the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Synthetic mapping of XCO2 retrieval performance from shortwave infrared measurements: impact of spectral resolution, signal-to-noise ratio and spectral band selection
Abstract. Satellites have been providing spaceborne observations of the total column of CO2 (noted XCO2) for over two decades now and, with the need for independent verification of Paris Agreement objectives, many new satellite concepts are currently planned or being studied to complement or extend the already existing instruments. Depending on whether they are targeting natural and/or anthropogenic fluxes of CO2, the design of these future concepts vary greatly. The characteristics of their shortwave infrared (SWIR) observations notably explore several orders of magnitude in spectral resolution (from λ/Δλ~400 for Carbon Mapper to λ/Δλ~25000 for MicroCarb) and include different selections of spectral bands (from one to four bands, among which the CO2-sensitive 1.6 µm and/or 2.05 µm bands). Besides, the very nature of the spaceborne measurements is also explored: for instance, the NanoCarb imaging concept proposes to measure CO2-sensitive truncated interferograms, instead of infrared spectra as other concepts, in order to significantly reduce the instrument size. This study synthetically explores the impact of three different design parameters on XCO2 retrieval performance, as obtained through Optimal Estimation: (1) the spectral resolution; (2) the signal-to-noise ratio (SNR) and (3) the spectral band selection. Similar performance assessments are completed for the exactly-defined MicroCarb, Copernicus CO2 Monitoring (CO2M) and NanoCarb concepts. We show that improving SNR is more efficient than improving spectral resolution to increase XCO2 precision when perturbating these parameters across two orders of magnitude, and that low-SNR and/or low spectral resolution yield XCO2 with vertical sensitivities giving more weight to atmospheric layers close to the surface. The exploration of various spectral band combinations illustrates, especially for lower spectral resolutions, how including an O2-sensitive band helps to increase optical path length information, and how the 2.05 µm CO2-sensitive band contains more geophysical information than the 1.6 µm band. With very different characteristics, MicroCarb shows a CO2 information content only slightly higher than CO2M, which translates into lower XCO2 random errors, by a factor ranging from 1.1 to 1.9 depending on the observational situation. The NanoCarb performance for a single pixel of its imager compares to concepts that measure spectra at low-SNR and low-spectral resolution but, as this novel concept would observe a given target several times during a single overpass, its performance improves when combining all the observations. Overall, the broad range of results obtained through this synthetic XCO2 performance mapping hints at the future intercomparison challenges that the wide variety of upcoming CO2-observing concepts will pose.
- Preprint
(2654 KB) - Metadata XML
-
Supplement
(3965 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on amt-2023-233', Anonymous Referee #1, 19 Dec 2023
Review of “Synthetic mapping of XCO2 retrieval performance from shortwave infrared measurements: impact of spectral resolution, signal-to-noise ratio and spectral band selection” by Dogniaux and Crevoisier, amt-2023-233
The present manuscript presents a simulation study of the impact of spectral resolution, spectral band selection and signal-to-noise (SNR) ratio on XCO2 retrieval performance for space borne passive spectrometers in the shortwave infrared, with a focus on the CO2M, MicroCarb and NanoCarb sensors. A large part of aforementioned parameter space is explored synthetically by simulating top-of-atmosphere radiances with the 5AI radiative transfer model and then operating an optimal estimation based inversion on the synthetic measurements to discuss XCO2 retrieval performance with respect to random errors, vertical sensitivity and information content among others. The paper addresses relevant questions within the scope of AMT and presents new data that warrant publication. The language is fluent, the authors give credit to related work and substantial conclusions are reached. However, I believe that the manuscript could be improved by addressing some methodological questions which should strengthen the interpretation of the results.
General comments:
G1: Recent studies have shown that coarse-resolution SWIR instruments tend to be affected by strong trace gas biases induced by surface albedo features (e.g. for XCO2 or XCH4). I feel that this is not sufficiently addressed in your discussion of retrieval performance as a function of spectral resolution and SNR (section 4.1), in the discussion of information entanglement (section 4.2.3) and your observations regarding XCO2 precision as a function of Degrees of Freedom for Signal (DoFS): In your synthetic study it may be true that precision can be easily “bought” at the expense of spectral resolution by increasing SNR. However, in the real world, for resolving powers below ~1000 I would expect that this stops being true. It would probably go beyond the scope of this paper to include a more realistic albedo model in your framework, but it sure would be interesting. You should at least discuss this more in-depth throughout the manuscript (e.g. near Figure 3, etc.).
G2: The fact that turning the aerosol fit off produces strong albedo correlations for NanoCarb indicates to me that you might be overfitting and/or need to adjust the retrieval scheme for NanoCarb. Perhaps the aerosol parameters are simply within the null space. If we think of NanoCarb as a low resolving power spectrometer, it seems physically very plausible that aerosol and albedo parameter fits would strongly impact each other. Please double check the retrieval and discuss this observed behavior. Which retrieval setting is recommended for NanoCarb? How do the retrieved values and standard deviations of XCO2 and albedo change when you turn the aerosol fit off and on?
G3: How does the above observation extend to the CVAR experiment? Do you fit aerosol parameters across the spectral-resolution space that you cover? If yes, would it not make sense to turn off the fine mode parameter at some low resolving power (see Fig. 6)? I am surprised that you appear to find sufficient DoFS for the coarse mode parameter even at a resolving power of 200 - how do you explain the very different behavior of the two aerosol retrieval parameters?
G4: As far as I can tell, the paper currently does not address the accuracy of XCO2 retrievals as a function of SNR, resolving power and band selection. I believe it would be worthwhile to include this.
Minor comments:
M1: In Figure 1, it would be helpful to indicate the FWHM of the band-pass filters of NanoCarb and remind the reader of the FWHM of the NanoCarb bands.
M2: Section 3.1: Please explain in the text why you confine aerosol particles to the lowermost 4 km of the atmosphere and add details on what kind of aerosol size distributions you used.
M3: Section 3.1: The albedo values you chose for the SWIR seem pretty high in my opinion. Is this really the value over all ASTER samples?
M4: Figure S1: It would be useful to magnify the albedo Jacobian for NanoCarb, especially in the context of albedo biases and aerosol retrievals (see above).
M5: Page 7, line 147: add reference for CO2M XCO2 random error value of 0.7 ppm
M6: Page 9, line 189: add reference for the “latest design of the NanoCarb instrument”
M7: Section 2.3: Mention a few more details on NanoCarb so the reader doesn’t have to go to other references to find out about the ground sampling distance, swath, etc.
M8: Section 3.2.2: Why did you use the GEISA 2015 database as opposed to HITRAN (and as opposed to a newer version)?
M9: Section 3.2.2: How about the treatment of the 1.27 um band and airglow emission? Did you leave it out or did you model it for MicroCarb?
M10: Table 4: To which parameter does the following statement refer? “Not included in the state vector for NanoCarb”. Please clarify.
M11: Page 14, lines 311-313: The XCH4 biases in Borchardt et al. (2021) are caused by albedo interferences, not by the prior information. Please reword.
M12: Figure 8: This is related to G1. Your albedo perturbation is not doing anything to the retrieved XCO2 under any scenario. I think you should introduce a linear or periodic perturbation and observe the effect. This would be a very interesting experiment!
M13: Figure 3: Maybe add a “disclaimer” to the caption, explaining that NanoCarb is symbolically added at very low resolving power (as in Page 23, lines 503-504).
M14: Page 27, lines 607-608: I am not convinced that this statement is generally true across the resolving power space you explore. Potentially reword.
M15: I find that you could try to summarize your results more concisely in the conclusion: e.g. The SNR-resolving power-space holds very little performance gain for sensors with two SWIR bands, the 2 um CO2 band carries higher XCO2 and aerosol DoFS than the 1.6 um band, etc…
Technical comments:
T1: check entire reference list for spelling errors and typesetting issues
T2: Abstract, line 22: perturbating -> perturbing
T3: Page 2, line 35: …carbon cycle compares results… -> …carbon cycle is based on comparisons of results…
T4: Page 2, line 44: pioneer -> pioneering
T5: Page 2, line 59: gather -> are responsible for
T6: Page 3, line 64: atmospheric schemes -> atmospheric inversion schemes ?
T7: Page 7, line 165: measure a continuous spectra -> measure continuous spectra
T8: Page 11, line 246-247: A also enables to compute the … -> A also enables computation of …
T9: Page 11, line 255: layers -> levels
T10: Page 15, line 336: “which sensitivities also correlate” reword
T11: Page 23, line 516: variable Jacobians -> Jacobian variables ?
Citation: https://doi.org/10.5194/amt-2023-233-RC1 -
RC2: 'Comment on amt-2023-233', Anonymous Referee #2, 21 Dec 2023
The paper “Synthetic mapping of XCO2 retrieval performance from shortwave infrared measurements: impact of spectral resolution, signal-to-noise ratio and spectral band selection” provides detailed analysis of the characteristics of CO2 measurements with SWIR spectrometers with a range of resolutions, signal to noise characteristics, and spectral bands. Characteristics of actual spectrometer designs (CO2M, MicroCarb, NanoSat), as well as a large set of hypothetical instrument configurations are considered.
The paper reports on detailed analysis of the XCO2 precision, the degrees of freedom for CO2, the vertical sensitivity, the sensitivity for and possible interference due to parameters such as temperature, water vapor, albedo, and aerosols. A wide range of scenarios (or situations) are explored, where surface reflectance and solar geometry are systematically changed.
Overall, this paper is very well constructed. The experiments and assessments are carefully structured and the key findings are clearly described. The graphics are effective, the completeness of the analysis is impressive, and the writing is clear.
General review comments:
Overall, this is an impressive and comprehensive piece of work. It can serve as a reference for instrument developers as they seek to optimize performance and evaluate the trade space of resolution, signal to noise, and band pass. The methodology is clearly described, including the input data and calculations that are performed.
One weakness I find in this paper is the treatment and description of the CO2M instrument. There should be language included to clarify that this work is assessing just the spectrometer element of CO2M, which will also integrate a multi-angle polarimeter. The assessment of XCO2 precision and error related to aerosols is a correct analysis for the CO2M spectrometer alone, but not the planned CO2M mission. I would suggest that this point is made clear at the beginning, and perhaps they use the phrase CO2M spectrometer in the paper.
To gain confidence in the methodology, it would be useful for the authors to point out where OCO-2 and GOSAT(-2) are in these SNR/resolution plots, and to compare to published results for precision, DOF, etc. I suggest this because sources of error such as spectroscopic mischaracterization or errors in instrument characterization are not well captured in the analysis presented here, yet may be important contributors to error. The mismatch for actual missions may provide some insight into the errors not captured in this analysis.
Specific comments:
1) Lines 210 – use soot and minerals as their aerosols – what justifies these choices? For many parts of the world, these are not representative. How do the absorption and scattering characteristics impact the results?
2) The paragraph that starts at line 296 discusses Figure 3. The authors use the word “break”. I think the changes in slope of these lines is not all that significant, so break is not a good choice of phrasing. I would suggest a phrase like “change of slope”
3) Sections 4.1 and 4.2 could have a short introductory paragraph to introduce the structure of the subsections that follow.
4) Line 330: The starting sentence of this section (4.2.1) is nearly the same text as is used to start section 4.1.1
- On line 270
270 For the atmospheric situation VEG-50o, Figure 3 shows the 𝑋CO2 precision (or random error and degrees of freedom (hereafter DOFs) as a function of both the resolving power 𝜆/Δ𝜆 and the signal-to-noise ratio (SNR) for CVAR, and for the exact CO2M, MicroCarb and NanoCarb concepts (results for exactly-defined concepts are discussed in Sect. 5)
- Line 330
For the atmospheric situation VEG-50o[…], Figure 5 shows the 𝑋CO2 precision and DOFs as a function of both the resolving power 𝜆/Δ𝜆 and spectral band selection for CVAR (with SNR fixed at its reference value), and for the exact CO2M, MicroCarb and NanoCarb concepts (results for exactly-defined concepts are discussed in Sect. 5).
- To address this, Section 4.2.1 could have a sentence to first introduce the focus of the analysis. Perhaps “In this section we assess the impacts of changing the spectral bands”. (and section 4.1.1 could be introduced with “ Here we look at SNR impacts on precisions and DOF.
5) Lines 382 and following: I find this language to be very convoluted, and suggest a rewrite.
Currently “While methodologies are hardly comparable (because this study is only based on synthetic simulations), both works agree that a sharp change in how 𝑋CO2 precision evolves with resolving power is to be expected around 𝜆/Δ𝜆 = 1000 – 2000, when solely using the 1.6 or 2.05 μm CO2 bands”
Suggest:
“While methodologies are hardly comparable (because this study is only based on synthetic simulations), both works agree that the 𝑋CO2 precision and resolving power relationship has a change of characteristic around 𝜆/Δ𝜆 = 1000 – 2000, when solely using the 1.6 or 2.05 μm CO2 bands”
6) Figure 11 – I can not differentiate the colors of MicroCarb B1234 and NanoCarb comp.
7) Line 608 – The phrase “more easily gained” implies that we just need to make higher SNR instruments and we can easily get better precision. But this paper just studies the sensitivities. I would suggest rephrasing to “Overall, precision is more sensitive to SNR improvements than spectral resolution improvements.”
Editorial comments:
Line 525 – the word Temperature is capitalized mid-sentence.
Lines 673, 674, 688, 716, 719 and 720 (and maybe others) – formatting issues in the references – looks like latex formats not properly converted???
Citation: https://doi.org/10.5194/amt-2023-233-RC2 -
RC3: 'Comment on amt-2023-233', Anonymous Referee #3, 05 Jan 2024
General comments
In this paper, the authors perform a quantitative study assessing the estimated performance of a hypothetical shortwave infrared (SWIR) CO2 satellite instrument, considering the impact of a range of instrument design parameters: spectral resolution, signal-to-noise ratio (SNR), and spectral band selection. They achieve this by applying an optimal estimation retrieval algorithm to synthetic spectra generated assuming a wide range of fictitious instrument concepts, defined by varying each of these parameters, and a number of different observation scenarios. In addition, they apply the same performance assessment framework to some ready-defined future mission concepts – MicroCarb, CO2M, and NanoCarb – providing useful context for the hypothetical concept assessment study. This paper is timely given the wide interest in new methodologies for measuring CO2 emissions, driven by the need to independently verify Paris Agreement objectives, which are likely to include satellite remote sensing as a significant component. There are some particularly interesting conclusions which should help inform the conception and design of future SWIR CO2 satellite missions, namely the relative importance of improving SNR vs. resolving power in order to improve XCO2 precision, the importance of including an O2 absorption band in a mission concept to account for aerosol absorption, and the sensitivity of low SNR and resolving power instrument concepts to a priori mis-knowledge. I think that this paper is suitable for publication in Atmospheric Measurement Techniques, and have a few suggestions for improvements which will hopefully help strengthen the paper’s conclusions further.
Specific comments
- As mentioned above, the inclusion of ready-defined mission concepts provides useful context for the fictitiously varying CO2M (CVAR) concept study. I think that the paper overall would benefit by also considering an existing mission – OCO-2 for example – along with the ready-defined future missions already included. This would provide additional context for the CVAR study by comparing their performances alongside the current “state-of-the-art”, whilst also demonstrating that the assumed observation scenarios and the forward and inverse setups produce realistic results when compared with real observational data;
- I think some further justification/clarification would be useful for the atmospheric situations used in the study. For example, are the temperature and water vapour profiles from the TGIR climatology representative of the current climate? Similarly, I think it would strengthen the conclusions if a realistic profile of CO2 concentration were used instead of a constant profile, especially given that the study considers the vertical sensitivity of the instrument concepts;
- Whilst this study does not explicitly consider spatial resolution, I think it would be worth commenting on the implications of some of the conclusions on the feasibility of CO2 imaging concepts, which trade off reduced SNR and/or resolving power in favour of high spatial resolution in order to be able to quantify emissions from ever-smaller plumes of CO2 emitted by point sources. To pick one example from the results in Section 4, Figure 9 shows how concepts with low resolving power would be quite sensitive to a priori mis-knowledge of aerosol optical depths, depending on the spectral band selected and whether an O2 absorption band is incorporated into the instrument concept. Further investigation looking at the ability of SWIR hyperspectral imagers to image emissions plumes and infer CO2 emission rates, using the performance assessment framework described here across a range of instrument parameters including spatial resolution would be very interesting, but I appreciate that would be beyond the scope of this study.
Technical corrections
Line 232: replace “Its” with “It is”;
Line 243: replace “degree” with “degrees”;
Line 372: please provide a reference for the “usual” hypothesis that aerosol properties are fixed across spectral bands;
Line 493: replace “MC123” with “MC234”.
Citation: https://doi.org/10.5194/amt-2023-233-RC3
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
154 | 49 | 15 | 218 | 17 | 10 | 8 |
- HTML: 154
- PDF: 49
- XML: 15
- Total: 218
- Supplement: 17
- BibTeX: 10
- EndNote: 8
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1