the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Feb 2025
Harmonized Cloud Datasets for OMI and TROPOMI Using the O2‐O2 477 nm Absorption Band
Abstract. We present a new cloud retrieval algorithm using the O2-O2 absorption band at 477 nm, designed to provide harmonized cloud datasets from OMI and TROPOMI. The goal of these derived cloud data is to mitigate the influence of clouds on the retrieval of tropospheric trace gases from UV-visible nadir satellite spectrometers. The retrieval process consists of two main steps: first, spectral fitting is performed using the Differential Optical Absorption Spectroscopy (DOAS) method to determine the O2-O2 slant column and calculate the reflectance at the center of the fitting window. Second, these parameters are used to derive cloud fraction and cloud pressure.
This retrieval algorithm builds on the OMI O2-O2 operational cloud algorithm (OMCLDO2) with several improvements. The fitting procedure uses a broader fitting window, incorporating the O2-O2 absorption bands at 446 and 477 nm, to more accurately derive O2-O2 slant column densities (SCD). A de-striping correction is applied to address across-track variability, and an offset correction motivated by radiative transfer simulations is introduced to correct the O2-O2 SCD bias between OMI and TROPOMI. Additionally, a temperature correction factor is included to account for the temperature dependence of both the O2-O2 SCD and the O2-O2 absorption cross-section. Consistent auxiliary data, such as meteorological information and surface albedo database, are used for both sensors. Due to the suboptimal quality of solar irradiance measurements by OMI, a fixed annual averaged irradiance for 2005 is used as a reference for the reflectance spectra in the spectral fittings.
To evaluate the performance of our retrieval approach, we compare it with the OMCLDO2 algorithm for both OMI and TROPOMI. The cloud fraction retrievals demonstrate good agreement, whereas the cloud pressure retrievals show a systematic bias, particularly in nearly cloud-free scenes. Our cloud pressure estimates tend to be higher than OMCLDO2 for OMI and lower for TROPOMI. Notably, our approach demonstrates improved consistency in cloud parameters, especially cloud pressure, between the two sensors compared to OMCLDO2. However, a consistent bias of approximately 0.05 in cloud fraction retrievals is observed, primarily attributed to differences in L1b data. Applying these cloud corrections to NO2 retrievals reveals that the average impact of cloud corrections ranges from -6 % to 11 % in polluted regions. Differences in NO2 AMF resulting from varying cloud correction methods can exceed 10 %. Importantly, the new correction approach achieves better consistency in NO2 retrievals between OMI and TROPOMI.
Competing interests: Michel Van Roozendael is an editor for Atmospheric Measurement Techniques
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.- Preprint
(10459 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 26 Apr 2025)
-
RC1: 'Comment on egusphere-2025-478', Anonymous Referee #1, 17 Mar 2025
reply
This paper presents a cloud retrieval algorithm utilizing a broader O2-O2 absorption bands between 435-495 nm, along with correction methods including de-striping, offset correction for slant column density, and temperature corrections. It also incorporates auxiliary data, such as meteorological data and surface albedo climatology, in the retrievals. The effect of each improvement (except for the wider fitting spectral range) is analyzed using real measurements from TROPOMI and OMI, comparing the results with those from the classic OMCLDO2 algorithm. The TROPOMI and OMI retrievals align better using the proposed algorithm. NO2 retrievals are also corrected using the proposed cloud results, OMCLDO2 cloud results, and cloud results from the FRESCO algorithm. The corrected NO2 retrievals are then compared with one another. The theory is sound, and the analysis is thorough. I recommend the paper be accepted after minor revisions.
- Section 3.3: Does the correction for across-tract variability (de-striping correction) affect the retrievals of cloud fraction? This seems not to be demonstrated in the section.
- Section 3.4 introduces the offset correction for slant column density. However, much of the content describing the temperature correction seems to belong in Section 3.5.3. This overlap may cause confusion for the readers. Additionally, why is an offset of -0.08x1043molec2cm-5 applied to the TROPOMI SCDs? This offset appears to be the corrected value intended for this section, but how this value is derived is not explained.
- Is radiometric cloud fraction the same as effective cloud fraction? The two terms are used interchangeably, but the relationship between them is not clarified in the paper.
- 545 “We”-> “we”
Citation: https://doi.org/10.5194/egusphere-2025-478-RC1 -
RC2: 'Comment on egusphere-2025-478', Anonymous Referee #2, 14 Apr 2025
reply
General comments
The article (Harmonized Cloud Datasets for OMI and TROPOMI Using the O2-O2 477 nm Absorption Band, by Huan Yu et al.) describes an essential (and timely) step that leads to a closer agreement between the retrieved OMI and TROPOMI cloud properties. However, for the derived cloud pressures, the OMI/TROPOMI consistency relies (in my opinion, in a major way) on the TROPOMI slant-column adjustment. Without it, the OMI/TROPOMI differences could be as large as in the already implemented OMCLDO2 retrievals.
The Ozone absorption greatly impacts the wavelength range used in the described O2-O2 retrievals. As a ‘sanity check’, I may advise comparing the O3 SCDs retrieved in the proposed approach to the OMI O3 SCDs (OMDOAO3). If there are any large, systematic (in essence, VZA and SZA-dependent) differences between the two O3 products, this may imply biases in the retrieved O2-O2 SCDs. E.g., one may note the large high-latitude differences between different approaches (Figure 2). Could these be related to the Ozone interference, besides the implied differences in the surface-albedo characterization?
------------------------------------------------------------------------------------------
Specific comments
Abstract:
“Due to the suboptimal quality of solar irradiance measurements by OMI…” Consider changing this to: “Due to the inadequate signal-to-noise ratios in the daily solar irradiance measurements by OMI…”. ‘Suboptimal’ may sound like a polite ‘inferior’.
“Notably, our approach demonstrates improved consistency in cloud parameters, especially cloud pressure, between the two sensors compared to OMCLDO2.” This comes in contradiction with the statement from Section 4.1.2 that compares the cross-track trends (thus the offsets) between the BIRA-IASB and OMCLDO2 approaches, where line 503 states “This analysis does not allow us to determine which algorithm achieves better agreement between the two sensors.” Indeed, the metrics are many, and many of them show mixed performances. As presented, it mostly relies on the systematic -0.08×1043 molec2 cm−5 TROPOMI SCD offset. This should be explicitly mentioned in the Abstract and the main text.
“… primarily attributed to differences in L1b data.” Consider expanding this to “…primarily attributed to differences in L1b data that show systematic biases between the OMI and TROPOMI reflectances”.
Table 1. Consider expanding/amending the comments
- “ I0 correction is applied with SCD of 5·1015molec./cm2” Does this mean: an offset of 5·10^15molec./cm2 is applied to the cross-sections? Or something else?
- same as for a)
- “additional cross-section taken as the inverse of the reference spectrum.” It is not clear how this could be applied to the wavelength shifts.
Table 1. Also, mention the intensity offset, if applicable (e.g., line 183).
l.167 Please comment on how the broader fitting window helps to improve the low-SCD retrievals.
Figure 2. Were the shown BIRA-IASB OMI retrievals de-striped? For better consistency (clarity of the trends), they should not be. Please clarify. Please also comment on the x-track features seen in the BIRA-IASB-OMCLDO2 differences (panel b). Are the shown (panel d) TROPOMI SCDs adjusted by -0.08×1043 molec2 cm−5 ? If they are not, then in this particular example I would insist on applying such an adjustment prior to comparison.
How typical are the differences shown in Figure 2? Are these season-dependent? Please comment on.
l.221 “Use the median SCDs from rows 2–21 in the desired year, subtracting the mean of the corresponding values…” What are these median SCDs? Are these SCDs adjusted for the linear fits? Are the medians derived row-by-row? What are these corresponding values: the median(s) or something else? The description must be expanded, keeping in mind that at some point it could be implemented by somebody else.
I may suggest normalizing (beforehand) all SCDs by tan(VZA) and, possibly (?), by cos(SZA); or, alternatively, by the geometric air mass factor. This could further simplify the approach.
l.227 “The data selection method used in Figure 3(a) avoids regions with significant variations in surface albedo and surface pressure.” The described destriping approach uses SCDs, and SCDs only. If SCDs are additionally filtered for these two factors, please add a comment. Otherwise, I’m not able to follow the statement – please explain.
l.237 Even with the minimal inter-annual variability of the striping patterns, there is no guarantee that these patterns remain stable intra-annually. Please either comment on or provide additional (Appendix) evidence of the short-term (seasonal) stability.
Figure 4. It is impossible to assess the impact of de-striping from 2D plots. Please add 1 panel that shows the average (say, for scanlines 1150-1200) cloud pressures for the three products (b-d).
l.280 Is/are there any reason(s) why OMI is considered the ‘truth’, thus prompting the TROPOMI adjustment? Considering that TROPOMI retrievals use more advanced surface characterization, one may side with TROPOMI as a ‘truth’. Please comment on.
l.300 “…this study uses the TOA reflectance, which accounts for both Rayleigh scattering and atmospheric absorption…” Ozone is an important contributing factor, especially at high VZA and SZA angles. Is it accounted for? Please mention what atmospheric absorptions are taken into account.
l.319 “…where the cloud fraction lies within [0, 1.5]…” What necessitates the range extension to (completely unphysical ) 1.5 – please explain.
l.345 “Here, T represents the atmospheric temperature representative of satellite observations…” The key word is ‘representative’. Considering the multitude of contributing factors, please provide more details on choice of the representative temperature. Is this c(T) related to C(T(p)) from eq. 7? If true, then change eq. 6 to c(T(p)=…(T(p)-T_0)… Otherwise, explain how c(T(p) and c(T) are brought together.
l.355 “…the temperature correction for the O2-O2 cross-section must be accounted for when creating the inverted LUT.” Please provide more details on how this is done. Via an additional LUT dimension?
Figure 7 “…compares temperature correction factors calculated using the BIRA-IASB and OMCLDO2 algorithms…” Note that the only unbiased way to assess the changes caused by the different temperature corrections is to use these two corrections within the framework of the same retrieval algorithm. Otherwise, other contributing factors (LUTs and the adopted interpolation approach, surface characterization, general DOAS setup, etc., etc.) may not be cleanly disentangled from the temperature impact. Consider either performing such a test or removing Figure 7(b) and the related discussion.
l.422 “Additionally, cloud pressure retrievals are shown only for pixels with cloud fractions above 0.01.” Later, in the caption of Figure 15, the authors say: “To ensure reliable cloud pressure comparisons, pixels with a cloud fraction below 0.05 are excluded, as very low cloud fractions lead to high uncertainties in cloud pressure retrievals.” Agreed! Please make sure that this 0.05 threshold is consistently implemented in the revised text (e.g., in Figures 10, 13, 14 and elsewhere).
Section 4.1.1 If the adjustment (-0.08×1043 molec2 cm−5 ) was applied to the TROPOMI data, mention this early in the Section.
Sect. 4.1.1, Figures 11-14. It would be very helpful to provide a table with the relevant characteristics (offset, slope, correlation) for all the performed comparisons, cloud fractions first, followed by the cloud pressures. The discussion could be shortened and streamlined hereafter.
Figure 13. There are extremely large (300-400 hPa ) offsets in the shown TROPOMI low-cloud-fraction retrievals, both over land and oceans. The introduced SCD offset may account for a very small fraction (~50 hPa) of these differences. Where does this (very concerning!) BIRA-IASB and OMCLDO2 difference come from?
Figure 14. The sharp increase of cloud fractions toward the edges of the swaths is puzzling, especially considering the absence of such prominent trends in the OMI retrievals based on an alternative approach (Vasilkov et al. 2018). This also applies to the west-east cloud-pressure differences seen in the discussed retrievals. I hope these subjects will be followed upon by the authors.
l.603 Please remind the reader that the OMCLDO2 OMI/TROPOMI retrievals use [very!] different surface albedo datasets. The differences could be mostly related to this aspect.
Technical comments
l.122 Mention that Band 4 is used for the O2-O2 retrievals. Also mention how the saturation flags are accounted for in the retrievals.
l.193 “In contrast, for OMI, the limited quality of the daily solar measurements…” – consider adding ‘daily’.
Figure 5, the caption: “Data are binned by cloud fraction intervals of 0.1…” – use 0.1 instead of 0.01
Figure 6(a) – Please check the match between the color legend and the plotted curves.
l.478 “The retrieved cloud fraction is constrained between 0.05 and 1…”
Citation: https://doi.org/10.5194/egusphere-2025-478-RC2 -
RC3: 'Comment on egusphere-2025-478', Anonymous Referee #3, 16 Apr 2025
reply
In this manuscript, the authors report on an improved version of the O2-O2 cloud retrieval for OMI and TROPOMI, aiming at consistent trace gas retrievals for the two instruments. The main improvements of the algorithm are the use of consistent a priori data, a larger O4 fitting window, treatment of the temperature dependence of the O4 cross-sections, destriping for OMI and an offset correction of the TROPOMI O4 columns. With these updates, the cloud retrieval results in more consistent cloud parameters between the two instruments than with the current implementations of the O2-O2 retrieval, in particular with respect to cloud pressure. Application to tropospheric NO2 columns also shows good consistency between the two instruments.
The manuscript reports on an interesting study, which fits well into the scope of AMT. It is well written, has a good level of detail and is of relevance for the UV/vis satellite community. I, therefore, recommend it for publication after revisions.General comments
Although I overall liked the paper and enjoyed reading it, there are several important aspects which in my opinion need to be addressed before publication.
1) Many aspects of the O4 retrieval have been changed, and it is not entirely clear, which one is the driver of the better agreement between the two instruments that the authors find. The use of the larger fitting window is not really motivated in the manuscript, the temperature correction is physical but was not used on either of the instruments in the past, so the main differences are the use of consistent a prioir information and the (not very satisfying) ad hoc offset correction. I suggest that the effect of the different changes is tested separately on a subset of data to clarify this point.
2) A significant effort was taken to correct the row dependence of the OMI O4 columns, and the results are certainly an improvement. However, in the final products (Fig. 14), the row dependency of the cloud pressure is smoother but not really smaller than in the original OMI product. It also is different between OMI and TROPOMI. This is a concern as it points at systematic problems in the retrieval. Please discuss this result in more detail.
3) While the results of the improved algorithm are compared in detail with those of the existing O2-O2 algorithm, the real alternative to the new algorithm is the existing FRESCO cloud algorithm. In contrast to the O2-O2 data stored in TROPOMI NO2 files, the FRESCO results are used by the current TROPOMI NO2 and HCHO retrievals, and this is what the new algorithm should (also) be compared to! I think that all comparisons dealing with cloud fractions and cloud pressures need to include the currently used TROPOMI cloud algorithm.
4) I’m not comfortable with the discussion of the effect of using different cloud products on the NO2 tropospheric columns. On the one hand, the way the numbers are discussed in the text focuses on the largest differences, which are driven by the TROPOMI O2-O2 algorithm which has to my knowledge never been used for tropospheric NO2 retrievals. In that sense, the importance is exaggerated, and in my opinion, differences of a few percent are small considering all the other uncertainties in the retrievals. A more appropriate message would therefore be “cloud correction does not make a big difference for NO2 tropospheric retrievals under the assumptions made here”.
On the other hand, the data are monthly averages, and for individual scenes, I expect much larger differences. Therefore, I’d suggest to add histograms of differences to the mean values for a more balanced view.
Finally, the results in Fig. 18b are quite surprising when comparing the cloud corrections TROPOMI-BIRA and operational – with the exception of Eastern China, they always have opposite sign! I think that warrants more discussion than just referring to “sampling effects”. How can we trust cloud corrections at all, if two different, state of the art approaches lead to opposite effects?Detailed comments:
L28: … most are partially covered by clouds.
L31: wasn’t the IPA assumption already used in Martin et al., 2002?
L37: Although most people will know OMI, I think a very short introduction of OMI and in the next paragraph TROPOMI is needed here
L38: Not sure if a trace gas with an exponential mixing ratio profile should be called “well mixed”. What about “with a known vertical distribution”?
L40: “collision-induced absorption” sounds a bit strange as the absorption of the photon is not caused by the collision of the molecules, just the collisional complex has an absorption cross-section which has larger values at this wavelength.
L44: There is an imbalance in the discussion of OMI and TROPOMI here, as for OMI, cloud fraction retrieval is not explained
L99: “consistently sized rows” – not clear what is meant with this
L120: Please give full version number of lv1 product
L140: …can implicitly correct… => … can implicitly correct part of the …
Figure 1: 10E-46 on y-axis label
Figure 2: Which cloud fraction is used for (c) and (f)?
L225: Were the data also filtered for sun glint?
L259: ... at the 465 … => … at 465 …
L265. Add horizontal resolution of the model
L312: I do not understand, how this reduces interpolation errors, and also did not find any additional explanation in Wang et al, 2020. Can you please explain?
L327: … proportionally to with the … => … proportionally with …
L375: … slightly highly … => … slight high …
L418: … the algorithm is not sensitive to the high … => … the algorithm is not sensitive under high …
Figure 10: There are clear orbital structures in the western part of the TROPOMI image and in many places of the OMI images, both for cloud fraction and cloud pressure. Please discuss.
L481: Isn’t that just a geometric effect, that apparent cloud fraction increases over broken clouds under slant view if clouds are 3d-objects?
L489: Shouldn’t we expect a bias in OMI cloud fractions as the pixels are larger than those of TROPOMI?
L524: I think it would be worthwhile to point out, why you used the 0.2 threshold – I assume that’s because this corresponds to 50% cloud radiance fraction, and therefore is the limit used in the tropospheric NO2 discussion. This point is then relevant when discussing figures such as Figure 16a, where large discrepancies are found in exactly the range of values relevant for cloud correction in trace gas algorithms.
L579: Why is weighting with cloud fractions a good idea? I think that this artificially makes results look better, as high cloud fractions show better agreement (Figure 13) and weighting will bias everything towards large cloud fractions.
Figure 17: remove “are considered”
Figure 19: Figure headers “select data” => “selected data”
Figure 20: axis labels “refletance” => “reflectance”
Citation: https://doi.org/10.5194/egusphere-2025-478-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 150 | 35 | 7 | 192 | 4 | 6 |
- HTML: 150
- PDF: 35
- XML: 7
- Total: 192
- BibTeX: 4
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 80 | 40 |
| Belgium | 2 | 25 | 12 |
| China | 3 | 24 | 12 |
| Netherlands | 4 | 9 | 4 |
| Germany | 5 | 6 | 3 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 80