Figure 1. Monthly emissions. The bars in this figure represent carbon dioxide emissions for each month in PgC yr-1 from the specified region. OCO-2 models simulated three types of surface-to-amosphere exchange of CO2, each of which is shown in a different color: fossil fuel emissions (tan), terrestrial biosphere flux possibly including implicit fires (green), and air-sea gas exchange (blue). Negative emissions indicate that the flux removes CO2 from the atmosphere, and such sinks have bars that extend below zero. The net surface exchange, computed as the sum of these three components, is shown as a thick black line.

• Figure 1a: Prior emissions and their spread.
• Figure 1b: Posterior emissions and their spread for the IS experiment (assimilating traditional in situ measurements)
• Figure 1c: Posterior emissions and their spread for the LNLG experiment (assimilating OCO-2 retrievals in Land Nadir and Land Glint mode)
• Figure 1d: Posterior emissions and their spread for the LNLGIS experiment (assimilating traditional in situ measurements and OCO-2 retrievals in Land Nadir and Land Glint mode)
• Figure 1e: Posterior emissions and their spread for the LNLGOGIS experiment (assimilating traditional in situ measurements and all OCO-2 retrievals in all modes)
• Figure 1f: Posterior emissions and their spread for the OG experiment (assimilating OCO-2 retrievals in Ocean Glint mode)

In the OCO-2 model intercomparison leading to this Level 4 product release, bottom-up estimates of fossil fuel were applied directly without optimization. Inverse modelers were asked to estimate optimal fluxes from the ocean and land biosphere. This optimization for each model generally starts with a first guess of the flux component and its uncertainty. The distribution of prior estimates is shown in panel 1a. The first-guess fluxes are subsequently evaluated against atmospheric CO2 observations and/or retrievals. These fluxes are then modified to be consistent with those observational constraints. The optimized results (panels 1b through 1e) have different mean fluxes and uncertainty ranges, indicating the influence of information brought by the observations. The bars extend to the across-model mean flux and the error bar represents one standard deviation of the model suite. The one standard-deviation level of uncertainty for the land flux is shown with a thin green vertical line, and the uncertainty for oceanic exchange is shown with a thin blue vertical line. The net flux as the sum of these three components uncertainty is shown in black. The uncertainty on the net flux (black vertical line) is the sum in quadrature of land biosphere and oceanic flux uncertainties.

---

Figure 2. Annual and six-year mean emissions. The bars in this figure represent annual-mean carbon dioxide emissions for 2015 through 2020, and the 2015-2020 mean in PgC yr-1 from the specified region. Colors and flux components are as in Figure 1. The fluxes are arranged in groups by experiment (Prior, IS, LNLG, LNLGIS, LNLGOGIS, OG), as defined in Figure 1.

Figure 3. Climatological monthly emissions. As in Figure 1, except the bars in this figure represent climatological carbon dioxide emissions for each month in PgC yr-1 from the specified region. OCO-2 models simulated three types of surface-to-amosphere exchange of CO2, each of which is shown in a different color: fossil fuel emissions (tan), terrestrial biosphere flux possibly including implicit fires (green), and air-sea gas exchange (blue). Negative emissions indicate that the flux removes CO2 from the atmosphere, and such sinks have bars that extend below zero. The net surface exchange, computed as the sum of these three components, is shown as a thick black line.

• Figure 3a: Prior emissions and their spread.
• Figure 3b: Posterior emissions and their spread for the IS experiment (assimilating traditional in situ measurements)
• Figure 3c: Posterior emissions and their spread for the LNLG experiment (assimilating OCO-2 retrievals in Land Nadir and Land Glint mode)
• Figure 3d: Posterior emissions and their spread for the LNLGIS experiment (assimilating traditional in situ measurements and OCO-2 retrievals in Land Nadir and Land Glint mode)
• Figure 3e: Posterior emissions and their spread for the LNLGOGIS experiment (assimilating traditional in situ measurements and all OCO-2 retrievals in all modes)
• Figure 3f: Posterior emissions and their spread for the OG experiment (assimilating OCO-2 retrievals in Ocean Glint mode)

In the OCO-2 model intercomparison leading to this Level 4 product release, bottom-up estimates of fossil fuel were applied directly without optimization. Inverse modelers were asked to estimate optimal fluxes from the ocean and land biosphere. This optimization for each model generally starts with a first guess of the flux component and its uncertainty. The distribution of prior estimates is shown in panel 1a. The first-guess fluxes are subsequently evaluated against atmospheric CO2 observations and/or retrievals. These fluxes are then modified to be consistent with those observational constraints. The optimized results (panels 3b through 3e) have different mean fluxes and uncertainty ranges, indicating the influence of information brought by the observations. The bars extend to the across-model mean flux and the error bar represents one standard deviation of the model suite. The one standard-deviation level of uncertainty for the land flux is shown with a thin green vertical line, and the uncertainty for oceanic exchange is shown with a thin blue vertical line. The net flux as the sum of these three components uncertainty is shown in black. The uncertainty on the net flux (black vertical line) is the sum in quadrature of land biosphere and oceanic flux uncertainties.

---

---

Experiment	Year	Fossil Emission	Land	Ocean	Net
Prior	2015	1.46	-0.21 ± 0.70	-0.01 ± 0.01	1.25 ± 0.70
Prior	2016	1.44	-0.29 ± 0.65	-0.01 ± 0.01	1.14 ± 0.65
Prior	2017	1.42	-0.25 ± 0.66	-0.01 ± 0.01	1.16 ± 0.66
Prior	2018	1.46	-0.18 ± 0.71	-0.01 ± 0.01	1.28 ± 0.71
Prior	2019	1.43	-0.23 ± 0.71	-0.01 ± 0.01	1.19 ± 0.71
Prior	2020	1.27	-0.24 ± 0.69	-0.01 ± 0.01	1.03 ± 0.69
Prior	Mean	1.41	-0.23 ± 0.69	-0.01 ± 0.01	1.18 ± 0.69
IS	2015	1.46	-0.27 ± 0.68	-0.01 ± 0.03	1.18 ± 0.68
IS	2016	1.44	-0.41 ± 0.64	-0.01 ± 0.03	1.01 ± 0.64
IS	2017	1.42	-0.54 ± 0.64	-0.01 ± 0.03	0.87 ± 0.65
IS	2018	1.46	-0.41 ± 0.58	-0.01 ± 0.03	1.05 ± 0.58
IS	2019	1.43	-0.59 ± 0.61	-0.01 ± 0.04	0.83 ± 0.62
IS	2020	1.27	-0.59 ± 0.59	-0.00 ± 0.04	0.68 ± 0.59
IS	Mean	1.41	-0.47 ± 0.62	-0.01 ± 0.03	0.94 ± 0.62
LNLG	2015	1.46	-0.50 ± 0.66	-0.01 ± 0.02	0.95 ± 0.66
LNLG	2016	1.44	-0.47 ± 0.60	-0.01 ± 0.02	0.96 ± 0.60
LNLG	2017	1.42	-0.48 ± 0.67	-0.01 ± 0.02	0.94 ± 0.67
LNLG	2018	1.46	-0.40 ± 0.65	-0.01 ± 0.02	1.06 ± 0.65
LNLG	2019	1.43	-0.54 ± 0.66	-0.01 ± 0.03	0.88 ± 0.66
LNLG	2020	1.27	-0.39 ± 0.67	-0.01 ± 0.02	0.87 ± 0.67
LNLG	Mean	1.41	-0.46 ± 0.65	-0.01 ± 0.02	0.94 ± 0.65
LNLGIS	2015	1.46	-0.36 ± 0.62	-0.01 ± 0.03	1.09 ± 0.63
LNLGIS	2016	1.44	-0.43 ± 0.54	-0.01 ± 0.03	1.00 ± 0.54
LNLGIS	2017	1.42	-0.55 ± 0.60	-0.00 ± 0.03	0.86 ± 0.60
LNLGIS	2018	1.46	-0.37 ± 0.57	-0.00 ± 0.03	1.10 ± 0.57
LNLGIS	2019	1.43	-0.50 ± 0.59	-0.00 ± 0.04	0.92 ± 0.59
LNLGIS	2020	1.27	-0.47 ± 0.57	-0.00 ± 0.03	0.80 ± 0.57
LNLGIS	Mean	1.41	-0.44 ± 0.58	-0.00 ± 0.03	0.96 ± 0.58
LNLGOGIS	2015	1.46	-0.37 ± 0.63	-0.01 ± 0.03	1.09 ± 0.63
LNLGOGIS	2016	1.44	-0.53 ± 0.55	-0.01 ± 0.03	0.90 ± 0.55
LNLGOGIS	2017	1.42	-0.57 ± 0.58	-0.01 ± 0.03	0.85 ± 0.58
LNLGOGIS	2018	1.46	-0.42 ± 0.60	-0.00 ± 0.03	1.04 ± 0.61
LNLGOGIS	2019	1.43	-0.59 ± 0.58	-0.00 ± 0.03	0.83 ± 0.58
LNLGOGIS	2020	1.27	-0.53 ± 0.55	-0.00 ± 0.03	0.74 ± 0.55
LNLGOGIS	Mean	1.41	-0.50 ± 0.58	-0.00 ± 0.03	0.91 ± 0.58
OG	2015	1.46	-0.22 ± 0.72	-0.01 ± 0.03	1.23 ± 0.72
OG	2016	1.44	-0.26 ± 0.73	-0.01 ± 0.02	1.17 ± 0.73
OG	2017	1.42	-0.47 ± 0.83	-0.01 ± 0.02	0.94 ± 0.83
OG	2018	1.46	-0.28 ± 0.79	-0.01 ± 0.03	1.18 ± 0.79
OG	2019	1.43	-0.40 ± 0.79	-0.01 ± 0.03	1.01 ± 0.79
OG	2020	1.27	-0.28 ± 0.80	-0.01 ± 0.03	0.99 ± 0.80
OG	Mean	1.41	-0.32 ± 0.78	-0.01 ± 0.03	1.09 ± 0.78

Table 1. Annual and six-year mean emissions. Fluxes in PgC yr-1 from the specified region. Annual means for 2015 through 2020 are listed, as well as the six-year mean, by terrestrial biosphere ("land") air-sea gas exchange ("ocean") and their sum ("net"). The fluxes are arranged in groups by experiment (Prior, IS, LNLG, LNLGIS, LNLGLOGIS, OG), as defined in Figure 1. Uncertainties are the 1 standard deviation spread in the model ensemble.

---