankenberg, B. M. Fisher et al. "Orbiting Carbon Observatory-2 (OCO-2) data product user’s guide, operational L1 and L2 data versions 7 and 7R." Jet Propulsion Laboratory, Pasadena, CA, USA (2016). Available online at