the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Algorithm evaluation for Polarimetric Remote Sensing of Atmospheric Aerosols
Pavel Litvinov
Guangliang Fu
Cheng Chen
Oleg Dubovik
Abstract. From a passive satellite remote sensing point-of-view, the richest set of information on aerosol properties can be obtained from instruments that measure both intensity and polarization of back- scattered sun light at multiple wavelengths and multiple viewing angles for one ground pixel. However, it is challenging to exploit this information at a global scale because complex algorithms are needed with many fit parameters (aerosol and land/ocean reflection), based on online radiative transfer models. So far, two of such algorithms have demonstrated capability at a global scale: the Generalized Retrieval of Atmosphere and Surface Properties (GRASP) algorithm and the Remote Sensing of Trace gas and Aerosol Products (RemoTAP) algorithm. In this paper, we present a detailed comparison of the most recent versions of RemoTAP and GRASP. We evaluate both algorithms for synthetic observations, for real PARASOL observations against AERONET for common pixels, and for global PARASOL retrievals for the year 2008. Both RemoTAP and GRASP show good agreement against AERONET. For the Aerosol Optical Depth (AOD) over land, both algorithms show a Root Mean Square Error (RMSE) of 0.10 (at 550 nm). For Single Scattering Albedo (SSA), both algorithms show a good performance in terms of RMSE ( 0.04) but RemoTAP has a smaller bias (0.002) compared to GRASP (0.021). For Angstrom Exponent (AE), GRASP has a smaller RMSE (0.367) than RemoTAP (0.387), mainly caused by a small overestimate of AE at low values (large particles). Over ocean both algorithms perform very well. For AOD, RemoTAP has an RMSE of 0.057 and GRASP an even smaller RMSE of 0.047. For AE, the RMSE of RemoTAP and GRASP are 0.285 and 0.224, respectively. Based on the AERONET comparison, we conclude that both algorithms show very similar overall performance, where both algorithms have stronger and weaker points. For the global data products, we find a Root Mean Square Difference (RMSD) between RemoTAP and GRASP AOD of 0.12 and 0.038 over land and ocean, respectively. The largest differences occur over the biomass burning region in equatorial Africa. The global mean values are virtually unbiased with respect to each other. For AE the RMSD between RemoTAP and GRASP is 0.33 over land and 0.23 over ocean. For SSA, we find good agreement over land (RMSD=0.043) for retrievals with AOD > 0.2. Over ocean the agreement is poor with a bias of 0.053 (where RemoTAP retrieves higher SSA) and an RMSD of 0.074. As expected, the differences increase towards low AOD, both over land and ocean. We also compared the GRASP and RemoTAP AOD and AE products against MODIS. For AOD over land, the agreement of either GRASP or RemoTAP with MODIS is worse than the agreement between the 2 PARASOL algorithms themselves. Over ocean, the agreement is very similar among the 3 products for AOD. For AE, the agreement between GRASP and RemoTAP is much better than the agreement of both products with MODIS. Overall, the good agreement between RemoTAP and GRASP, and between the individual algorithms with AERONET, demonstrates high fidelity of both data products for PARASOL. The agreement of the latest product versions with each other and with AERONET improved significantly compared to the previous version of the global products of GRASP and RemoTAP. The results demonstrate that dedicated effort on algorithm development for MAP aerosol retrievals still leads to substantial improvement of the resulting aerosol products and this is still an ongoing process.
- Preprint
(44860 KB) - Metadata XML
- BibTeX
- EndNote
Otto Hasekamp et al.
Status: final response (author comments only)
-
RC1: 'Comment on amt-2023-203', Anonymous Referee #4, 19 Oct 2023
The authors evaluate the performance of the most recent version of two global aerosol algorithms, GRASP and RemoTAP, applied to POLDER/PARASOL onboard the A-Train, a space-borne sensor that measured intensity and polarization of backscattered light at multiple wavelengths and viewing angles.
Major comments
This paper is very well structured and written. The scientific analysis is robust, the results comprehensive are clearly laid out. Such a study and its results are essential to pave the way for the next generation of spaceborne aerosol and cloud missions that will include a polarized sensor (e.g., HARP/ SPEXOne on PACE, 3MI, AOS etc.).
We recommend that the authors:
- Clearly spell out, define, and consistently use their abbreviations throughout the text for all recent vs. older versions of GRASP and RemoTAP (e.g., GRASP-CC vs. GRASP-O, -HP and -M)
- Discuss how they were able to validate their spaceborne aerosol properties using AERONET stations over ocean. They must have used coastal AERONET stations, in which case they might want to discuss the performance of their spaceborne observations over complex surfaces and how this would not be an issue for MAP compared to spaceborne sensors such as MODIS/ VIIRS
- Should briefly discuss their choice of aerosol type in Table 1 i.e., with an RRI of 1.54 and an IRI of 0.005 more representative of maybe mineral dust. Same comment applies to Table 3 – why choose these AEROENT stations? Maybe they provide a wide enough range of geometries
- It is not clear why the authors compare POLDER and MODIS products (section 4.4.1) under section 4.4 (i.e., surface properties)
Detailed comments:
Abstract: The authors need to spell out MAP once
Line 57: Maybe add “from space” after “the only MAP instrument…”
Line 108: SPEX and airMSPI need to be spelled out.
Line 151: “includes” is repeated twice
Line 154: “a posterior filter removing retrieval”
Line 163-169: See major comment #1 above
Line 173: We suggest “we use the set of aerosol properties provided in Table 1 (…) in Table 2”. Also see major comment #3 above.
Line 200: based on direct sun
Line 203: Doesn’t the AERONET SSA accuracy depend on the AOD range?
Line 230: The reader could use a reminder of the POLDER-3/PARASOL measurement uncertainty
Line 244: This is the first time in the text that ECHAM is called “SRON-ECHAM”.
Line 263: Tables are often introduced in the order that they are numbered (e.g., Table 4 before 5)
Line 266 and legend of Table 4: The reader could use a reminder of the current spatial gridding in this study (0.1º)
Line 268-69: See major comment #1 above
Section 4.1.2: See major comment # 2
Figure 7: Add “over ocean”
Figure 8: The reader might wonder which one is right/ wrong when looking at the AOD, AE or SSA bias between RemoTAP and GRASP. Would it be possible to quickly refer to the AERONET evaluation results for a few stations as an illustration (like it is done for Fig. 10)?
Figure 9, 11, 13, 22: blue and black are not easy colors to differentiate in the histogram plot.
Figure 12 and 17: y-axis says “sron” instead of RemoTAP
Line 347: “that” is repeated twice
Line 385: “significantly”
Line 410: “as noted”
Line 427: authors might want to discuss potential “ground truth” for such spaceborne BRDF/ BPDF retrievals
Figure 20, 21: “Same as Fig. 19”
Line 429: A few words explaining why it is important here to compare POLDER-derived to MODIS-derived aerosol properties would help.
Line 431: authors might want to say how they combined DT and DB from MODIS
Line 445: “based on a discrete”
Line 448: you might want to also refer to Reid et al., 2022; Reid, Jeffrey S., et al. "A coupled evaluation of operational MODIS and model aerosol products for maritime environments using sun photometry: evaluation of the fine and coarse mode." Remote Sensing 14.13 (2022): 2978.
Line 459: spell out CO2M
Line 462-3: sentence is a little too convoluted
Line 467: “related to a small”
Line 514: authors might want to add aerosol retrievals below thin clouds as a future research product
Line 518: authors might want to ask for more aerosol validation over ocean, where we do not have AERONET stations (e.g., airborne lidar and/ or sunphotometer)
Citation: https://doi.org/10.5194/amt-2023-203-RC1 -
RC2: 'Comment on amt-2023-203', Anonymous Referee #1, 04 Nov 2023
In this manuscript titled “Algorithm evaluation for Polarimetric Remote Sensing of Atmospheric Aerosols“ by Hasekamp et al, two advanced retrieval algorithms which exploit the rich aerosol information from measurements with multiple angles, wavelengths and polarizations are compared based on both synthetic data and global scale PARASOL data. Collocated AERONET data are used to evaluate the retrieval qualities. These two algorithms are the Generalized Retrieval of Atmosphere and Surface Properties (GRASP) algorithm and the Remote Sensing of Trace gas and Aerosol Products (RemoTAP) algorithm. Both are well received in the community and widely used with broad applications.
Overall, good agreements are found between RemoTAP and GRASP, and between the individual algorithms with AERONET data product. The differences in the RemoTAP and GRASP algorithms, at different regions, AOD ranges, etc, are also discussed. The comparison between the two algorithm is important for the community to understand the advantage of the polarimetric remote sensing and impacts from the assumptions in the aerosol and surface properties. Therefore, this work provides useful guidance for the future improvements. The manuscript well written with comprehensive information on the algorithm, synthetic and real observations, comparison details of key aerosol optical properties, such as AOD, SSA, AE etc. I would recommend the publication of this manuscript after addressing a few suggestive comments for clarities.
General comments:
1. It is nice to see the converge of the two leading retrieval algorithms in many of the aerosol properties. However, it is not clear whether the retrieval uncertainties and differences are due to inversion algorithm (cost function, convergence, numerical accuracy of forward model etc), or measurement and forward model uncertainty assumptions. Can the author clarify whether the ideal retrieval uncertainties from error propagation have been derived and compared? Such comparisons can also help us to understand the sources to explain the differences between the two algorithms, and any potential to improve on the properties where both algorithms already agree on.
2. It seems different cost functions are used in the inversion. RemoTAP relies on the reflectance and DoLP, and GRASP replies on reflectance, q, and u (Page 5, Line 149). Different uncertainties are also used in DoLP and q,u. Since the cost function is fundamental in the inversions, can the authors comment that how much impacts of the choice of these different cost functions?
3. It seems GRASP algorithms use a multi-pixel approach with additional constraints on spatial and temporals, can the authors clarify where are the comparisons between single pixels retrievals and where are with multi-pixel retrievals? Can you quantify how much the multi-pixel algorithm helps the aerosol retrievals and how that impacts this work?
Detailed comments:
1.Page 3, Line 80: “Of the full inversion approaches only the RemoTAP and GRASP algorithms have demonstrated capability at a global scale. “
Among the algorithm mentioned in the manuscript, it seems MAPOL has also been used in a global scale processing but on synthetic data (https://egusphere.copernicus.org/preprints/2023/egusphere-2023-1843/).
2.Is processing resource a factor which impact the capability to process global scale data? can you provide an estimate of the resources used between these two algorithms to process the whole year of PARASOL data?
3.Page 3, Line 88: “although based on theoretical information content a better agreement is expected”.
Similar to the general comments, can you provide more information on the gaps and possible causes?
4. Page 4, line 118: “This allows to improve retrieval by using additional a priori constraints on spatial or temporal variability of any retrieved parameter in different pixels”, also later stated in Page 5, Line 127.
Also related to the general comments, can you confirm whether such multi-pixel approach is used in this study, and how that impact this study?
5. Page 5, Line 152, ” The assumed uncertainty is 1% on I and 0.007 (absolute) on DoLP “ (for RemoTAP), and “The assumed uncertainty is 1 % on I and 0.002 (absolute) on q and u. “ (for GRASP).
Why is there higher accuracy for q and u, and how that impacts retrieval results?
6. Page 6, Line 162: “The version of GRASP is based on the recently developed Chemical Component (CC) approach (Li et al., 2019; Zhang et al., 2021), while the previous global versions of GRASP (Chen et al., 2020) were based on the GRASP "Optimized", "High Precision", and "Models" approaches, referred to as GRASP-O, GRASP-HP, and GRASP-M.”
The authors later mentioned different performance of these algorithms. Does it due to a better choice of priori constraints in the aerosol models? Better alignment with assumptions in AERONET inversions?
7. Page 8, line 201: “Here, it should be noted that the AOD (and hence AE) AERONET product is based om direct sun measurements achieving high AOD accuracy (0.01-0.02) whereas the SSA is based on inversion of diffuse sky measurements which rely on several retrieval assumptions leading to a moderate accuracy of (0.03).”
Thanks for mentioning the AERONET product uncertainty. It seems both GRASP and RemoTAP demonstrated a similar RMSE in the order of 0.03-0.04 (Table 6 and Table 8) as comparing with the AERONET data. Since the value is so close to the AERONT uncertainty, can the difference come from the AERONET data itself? It would be very interesting to understand whether such accuracy is at the limit of the information content of the PARASOL data itself.
Similarly, to the AOD as well, as both algorithms seems have a larger RMSE in the order of 0.1, while AEROENT accuracy is 0.01-0.02. It seems a lot room to improve the retrieval product if there is enough information.
8. Page 8, line 223, “Figure 1 shows the forward model comparisons between RemoTAP and GRASP. The bias and standard deviation of the differences in radiance are mostly below 2% and for DoLP mostly below 0.005.”
Can the authors comment about whether the differences are coming from physical models or also related to numerical accuracy in the RT simulations?
Since GRASP use q and u in the retrievals, are their uncertainties also quantified?
9. Page 9, Figure 1: the plot seems noisy, do you add noise to the simulation?
10. Page 10, Figure 2: “Results have been filtered for RemoTAP chi2 < 1 and GRASP minimum residual< 3% “, why GRASP only report a residual?
Why RemoTAP choose chi2<1? For chi2<1, it would suggest overfitting, or the uncertainties used in the retrieval is underestimated.
Does GRASP also use a chi2 type of cost function? Choosing a minimum residual < 3%, would lead to a chi2 cost function value of (3%/1%)^2 = 9 for reflectance? Is this a correct estimation?
11. Page 14, Fig 4, Page 17, Fig 7:
It seems there is not much sensitivity for AOD<0.1. What is the possible reason? I would expect a higher accuracy due to the high accuracy in the reflectance measurements (1%).
12. Page 17, Table 7:
It is nice to report the uncertainty with respect to the AOD ranges! Is there such table for retrievals over land?
13. Page 32, Line 521: it seems only RemoTAP data is available and GRASP data access need permissions.
14. Page 35, Table A2: There are a long list of parameters, but it is not mentioned what they are?
Citation: https://doi.org/10.5194/amt-2023-203-RC2 -
RC3: 'Comment on amt-2023-203', Anonymous Referee #3, 06 Nov 2023
Review of “Algorithm evaluation for Polarimetric Remote Sensing of Atmospheric Aerosols”
General comments:
This study evaluates the performance of the two popular multi-angle polarimetric aerosol retrieval algorithms GRASP and RemoTAP using both synthetic data and PARASOL real measurements. Results show that retrievals from the two algorithms are not only in agreement with each other but also validated well against AERONET data. This manuscript is well organized.
Specific comments:
- The largest AOD difference between GRASP and RemoTAP exist over Sahara/Arabia and equatorial Africa. I suggest discussing the possible reasons. Are surface reflectance differences also large at these regions.
- The caption of Fig. 4 should be put below Fig. 4.
- Colorbar are missing in many figures (Figs. 9, 11, 12, 13, 14, and 16 - 23).
Citation: https://doi.org/10.5194/amt-2023-203-RC3 -
RC4: 'Comment on amt-2023-203', Anonymous Referee #5, 08 Nov 2023
General Comments:
This manuscript presents a thorough evaluation of two prominent multi-angle polarimetric aerosol retrieval algorithms, GRASP and RemoTAP, utilizing both synthetic data and real measurements from the PARASOL satellite. The study demonstrates that the results obtained from these two algorithms not only exhibit strong agreement with each other but also show robust validation against AERONET data. The manuscript is well-structured and effectively communicates the scientific analysis and comprehensive results. Such research is of great significance, laying the foundation for forthcoming spaceborne aerosol and cloud missions equipped with polarized sensors. Therefore, it is recommended for the acceptance of this study after the following issues have been addressed.
Specific Comments:
Abstract: The full names of several abbreviations should be introduced, such as of “MAP”. Similar issues are also present in the main text (SRON, CO2M, etc.).
Line 151: There are two “includes”.
Figure 1: Using different line styles in the figures is preferable to using different colors (e.g., in Figure 5).
Table: Tables should be formatted as three-line tables whenever possible.
Figure 4: The figure caption should be placed below the figure.
Figure 9, 11, 13, 14, 16, 22: Same issues with in Figure 1 mentioned above.
Line 346: There are two “that”.
Citation: https://doi.org/10.5194/amt-2023-203-RC4 -
RC5: 'Comment on amt-2023-203', Anonymous Referee #6, 20 Nov 2023
This important paper concerns the important topic of polarimetric remote sensing, which represents the new state-of-the-art in passive remote sensing of the Earth’s atmosphere-ocean/surface system. This paper compares two advanced and powerful polarimetric aerosol remote sensing retrieval algorithms, namely GRASP and RemoTAP, as applied to simulated polarimeter data, PARASOL/POLDER-3 satellite polarimeter data, and with comparisons made against AERONET ground-based aerosol retrievals and MODIS aerosol products (non-polarimetric aerosol retrievals). The paper is well-written and logically structured. The findings in this paper advance the state-of-the-art in polarimetric aerosol remote sensing, and the paper is complete in terms of content.
The paper is thus recommended for publication subject to minor revisions to address the following recommendations and comments:
Abstract: The term “good agreement” is qualitative. For example, an RMSD difference of 0.12 in AOD at 550 nm (or 0.1) does not seem “good”. What does “good agreement” mean? Does it mean to achieve as much agreement as can be expected between RemoTAP and GRASP aerosol and surface properties retrieved for a given instrument (e.g. PARASOL/POLDER-3) with its particular set of channels, angles and corresponding instrument measurement uncertainties?
Line 150: Justify/cite the choice of 1% radiometric measurement uncertainty. Is this a 1-sigma (one standard deviation) uncertainty?
Line 230: Add a reference for the PARASOL/POLDER-3 instrument uncertainties that are mentioned.
Section 4.4 (Surface properties). Distinguish between surface properties over land vs ocean. Is there evidence to suggest that the differences between the ocean surface models in RemoTAP and GRASP can be responsible for the considerable differences in retrievals of aerosol SSA over the ocean? Or is it something else?
Is the “BPDF scaling parameter” the same as the “Maignan scaling” parameter in Table 2? Recommend using one term consistently, and adding the Greek/Mathematical symbols corresponding to the written parameters in Table 2.
Figures 19-21: the colorbar appears to be missing. What does the white line represent? Should “delta” be “\Delta” in Latex?
Minor corrections:
Line 142: lambda -> \lambda
Line 201: om -> on
Line 246 (and a few other places): correctly use ` as the opening single quote
Citation: https://doi.org/10.5194/amt-2023-203-RC5
Otto Hasekamp et al.
Otto Hasekamp et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
253 | 96 | 16 | 365 | 3 | 5 |
- HTML: 253
- PDF: 96
- XML: 16
- Total: 365
- BibTeX: 3
- EndNote: 5
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1