the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 Sep 2026
Feasibility study of atmospheric carbon dioxide satellite retrievals over snow in the context of the Copernicus Anthropogenic CO2 Monitoring Mission
Abstract. Northern high latitudes pose significant challenges to reliable space-based observations of total column carbon dioxide (XCO2). In addition to large solar zenith angles and frequent cloud coverage over the Arctic and boreal regions, snow-covered surfaces absorb strongly in the shortwave-infrared wavelengths coinciding with the CO2 absorption channels used by several current missions, e.g., Japanese Greenhouse Gases Observing Satellite and Nasa Orbiting Carbon Observatory-2, and the upcoming Copernicus Anthropogenic CO2 Monitoring Mission (CO2M). Because of the resulting low radiances of the reflection measured by the satellite instruments, retrievals over snow may be less reliable and, for current missions, are typically filtered or flagged for potentially poor quality. In this work, we present the first feasibility study dedicated to XCO2 retrievals over snow. We introduce a measurement-based snow reflectance model, develop this into a kernel format following the Ross-Thick-Snow (RTS) kernel formulation, and use this to simulate CO2M radiances. Furthermore, we also present a sample of simulated XCO2 retrievals for the CO2M with the University of Leicester Full Physics (UoL-FP) retrieval framework, considering a variety of solar zenith angles as well as nadir and specular reflection (glint) observing geometries, the two planned observation modes of CO2M. The results indicate the advantages of using a snow kernel model in the retrievals of XCO2, in terms of improved convergence. Glint-mode observations are shown to lead to higher signal-to-noise ratios and improved retrievals in terms of reduced errors compared to the nadir geometry in the Northern high latitudes over snow. The signal-to-noise ratio increases towards forward-scattering direction, which indicates that the exact specular reflection (glint) geometry is not required but forward scattering directions in general have advantages at large solar zenith angles. However, we also find that constraining the snow reflectance model is essential for the retrieval improvements. The results show the potential to significantly increase the number of observations at high latitudes and widen the seasonal coverage by approximately 1-3 months. The relevance of a potentially increased coverage does not limit to CO2M but extends to other current and future CO2 missions as well as other atmospheric constituents. Ultimately, the increased coverage may contribute to the quantification and an improved understanding of high-latitude CO2 sources and sinks, especially in the late winter and spring seasons where a reliable quantification of the XCO2 seasonal cycle is essential.
- Preprint
(2550 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-2722', Anonymous Referee #1, 11 Oct 2026
-
RC2: 'Comment on egusphere-2026-2722', Anonymous Referee #2, 20 Oct 2026
The manuscript addresses an important and timely problem for satellite remote sensing of greenhouse gases at northern high latitudes. Its investigation of BRDFs, forward scattering geometries, and their effects on simulated retrievals could make a useful contribution to mission planning and high latitude observing systems for carbon cycle. The manuscript demonstrates the potential for improved retrievals under selected idealized conditions, but does not yet establish that the proposed approach will deliver the claimed increase in reliable, unbiased satellite observations under realistic conditions. The authors should address the following comments before the manuscript can be considered for publication.
1. The manuscript develops an RTS kernel model from measurements and extrapolates it across angular regions that are sparsely sampled. The authors explain that the kernel formulation was preferred over Legendre polynomials because it behaves more robustly during extrapolation. However, smooth and physically plausible extrapolation does not, by itself, demonstrate that the extrapolated BRDF is accurate. Any systematic error in the magnitude or angular distribution of this enhancement could directly affect simulated radiances. I recommend authors quantify the quality of the BRDF fit, assess extrapolation uncertainty, and establish robustness of the kernel choice.
2. The authors should also discuss the transferability of the three snow classes. Explain how the selected measurements represent snow conditions in boreal forests, tundra, mountainous terrain, and open snow-covered surfaces.
3. How do the vegetation, exposed ground, patchy snow, surface roughness, and topography might alter the effective satellite-scale BRDF? A quantitative sensitivity experiment with a simple mixed snow vegetation surface would strengthen this manuscript.
4. The retrieval experiments may overstate the performance because the forward model and retrieval use the same BRDF assumptions. While the surface mismatch experiment tests only one direction of snow-type mismatch. The authors should evaluate bidirectional mismatches across all three snow types and both viewing geometrics, quantify sensitivity to plausible BRDF parameter uncertainties, and justify retrieving only first kernel parameter while holding the other coefficients fixed.
Also, clarify the representativeness of the imposed 10 ppm CO2 perturbation. These analyses would better establish the robustness of the reported retrieval improvements.
5. The reported biases of up to 2-3 ppm under incorrect BRDF assumption are important for the atmospheric CO2 applications. The authors should distinguish posterior precision, systematic bias, and overall accuracy throughout the manuscript.
6. The results indicate that glint geometry increases radiance and can improve retrieval performance, particularly at larger solar zenith angles. However, the manuscript should more clearly distinguish the idealized geometry used in the simulations from the viewing geometries that CO2M can practically acquire over land. The authors should explain how glint is defined the simulations, and how viewing and relative azimuth angles affect the results.
7. The manuscript should more clearly discuss demonstrated improvements in simulated radiances and XCO2 retrievals from the potential downstream benefits for biospheric flux inversions and anthropogenic emission monitoring. Please temper the claims in the abstract and conclusions accordingly.
8. The manuscript uses both a single-scattering model and a multiple-scattering model for different experiments. Please explain which conclusions depend on the single-scattering approximation and how atmospheric multiple scattering could affect the radiance and retrieval results, particularly at large solar zenith angles.
9. The authors note that expanding retrievals over snow can benefit downstream emission monitoring. The authors should also briefly discuss how this work can enable high-latitude regional and urban scale emission monitoring.
Relevant citation: Hamilton, S. D., Wu, D., Johnson, M. S., Turner, A. J., Fischer, M. L., Dadheech, N., & Jeong, S. (2024). Estimating carbon dioxide emissions in two California cities using Bayesian inversion and satellite measurements. Geophysical Research Letters, 51(20), e2024GL111150.
Citation: https://doi.org/10.5194/egusphere-2026-2722-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 370 | 125 | 51 | 546 | 56 | 59 |
- HTML: 370
- PDF: 125
- XML: 51
- Total: 546
- BibTeX: 56
- EndNote: 59
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study is motivated by a current low throughput of good quality data over snow-covered surfaces, for XCO2 retrievals in commonly used spectral bands, near 1.6 and 2 microns. Measurements at high latitudes are critical for CO2 monitoring because the region is rapidly evolving in a warming climate. A new snow reflectance model is proposed and tested with simulated CO2M radiances. The simulation experiments show significant signal enhancements over snow in glint geometry compared to nadir in the CO2M SWIR bands, and comparatively moderate enhancements in the NIR band. It also presents the XCO2 bias in response to the wrong choice of snow type BRDF, and to a perturbation in the a priori XCO2. The results presented here are significant for any remote sensing mission and make a strong case for prioritizing glint measurements over snow. The manuscript should be published after the following comments are addressed:
Major comments:
1 - The main drawback of the study is the lack of testing on real satellite observations to confirm the effectiveness of the proposed snow model. The paper being motivated by CO2M is only circumstantial, and the proposed snow reflectance model should improve retrievals over snow near glint regardless of the observing platform. There are OCO-2 land glint observations over snow, and likely chance occurrences in GOSAT observations despite not having a dedicated glint mode over land. The paper’s conclusions would be significantly strengthened if it could show one example application with real spectra from one of these instruments.
2 - A second drawback, which would be mitigated by addressing the first, is the limited parameter space considered for the simulation experiments. Although it is always easy to criticize a simulation experiment setup, comparing nadir to pure glint with VZA=SZA and a relative azimuth angle of 180°, which will rarely occur, may not be representative of real measurements, and may not include the best-case scenario. It would be valuable to quantify performance within 30° of the specular reflection. From Fig. 5 it seems like you would not lose much signal ~30-45 degrees around direct glint, but given aerosols like sulphate and sea salt have a narrow forward phase function peak, observing away from exact glint (increasing the scattering angle) could reduce the scattered light reaching the instrument, and also reduce airmass. Sampling where SNR is maximized and scattered light minimized may result in an optimal angle not at exact glint. At least, a discussion of the limitations of the chosen parameter space would improve the manuscript.
3 - You could discuss how you would detect which snow type BRDF model to select for the retrievals, would you need 3rd party data or make the determination based on the radiance level? You showed the second option might be difficult in the presence of thin ice clouds.
Minor comments
Fig1: the color bar should be discrete with 3 colors instead of continuous.
Eq1: you have not stated that R is the reflectance factor.
L145: give the short commit hash of the version used for the simulation in the paper (eg. e1cd23e if it is the current last commit). Consider making a github release and assigning it a DOI on a service like Zenodo.
L157: “O2-O2”, state so if you mean collision-induced absorption. Including it amongst the “vertically varying distributions” of trace gases sounds like it implies there is a “O2 dimer” profile separate from the O2 profile. And do you not account for O2-N2 CIA?
L160: cite https://doi.org/10.5194/amt-16-1121-2023 for the TCCON priors and/or https://doi.org/10.5194/essd-16-2197-2024 for TCCON/GGG2020.
Section 3.4: state if you do or do not add noise to the simulated spectra.
Table 1: add the total number of simulations resulting from the parameter space.
L226: why do you only retrieve fiso?
Fig3: The “(b)” label overlaps with the right panel. The figure could be regenerated with higher quality as the labels are not sharp, Fig A1 has the same quality issue.
Fig4: missing the vertical axis label. Is the high-frequency variability between 1.75-2 microns a measurement artifact?
Fig B1 and B2: comparing the cases with / without the ice cloud would be easier if each panel in B2 was sharing its vertical axis with the corresponding panel in B1. What causes the difference between the solid and dotted lines in B1 with no ice cloud? Rayleigh scattering? Is there an aerosol profile in these simulations?
L430 says B1 and B2 show the single- and multiple-scattering radiances. But the figure captions only mention surface reflectance contribution and total radiance. Is there an aerosol in B1 or is the difference between the solid and dotted line from Rayleigh scattering? Is B1 using Raysca while B2 and Fig8 use Siro? It was difficult to follow which model is used when, it may be good for clarity to specify it in the caption of each figure showing simulated radiances.
Many references are missing DOIs eg., Baum et al 2014 (https://doi.org/10.1016/j.jqsrt.2014.02.029), Hannula et al 2020 (https://doi.org/10.5194/essd-12-719-2020), check all references.
The Meijer et al 2020 reference should include the document’s issue and reference numbers (Issue 3.0; Ref EOP-SM/3088/YM-ym)
Style / typos
L168: “convolved using a Gaussian slit” => Gaussian instrument line shape / Gaussian slit function
L203: “passed through the instrument model” => convolved with the instrument line shape (unless it means something more, in which case state that too)
Fig B1 and B2 appear after the acknowledgements