the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Apr 2024
A global perspective on CO2 satellite observations in high AOD conditions
Abstract. Satellite-based observations of carbon dioxide (CO2) are sensitive to all processes that affect the propagation of radiation in the atmosphere, including scattering and absorption by atmospheric aerosols. Therefore, accurate retrievals of column-averaged CO2 (XCO2) benefit from detailed information on the aerosol conditions. This is particularly relevant for future missions focusing on observing anthropogenic CO2 emissions, such as the Copernicus Anthropogenic CO2 Monitoring mission (CO2M). To fully prepare for CO2M observations, it is informative to investigate existing observations in addition to other approaches. Our focus here is on observations from the NASA Orbiting Carbon Observatory -2 (OCO-2) mission. In the operational full-physics XCO2 retrieval used to generate OCO-2 level 2 products, the aerosol properties are known to have high uncertainty but their main objective is to facilitate CO2 retrievals. We evaluate the OCO-2 product from the point of view of aerosols by comparing the OCO-2 retrieved aerosol properties to collocated Moderate Resolution Imaging Spectro-radiometer (MODIS) Aqua Dark Target aerosol products. We find that there is a systematic difference between the aerosol optical depth (AOD, τ) values retrieved by the two instruments, such that τOCO−2 ∼ 0.4τMODIS. We also find a dependence of the XCO2 on the AOD difference, indicating an aerosol-induced effect in the XCO2 retrieval. In addition, we find a weak but statistically significant correlation between MODIS AOD and XCO2, which can be partly explained by natural covariance and co-emission of aerosols and CO2 but is partly masked by the aerosol-induced XCO2 bias. Furthermore, we find that issues in the OCO-2 aerosol retrieval may lead to misclassification of the quality flag for a small fraction of OCO-2 retrievals. Based on MODIS data, 4.1 % of low AOD cases are incorrectly classified as high AOD (low quality) pixels, while 16.5 % of high AOD cases are erroneously classified as low AOD (high quality) pixels. Finally, we investigate the effect of an AOD threshold on the fraction of acceptable XCO2 data. We find that relaxing the MODIS AOD threshold from 0.2 to 0.5 (at 550 nm), which is the goal for the CO2M, increases the fraction of acceptable data by 14 percentage points globally, and by 31 percentage points for urban areas.
- Preprint
(6351 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on amt-2024-77', Anonymous Referee #1, 28 May 2024
The Virtanen et al manuscript entitled “A global perspective on CO2 satellite observations in high AOD conditions” tries to evaluate the OCO-2 product from the point of view of aerosols by the combined analysis of the OCO-2 retrieved XCO2 and AOD, collocated MODIS AOD, and TCCON retrieved XCO2. This manuscript found a systematic difference between AOD retrieved by OCO-2 and MODIS, which can impact the retrieval of XCO2. The dependence of the OCO-2 retrieved XCO2 on the AOD difference was also found. But toward the future CO2M mission, in my opinion, besides simple data coverage variation with different AOD threshold, you should also discuss the quality control with the AOD threshold of 0.5.
Major comments:
You only analyzed the data coverage variation after adjusting the AOD threshold from 0.2 to 0.5. According to your analysis of the relationship between OCO-2 retrieved XCO2 and AOD, the accuracy of the former significantly depend on the uncertainty of the later. You should also discuss how to conduct quality filter, or what the data quality will be for CO2M, which will use 0.5 as AOD threshold.
The specific comments are listed below:
P2L24: Change the order of the sentences “An essential monitoring component will be the Copernicus Anthropogenic CO2 Monitoring Mission (Meijer et al., 2023).” And “While ground-based greenhouse gas measurements are mainly 25 available in developed countries – with limited coverage and representativeness – satellite-based XCO2 information will be irreplaceable in areas where ground-based measurements are not made.”
P2L33: Check the format of the citations “(e.g., (Houweling et al., 2015; Crowell et al., 2019))”, should they be “first name et al. (year)”?
P3L68: “In the collocated database we include only a limited selection of OCO-2 data fields, but the sounding ID in the combined daily files is equivalent with the original OCO-2 lite files, allowing addition of more data fields in an effective manner.” It makes me confused. Please further explain it. What is “sounding ID”? Why can you get “more data fields”?
P3L70: “The aerosol parameters of the ACOS algorithm include five scatterers, two cloud types (water and ice), two tropospheric aerosol types and a stratospheric aerosol type (sulfate).” This may should be revised as “The aerosol products of the ACOS algorithm include parameters from five scatterers”. Is “five scatterers” refer to “two cloud types (water and ice), two tropospheric aerosol types and a stratospheric aerosol type (sulfate)” If so, this may should be revised as “five scatterers, which are two cloud types (water and ice), two tropospheric aerosol types and a stratospheric aerosol type (sulfate).”
P3L72: There are five aerosol types mentioned in the brackets. Why did you write “Two representative types of tropospheric aerosols”?
P3L74: “Atmospheric Carbon Observations from Space (ACOS)”. Only use abbreviation is ok.
P3L73: “From the large number of data products…”.
P4L86: Add last accessed date after the website.
P4L106: What are “data fields which are not relevant for this study”?
P4L120: Why did you use latitude and longitude as the collocation criteria? Using this criterion, the distance between OCO-2 and TCCON is different at different latitudes.
P4L128: 0.1◦ is also the latitude and longitude threshold? Please describe it. Why did you not use the same spatial and temporal criteria with TCCON?
P4L129: “The OCO-2 observations are not spatially averaged.” Not necessary for AERONET part.
P6L148: What is the specific definition of XCO2 anomaly? How to calculate XCO2 anomaly from the median XCO2? The anomaly was calculated from all data, or for every year? After the calculation, have you removed the anomaly or how to use the anomaly? And you said “This is an alternative way to de-trend the data”, when did you use the LTC method and when did you use the XCO2 anomaly? Please supplement this section with more information.
P6L167: The dust loads may be identified by the areas, but “biomass burning aerosols increase AOD in central Africa and South-East Asia” lack of evidence.
P6L172: I think the largest differences appear in Central Asia and South Asia.
P6L173: “These positive difference values are related to the MODIS DT algorithm permitting small negative AOD values.” Any references?
P7 Fig.1: The colorbar of panel (b) should be adjusted because 0 and invalid data are both white.
P8L187: You should point out earlier.
P8L200: You acquired 770 nm AOD by averaging 675 nm AOD and 870 nm AOD. Can you really achieve that by the simple average? Please give more evidence or reference.
P8L210: How did you define urban areas?
P9L217: Please point out the spatial resolution of OCO-2 again.
P10L250: How did you handle with the AOD exactly equal to 0.2?
P13Figure 4 (b): You should distinguish 0 values and invalid values. Both of them are white.
P13L299: Table A3 or Figure A3?
P14L316: What does “the measured CO2 absorption is divided into too short distance” mean?
P15L340: Again, if you aquire OCO-2 XCO2 anomaly by calculating the difference between the OCO-2 XCO2 data and the median value along 500 km orbit, how can the anomaly be used to de-seasonalize and de-trend the data? Please describe more details about this.
P15Table2: “TCCON(1)” should be revised as “OCO-2(1)”?
P16L365: “OCO-2 seems to slightly overestimate XCO2 for low AOD values, and underestimate at high AOD values.”
-For panel (a) of Figure A8, you used MODIS AOD and XCO2 difference (deviation of OCO-2 from TCCON). MODIS AOD and TCCON XCO2 can be considered as references, so you can definitely get the statement about overestimate or underestimate of OCO2 XCO2 at different AOD.
-For panel (c), it seems that the XCO2 difference is close to zero when the AOD difference is in the range of 0 to 0.1, and when the AOD difference becomes higher or lower, the XCO2 difference will be minus, which means the underestimate of OCO2 XCO2. It seems to be not totally consistent with the statement of Figure 5 “If the aerosol load is underestimated in the retrieval (Q2), the light path is also underestimated, and the measured CO2 absorption is divided into too short distance, leading to overestimation of XCO2. Similarly, if AOD is overestimated, the light path is also overestimated, causing underestimation of XCO2.” Do you have any comments on it?
P18L388: The XCO2 and AOD also show negative correlation in North America and Europe. Do you have any comments on this?
P19L395: “Satellite XCO2 retrievals are known to have higher uncertainty in high aerosol conditions.” Where is this statement summarized from? From Figure A8 (a) and (b), it can be seen that the OCO-2 XCO2 also has deviation from TCCON XCO2 at low AOD.
P21L427: For future CO2M, you found the large increase of data coverage after adjusting AOD threshold from 0.2 to 0.5. After adjusting, there will be more underestimated and overestimated data when the AOD is below 0.5, i.e., the data in Q1 area you defined will include data in Q2 and Q4 areas. You should discuss how to conduct quality filter for CO2M, or the data quality with the AOD threshold of 0.5.
Citation: https://doi.org/10.5194/amt-2024-77-RC1 - AC1: 'Reply on RC1', Timo Virtanen, 05 Aug 2024
-
RC2: 'Comment on amt-2024-77', Anonymous Referee #2, 10 Jun 2024
I understood that this manuscript studies the possibility of CO2 retrieval in high AOD pixels by the AOD threshold change.
Because of the disadvantage of spatial coverage for CO2 satellites, this study will contribute to enhancing the global coverage of observation data for CO2 satellites. Although the purpose of the manuscript is acceptable, the detailed analysis and results are not so clear. For details..
1) Section 2: The study used the Dark Target (DT) algorithm to concentrate the urban surface. However, the DT algorithm have large uncertainty of AOD over land surface as compared to the ocean surface. For the detailed analysis of retrieval uncertainty related to the AOD, retrieval results over land surface are carefully handled. Do you have the same results when AOD from MODIS uses the Deep-blue algorithm?
2) L72-L73: For the OCO-2 AOD retrieval, two representative types selection is confused. Does this sentence mean that spatio-temporal variated climatological types are selected for the AOD retrieval? In addition, Is the AOD from OCO-2 hard to consider the 'case dependent' aerosol types?
3) L99: For the AERONET AOD reference, Eck et al. (1999) is too old to explain the Version 3. Giles et al. (2019) or Sinyuk et al. (2020) are more suitable.
Giles, David M., et al. "Advancements in the Aerosol Robotic Network (AERONET) Version 3 database–automated near-real-time quality control algorithm with improved cloud screening for Sun photometer aerosol optical depth (AOD) measurements." Atmospheric Measurement Techniques 12.1 (2019): 169-209.
Sinyuk, Alexander, et al. "The AERONET Version 3 aerosol retrieval algorithm, associated uncertainties and comparisons to Version 2." Atmospheric Measurement Techniques 13.6 (2020): 3375-3411.
4) Section 3.1: For the collocation, I can't find the cloud screening method before the grinding. The AOD retrieval products are very important to the cloud screening before the analysis, although respective AOD retrieval algorithm have there own cloud masking method.
5) Section 3.3 and more: For the collocation, the author has to check the consistency of data variability within the collocation range (spatially and temporally). Could you provide the reference for spatial and temporal collocation ranges?
6) Section 4: Did only use the Quality flag value from OCO-2? Why don't you use the quality flag for another satellite platform (MODIS AOD)?
In most of the results' subsections, the paper did not have sub summary or sub-conclusion. For this reason, it may be difficult to a connection among the results.
7) L184-L185: How to eliminate the cloud contamination? Cloud contamination affects the high AOD, and it affects the high correlation between OCO-2 and MODIS, when both algorithms have cloud contamination.
8) L194: I don't agree with the spectral conversion based on the MERRA-2. To use this method, the author has to analyze the intercomparison between MERRA-2 angstrom exponent and AERONET angstrom exponent.
9) Section 4.2: I clearly don't know the purpose of this section. Only quality checking? or making the threshold of AOD to define the high AOD?
In addition, for the AOD quality checking, gridded dataset is not adequate. If you use the gridded AOD data, the author has to make a finer resolution.10) L236: Do you have references or analyzed results? I agree with the cloud contamination. However, the effect of ice aerosol component is not clear. Please include some back-up result.
11) L240: Based on the statistical results and figures, this paragraph is not clear. The statistical score is possible to change due to the large number of data under low AOD grids. Statistical score change is not efficient in explaining the quality change of datasets.
12) L293: Do you have reference?
13) Figure 5: Showing the number of data in each bin as adding figure.
14) Section 4.4: So, from this section, does the author think that the AOD affects the XCO2 retrieval? How to be quantitatively separate the effects between the AOD effect and real XCO2 enhancement?
15) Section 5: I am confused about whether the AOD threshold change is acceptable.
For focusing on the comparison between XCO2 and MODIS AOD, the moderate AOD condition will make it possible to estimate the accurate XCO2 value. However, it is just the data based on the AOD from MODIS.
The AOD difference between OCO2 and MODIS is partially due to the AOD retrieval limitation by the OCO2.
In this case, high AOD conditions from OCO2 have high uncertainty. From this study, is this case can be clarified?Citation: https://doi.org/10.5194/amt-2024-77-RC2 - AC2: 'Reply on RC2', Timo Virtanen, 05 Aug 2024
-
RC3: 'Comment on amt-2024-77', Anonymous Referee #3, 20 Jun 2024
This paper focuses on the collocation of OCO-2 and MODIS data, and analyzes the relationship between CO2 retrievals from OCO-2 and AOD retrievals from both OCO-2 and MODIS. The authors demonstrate that errors in AOD retrievals affect XCO2 retrievals, and also show that excluding data points with moderate AOD (0.2-0.5) excludes many areas with high XCO2. As a result, the authors recommend relaxing the MODIS AOD threshold to 0.5.
Major comments
The authors note (in lines 181-185) that there is low correlation between MODIS AOD and OCO-2 AOD in Australia, the Sahel, the Western US, and Central Asia, using MODIS Dark Target observations. However, these areas seem to include bright land surfaces like large deserts and snowy mountain ranges, so perhaps using MODIS Deep Blue would be more appropriate for such areas. The Western US and Sahel also show high correlation between XCO2 and MODIS AOD in Figure 4b. Can this be explained by poor quality MODIS Dark Target observations? Would using MODIS Deep Blue for these areas change the analysis?
Section 4.5: The authors show that increasing the AOD threshold to 0.5 will increase the fraction of acceptable data, but do not show or discuss how this will affect the quality of XCO2 retrievals. It seems that improving the quality of XCO2 retrievals at higher aerosol loads remains an open challenge -- the authors should state this explicitly.
Minor comments/technical corrections
Line 299: Change (A3) to (see Fig. A3)
Citation: https://doi.org/10.5194/amt-2024-77-RC3 - AC3: 'Reply on RC3', Timo Virtanen, 05 Aug 2024
Data sets
Data and code for manuscript "A global perspective on CO2 satellite observations in high AOD conditions" by Virtanen et al. Timo H. Virtanen https://doi.org/10.57707/fmi-b2share.62269cb9cf944d5595692a5f8ea6b915
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 440 | 173 | 144 | 757 | 26 | 22 |
- HTML: 440
- PDF: 173
- XML: 144
- Total: 757
- BibTeX: 26
- EndNote: 22
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1