Retrieval of High Spatial Resolution Aerosol Optical Depth from HJ-1 A/B CCD Data
Round 1
Reviewer 1 Report
The paper "Retrieval of High Spatial Resolution Aerosol Optical Depth from HJ-1 A/B CCD Data" seeks to provide a single 30 m spatial retrieval algorithm for Aerosol Optical Depth using variants of more conventional MODIS Dark Target algorithms. The paper is interesting as a general approach but it is severely limited by poor English language and construction and requires extesive English proofreading. This has the undesirable affect of trying to follow the arguments very difficult and a constant struggle.
However, with enough struggle, I think I am able to generally understand and appreciate the basics of the algorithm approach which for clarity I describe below.
The main deficiencies that must be addressed in my opinion are as follows
1) The retrieval assumes a continental aerosol model. Unlike the MODIS DT retrieval algorithm using 3 bands including SWIR, which allows for more flexability in dealing with diverse aerosol types, this is not the case here. The authors do not discuss this. Are the cases being chosen for comparison selected to be specifically 'continental like' based on AERONET Microphysical processes
2) Based on the general structure of the algorithm, the authors through figure 4 are left with a situation where different AOD retrieval values will occur based on the choice of the 'constant' regression coefficient which is hard to lock down on physical grounds so the authors allow different offset values indexed by counting parameter {i} so that the retrieval at this stage allows for a set of AOD{i} and the specific value obtained must make use of additional constraints that are not part of the used blue / green channels used in the algorithm.
For this purpose, the authors introduce 2 spectral ratio type constraints including an ACI constraint and an NDVI constraint which are supposed to pin down the offset parameter {i}, To that end, the authors plot in Fig 4 the ACI vs offset {i} and provide some very loose arguments on how the mnature of the plot can be used to constrain the {i} but it is not at all clear.
The second constraint uses the NDVI obtained from existing products as a contraint taking into account of the seasonal changes that can be expected.
My "guess" is that the value of {i} and the subsequent tau{i} are then used for atmospheric correction to quantify NDVI{i} which can be compared to constain the {i} parameter.
Unfortunately, this is conjecture since the authors do not provide enough detail and show this approach from beginning to end on a specific pixel. This needs to be improved.
3) I am very dubious that retreivals can be done at 30m (single pixel) retrieval in any meaningful way. My pripr experience in applying sngle pixel retrievals to Landsat 30m using MODID like algorithms result in strong 'noisy' variability on any AOD retrieval and significant averaging of the radiances are needed reducing the resolution. To convince the community, the authors should
provide 'zoomed' in retrievals at 30 meter resolution where the AOD as well as the variability of the intermediate products such as the regression coefficients and the offset parameter can be seen and interpreted by the reader.
4) It seems difficult to me that a single AOD 'calibration' allows for the offset correction which is then used for all pixels of the image. Based on that, when doing an un-calibrated image, that should mean the for the entire image, the offset parameter is the same even for different sun-view angles and land types.
5) Perhaps the authors may provide some expertise and comments on whether this approach can be applied to Landsat.
Author Response
Responses to Review#1
We would like to thank the anonymous reviewer for his/her valuable comments that helped us revise and improve the presentation and the technical context of our paper. In the following, we addressed all comments and suggestions made by the reviewer. The corrections have been made in this revision are highlighted in blue. All numbered items (pages, equations, figures, and references) are consistent with those in the revised manuscript, excepted if otherwise stated. For convenience, the comments of the reviewer are repeated below in italics.
The paper "Retrieval of High Spatial Resolution Aerosol Optical Depth from HJ-1 A/B CCD Data" seeks to provide a single 30 m spatial retrieval algorithm for Aerosol Optical Depth using variants of more conventional MODIS Dark Target algorithms. The paper is interesting as a general approach but it is severely limited by poor English language and construction and requires extensive English proofreading. This has the undesirable affect of trying to follow the arguments very difficult and a constant struggle.
However, with enough struggle, I think I am able to generally understand and appreciate the basics of the algorithm approach which for clarity I describe below.
The main deficiencies that must be addressed in my opinion are as follows
1) The retrieval assumes a continental aerosol model. Unlike the MODIS DT retrieval algorithm using 3 bands including SWIR, which allows for more flexibility in dealing with diverse aerosol types, this is not the case here. The authors do not discuss this. Are the cases being chosen for comparison selected to be specifically 'continental like' based on AERONET Microphysical processes
Author’ reply: Thank you for your valuable suggestions and positive evaluation of our work. We have checked the English proofreading carefully, and rewritten parts of this manuscript. Please see the revised manuscript highlighted in blue for more details.
In this study, the atmospheric and aerosol types are considered as input parameters for 6S model. In fact, the aerosol lookup table can be simulated for a diversity of atmospheric and aerosol types (i.e., Continental, Maritime, Urban, Desert, Biomass, and Stratospheric). In this study, we just took the mid-latitude summer, continental aerosol type as an example for demonstration the ability and validity of algorithm with the measurements of AERONET sites in Beijing. For making it much clearer to understand, we have added a description about it in Section 2.4. Please see Page 6, Lines 174-175 for more details.
2) Based on the general structure of the algorithm, the authors through figure 4 are left with a situation where different AOD retrieval values will occur based on the choice of the 'constant' regression coefficient which is hard to lock down on physical grounds so the authors allow different offset values indexed by counting parameter {i} so that the retrieval at this stage allows for a set of AOD{i} and the specific value obtained must make use of additional constraints that are not part of the used blue / green channels used in the algorithm.
For this purpose, the authors introduce 2 spectral ratio type constraints including an ACI constraint and an NDVI constraint which are supposed to pin down the offset parameter {i}, To that end, the authors plot in Fig 4 the ACI vs offset {i} and provide some very loose arguments on how the mnature of the plot can be used to constrain the {i} but it is not at all clear.
The second constraint uses the NDVI obtained from existing products as a contraint taking into account of the seasonal changes that can be expected.
My "guess" is that the value of {i} and the subsequent tau{i} are then used for atmospheric correction to quantify NDVI{i} which can be compared to constain the {i} parameter.
Unfortunately, this is conjecture since the authors do not provide enough detail and show this approach from beginning to end on a specific pixel. This needs to be improved.
Author’s reply: Thank you for your valuable suggestion. We have reorganized and rewritten parts of this section (Section 2.2-2.5 in the revised manuscript). Please see Pages 3-8 for more details.
3) I am very dubious that retrievals can be done at 30m (single pixel) retrieval in any meaningful way. My prior experience in applying single pixel retrievals to Landsat 30m using MODID like algorithms result in strong 'noisy' variability on any AOD retrieval and significant averaging of the radiances are needed reducing the resolution. To convince the community, the authors should provide 'zoomed' in retrievals at 30 meter resolution where the AOD as well as the variability of the intermediate products such as the regression coefficients and the offset parameter can be seen and interpreted by the reader.
Author’s reply: A very good question. In this study, for avoiding the effects of sensor noises, the HJ-1 A/B CCD data was firstly resampled to 300 m, and then the AOD was retrieved from the low spatial resolution data. Finally, the AOD was resampled to 30 m for obtaining a high spatial resolution dataset. We have to admit that this processing procedure reduced the real spatial resolution of retrieved AOD dataset. However, the image quality of HJ-1 A/B CCD AOD is still superior than that of MODIS 3 km aerosol product. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 370-375 for more details.
We have provided a partially enlarged (zoomed) image of Figure 12(a) (Upper-right corner) in the revised manuscript and the corresponding AOD map retrieved from HJ-1 A/B CCD, MODIS aerosol product, regression coefficients a and b.
Fig. 1 Partially enlarged image and retrieved AOD maps. (a) Partially enlarged image; (b) AOD map retrieved from HJ-1 A/B CCD; (c) MODIS aerosol product (MOD04_3K); (d) Regression coefficient a; (e) Regression coefficient b. （see the attached pdf file）
4) It seems difficult to me that a single AOD 'calibration' allows for the offset correction which is then used for all pixels of the image. Based on that, when doing an un-calibrated image, that should mean the for the entire image, the offset parameter is the same even for different sun-view angles and land types.
Author’s reply: A very good question. The AOD retrieval accuracy can be significant improved once the measurements of an AERONET site were used as prior knowledge. The main reason for this phenomenon is that the estimation of offset step i of the regression intercept can be seriously affected by the issues of sensor calibration. Thus, the in situ measured AOD can be used as a calibration dataset for various of satellite imageries, get the optimal estimation of offset step i directly, and improve the estimation accuracy of AOD derived from satellite observations. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 364-369 for more details.
5) Perhaps the authors may provide some expertise and comments on whether this approach can be applied to Landsat.
Author’s reply: Thank you for your suggestion. The method proposed by this study can be applied to various remote sensing sensors and application scenarios. Once the sensor has visible and NIR bands, the AOD can be used for retrieved. It can be applied to commonly used satellite data, such as Landsat TM/ETM/ETM+, MODIS and POLDER etc., and also has great advantage for retrieving AOD from the sensors without SWIR bands, i.e., Quickbird CCD, Chinese Gaofen series satellites sensors, and low cost cameras onboard of unmanned aerial vehicle (UAV).We have added a discussion for it in Section 3.4. Please see Page 15, Lines 381-386 for more details.
We would like to thank the reviewer for the positive evaluation of our work and for his/her detailed remarks that helped us improve the presentation of our work.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper appears to be scientifically interesting but is difficult to understand because of errors in language, equations and figures. Also definitions are missing. Major revision is required to be acceptable.
The use of the geological word 'stratify' all over the text, including abstract, is odd for the description of the arrangement in a multidimensional table.
Line 34: Better 'variable surface properties create'
Line 50: Satellite name
Line 52: Please define all letters in the formula (total reflectance?).
Line 71: Say why only 2 satellites are used.
Lines 91f: Mention the iteration (i++ ?). Where is ACI in the flowchart? More details please how AOD is separated from surface effects.
Figure 1: POLDER! The regression LUT of the abstract is in the middle right, or? The formula for b in the middle box agrees with the text in line 172 but not with Fig. 3. Multiple use of 'i' is confusing.
Line 103f: I suppose this means multiple scattering, please be more precise here.
Eqn 3: Something appears to be wrong here. As written, S is zero always. Check line 107.
Lines 114f and Table 2: Is this the grid for the 6S-aerosol-lookup-table (Fig.1 upper left)? Is this
really four-dimensional? Is the angle grid also used for the regression LUT later (e.g. line 143)? Please improve wording and clarify.
Line 120: New page!
Eqn 4: What is the exact physical meaning of B? Upward reflected and scattered radiation in the spectral region (normalized)?
Line 133: Does this refer to Table 2 (AOD) or assumptions on the vertical distribution of aerosol? Bad wording, clarify.
Line 142: As in Table 2?
Line 151: Do you mean BRDF for HJ-1A? Improve language.
Line 152 and Table 3: Please provide the formula where Ci is used.
Line 163: and part 3?
Line 171 and 178: A factor is for multiplication, not addition. Please change wording and notation. i is a counting index.
Line 172: This fits to Fig.2 but is in contradiction to Fig.3.
Fig.3: The formulae do not represent the shown lines which appear to be as in Fig.9 (right parts)
Fig.4: AOD and 0.005i-0.05 appear to be not related by an equation, please more explanation.
Fig.5: The fit-formula is not valid in winter where it has a spike and a discontinuity, please mention.
Line 233: The formula letters of Eqn.7 are missing here.
Eqn. 8 and line 244: Something appears to be missing here. This shows the assumed error bar width but not a percentage. Please explain EE. Does a high EE mean that most data are within the error bars?
Line 247: I suppose you mean arranged or ordered and the second lookup-table. Is the grid as in Table 2, if yes refer to, if not, please provide.
Figure 8. Say something on the oscillations.
Line 289: Separated? More explanations please for Fig. 9 and the difference to Fig.3.
Line 353: 'surfaces' missing.
Line 347 and Fig. 12: Frames a and d don't show AOD from the legend, but the bands (dominating ones?). Please explain in more detail what is shown in caption.
Line 361: Is interpolation applied? The images should be available with gaps and with filling.
Fig. 13: Provide color bar or explain colors in caption.
Line 479, 480: Provide links or more bibliographical information.
Author Response
Responses to Reviewer#2
We would like to thank the anonymous reviewer for his/her valuable comments that helped us revise and improve the presentation and the technical context of our paper. In the following, we addressed all comments and suggestions made by the reviewer. The corrections have been made in this revision are highlighted in blue. All numbered items (pages, equations, figures, and references) are consistent with those in the revised manuscript, excepted if otherwise stated. For convenience, the comments of the reviewer are repeated below in italics.
1. The paper appears to be scientifically interesting but is difficult to understand because of errors in language, equations and figures. Also definitions are missing. Major revision is required to be acceptable.
Author’s reply: Thank you very much for your valuable suggestions. We have carefully checked the errors in language, equations, and figures for improving the presentation of our paper. The missing definitions were also added to the revised manuscript. Please see the answers to following questions and the revision highlighted in blue in the revised manuscript for more details.
2. The use of the geological word 'stratify' all over the text, including abstract, is odd for the description of the arrangement in a multidimensional table.
Author’s reply: We have changed the expression ‘stratify’ to ‘… to construct a lookup table for inter-band regression coefficients that vary with the solar/view angle’. We have also change the usage of this word in other places of this manuscript. Please see Page 1, Lines 13-14 for more details.
3. Line 34: Better 'variable surface properties create'
Author’s reply: We have changed this sentence to “results in large errors for quantitative inversion of surface variables”. Please see Page 1, Line 35 for more details.
4. Line 50: Satellite name
Author’s reply: The names of satellites are HJ-1 A/B. We have changed several expressions in this section. Please see Page 2, Lines 47-52 for more details.
5. Line 52: Please define all letters in the formula (total reflectance?).
Author’s reply: We have rewritten this sentence to “…derived the relationship between the visible red and blue bands”. Please see Page 2, Line 53 for more details.
6. Line 71: Say why only 2 satellites are used.
Author’s reply: The Chinese Huanjing-1(HJ-1) satellite system were consisted with three satellites: HJ-1 A, HJ-1 B, and HJ-1 C. However, the HJ-1 C satellite is a radar satellite which does not carry any CCD sensor. Therefore, only the CCD sensors onboard of HJ-1 A/B satellites can be used for retrieving the AOD. We have added an explanation about this issue in the revised manuscript. Please see Page 2, Lines 74-75 for more details.
7. Lines 91f: Mention the iteration (i++ ?). Where is ACI in the flowchart? More details please how AOD is separated from surface effects.
Author’s reply: We have redrawn Figure 1 and rewritten the description about it in the section the overall framework of the AOD retrieval algorithm (Section 2.2). Please see Page 3 for more details.
8. Figure 1: POLDER! The regression LUT of the abstract is in the middle right, or? The formula for b in the middle box agrees with the text in line 172 but not with Fig. 3. Multiple use of 'i' is confusing.
Author’s reply: We have corrected the spelling mistakes and inconsistent expression in Figure 1. Please see Figure 1 in the revised manuscript (Page 3).
9. Line 103f: I suppose this means multiple scattering, please be more precise here.
Author’s reply: We have corrected the definitions for variables in Eqn.(4). Please see Page 5, Lines 157-162 for more details.
10. Eqn 3: Something appears to be wrong here. As written, S is zero always. Check line 107.
Author’s reply: We are sorry for this mistake and have made correction for it in the revised manuscript. Please see Page 6, Line 168 for more details.
11. Lines 114f and Table 2: Is this the grid for the 6S-aerosol-lookup-table (Fig.1 upper left)? Is this really four-dimensional? Is the angle grid also used for the regression LUT later (e.g. line 143)? Please improve wording and clarify.
Author’s reply: (1) Yes. The 6S-aerosol-lookup-table is a table with four dimensions: aerosol optical depth, solar zenith, view zenith, and relative azimuth angle. We have added a description for the aerosol look-up table. Please see Page 6, Lines 172-175 for more details.
(2) The angle grid used for the inter-band regression coefficients was provided in Section 2.3. Please see Page 4, Lines 133-137 for more details.
12. Line 120: New page!
Author’s reply: We have revised it as the reviewer’s recommendation.
13. Eqn 4: What is the exact physical meaning of B? Upward reflected and scattered radiation in the spectral region (normalized)?
Author’s reply: It means surface reflectance of visible green and blue bands. We have rewritten this Equation to make it much easier to read. Please see Page 3, Line 102 and Page 4, Lines 103-106 for more details.
14. Line 133: Does this refer to Table 2 (AOD) or assumptions on the vertical distribution of aerosol? Bad wording, clarify.
Author’s reply: It refers to Table 2 (AOD). We have rewritten this part. Please see Section 2.4 for more details.
15. Line 142: As in Table 2?
Author’s reply: No. It means the POLDER sensor is able to collect observations from up to 16 viewing angles for each orbit. Please see Page 4, Lines 117-118 for more details.
16. Line 151: Do you mean BRDF for HJ-1A? Improve language.
Author’s reply: Yes. We have revised it as the reviewer’s recommendation. Please see Page 4, Lines 125-126 for more details.
17. Line 152 and Table 3: Please provide the formula where Ci is used.
Author’s reply: We have added the formula for the band conversions method. Please see Eqn.(3) in Page 4, Lines 127-130 for more details.
18. Line 163: and part 3?
Author’s reply: We have rewritten this part. Please see Section 2.2 for more details.
19. Line 171 and 178: A factor is for multiplication, not addition. Please change wording and notation. i is a counting index.
Author’s reply: We have changed the word ‘factor’ to ‘step’. Please see Page 6, Line 186 and other places in the revised manuscript.
20. Line 172: This fits to Fig.2 but is in contradiction to Fig.3.
Author’s reply: We have revised the expression to make them consistent. Please see Figure 4, Page 7, Line 192 for more details.
21. Fig.3: The formulae do not represent the shown lines which appear to be as in Fig.9 (right parts)
Author’s reply: Figure 3 is part Figure 9 for showing the effects of the variation of offset step i for the linear regression. We have reorganized the structure of this section, to make it much easier to understand. Please see Section 2.3 and 2.5 for more details.
22. Fig.4: AOD and 0.005i-0.05 appear to be not related by an equation, please more explanation.
Author’s reply: We have changed this figure for showing the relationship between ACI and AOD with different offset step i. Please see Figure 5, Page 8, Line 209 for more details.
23. Fig.5: The fit-formula is not valid in winter where it has a spike and a discontinuity, please mention.
Author’s reply: Thank you for your valuable comments. The temporal variation of NDVI constraints is only useful during the growing season of vegetation, and not applicable for winter. We have added it in the revised manuscript. Please see Page 8, Lines 215-216 for more details.
24. Line 233: The formula letters of Eqn.7 are missing here.
Author’s reply: We are sorry for this omission and have made correction in the manuscript. Please see Page 9, Lines 241-243 for more details.
25. Eqn. 8 and line 244: Something appears to be missing here. This shows the assumed error bar width but not a percentage. Please explain EE. Does a high EE mean that most data are within the error bars?
Author’s reply: We have revised the expression of EE to make it much easier to understand. Please see Page 10, Lines 251-253 for more details.
26. Line 247: I suppose you mean arranged or ordered and the second lookup-table. Is the grid as in Table 2, if yes refer to, if not, please provide.
Author’s reply: We have provided a description for the inter-band regression coefficients lookup-table in Section 3.2. Please see Page 4, Lines 135-137 for more details.
27. Figure 8. Say something on the oscillations.
Author’s reply: The oscillations was caused by the angular bin of solar/view angles. For each tiny angular bin (2° for zenith angle), the regression coefficient is constant. Meanwhile, the regression coefficients for neighbor angular bins are different. Thus, the correction effects at the boundary of the angle grid could be different, which results in such oscillations phenomenon.
28. Line 289: Separated? More explanations please for Fig. 9 and the difference to Fig.3.
Author’s reply: Figure 3 is part of Figure 9 in the previous manuscript. Figure 3 is used for showing the effects of the variation of offset step i for the linear regression. Figure 9 is used for showing the regression procedures for the HJ-1 A/B CCD BRDF dataset. Please see the Figure 2 and 4 in the revised manuscript for more details.
29. Line 353: 'surfaces' missing.
Author’s reply: We have rewritten this section. Please see Section 3.3 in the revised manuscript for more details.
30. Line 347 and Fig. 12: Frames a and d don't show AOD from the legend, but the bands (dominating ones?). Please explain in more detail what is shown in caption.
Author’s reply: Fig.12 a and d show the false color HJ-1 A/B CCD images (band 4 is displayed in red color, band 3 is displayed in green color, and band 2 is displayed in blue color). Please see Page 13, Lines 318-320 for more details.
31. Line 361: Is interpolation applied? The images should be available with gaps and with filling.
Author’s reply: In this study, for avoiding the effects of sensor noises, the HJ-1 A/B CCD data was firstly resampled to 300 m, and then the AOD was retrieved from the low spatial resolution data. Finally, the AOD was resampled to 30 m for obtaining a high spatial resolution dataset. We have to admit that this processing procedure reduced the real spatial resolution of retrieved AOD dataset. However, the image quality of HJ-1 A/B CCD AOD is still superior than that of MODIS 3 km aerosol product. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 370-375 for more details.
The inter-band correlation method is applicable for all kinds of land cover types, not for dark target only. Therefore, the aerosol derived by this method are spatially continuous without data missing.
32. Fig. 13: Provide color bar or explain colors in caption.
Author’s reply: We have added a color bar for it. Please see Page 15, Lines 355-358 for more details.
33. Line 479, 480: Provide links or more bibliographical information.
Author’s reply: We have added the links for them. Please see Page 17, Lines 454-457 for more details.
We would like to thank the reviewer for the positive evaluation of our work and for his/her detailed remarks that helped us improve the presentation of our work.
Author Response File: Author Response.pdf
Reviewer 3 Report
Review of “Retrieval of High Spatial Resolution Aerosol Optical Depth from HJ-1 A/B CCD Data” by Fan and Qu.
In this manuscript, the authors modified the Dark Target method to retrieve AOD from HJ-1A/B data at 30 m resolution over Beijing. Overall, the manuscript is interesting but it should be revised based on the following major and minor comments before considering for Publication in Remote Sensing.
Major comments:
1. What does mean by “if AOD of a specific point was given”? , and it is not clear how Fig. 11(b) was plotted.
2. It is very hard to remove surface artifacts in the final AOD output especially using high-resolution satellite data. So, how the authors get the smooth AOD spatial distribution – did the authors have used any filter or interpolation? How the surface artifacts were removed?
Minor Comments:
L20-22: L36: DT and DB are the operational algorithms. It is better if the authors could also mention non-operational algorithm which has been used for this region or other regions.
L112: It is not clear how S and T were defined.
Author Response
Responses to Reviewer#3
We would like to thank the anonymous reviewer for his/her valuable comments that helped us revise and improve the presentation and the technical context of our paper. In the following, we addressed all comments and suggestions made by the reviewer. The corrections have been made in this revision are highlighted in blue. All numbered items (pages, equations, figures, and references) are consistent with those in the revised manuscript, excepted if otherwise stated.
In this manuscript, the authors modified the Dark Target method to retrieve AOD from HJ-1A/B data at 30 m resolution over Beijing. Overall, the manuscript is interesting but it should be revised based on the following major and minor comments before considering for Publication in Remote Sensing.
Major comments:
1. What does mean by “if AOD of a specific point was given”? , and it is not clear how Fig. 11(b) was plotted.
Author’s reply: It means that the AOD measurements of an AERONET site are known and can be used as prior knowledge for getting the optimal offset step i. If the AOD measurements of an AERONET site were used for this procedure, the measurements of other AERONET sites were used for validation. We have added a description for this procedure in Section 3.2. Please see Page 12, Lines 308-312 for more details.
The AOD retrieval accuracy can be significant improved once the measurements of an AERONET site were used as prior knowledge. The main reason for this phenomenon is that the estimation of offset step i of the regression intercept can be seriously affected by the issues of sensor calibration. Thus, the in situ measured AOD can be used as a calibration dataset for various of satellite imageries, get the optimal estimation of offset step i directly, and improve the estimation accuracy of AOD derived from satellite observations. We have added a discussion about it in Section 3.4. Please see Page 16, Lines 364-369.
2. It is very hard to remove surface artifacts in the final AOD output especially using high-resolution satellite data. So, how the authors get the smooth AOD spatial distribution – did the authors have used any filter or interpolation? How the surface artifacts were removed?
Author’s reply: In this study, for avoiding the effects of sensor noises, the HJ-1 A/B CCD data was firstly resampled to 300 m, and then the AOD was retrieved from the low spatial resolution data. Finally, the AOD was resampled to 30 m for obtaining a high spatial resolution dataset. We have to admit that this processing procedure reduced the real spatial resolution of retrieved AOD dataset. However, the image quality of HJ-1 A/B CCD AOD is still superior than that of MODIS 3 km aerosol product. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 370-375 for more details.
In this study, the inter-band correlation method is applicable for all kinds of land cover types (i.e., surface artifacts), not for dark target only. Therefore, the aerosol derived by this method are spatially continuous without data missing.
Minor Comments:
L20-22: L36: DT and DB are the operational algorithms. It is better if the authors could also mention non-operational algorithm which has been used for this region or other regions.
Author’s reply: Thank you for your valuable suggestion. We have provided a description about related AOD estimation algorithms in Section 1. Please see Page 2, Lines 52-62 for more details.
L112: It is not clear how S and T were defined.
Author’s reply: We are sorry for this mistake and have made correction for it in the revised manuscript. Please see Eqn.(6), Page 6, Line 168 for more details.
We would like to thank the reviewer for the positive evaluation of our work and for his/her detailed remarks that helped us improve the presentation of our work.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
In general, the revised paper has been sufficiently improved with respect to english and grammatical structure. The rst of my response will be focused on how well the authors addressed my major concerns.
1) The retrieval assumes a continental aerosol model. Unlike the MODIS DT retrieval algorithm using 3 bands including SWIR, which allows for more flexibility in dealing with diverse aerosol types, this is not the case here. The authors do not discuss this. Are the cases being chosen for comparison selected to be specifically 'continental like' based on AERONET Microphysical processes
Author’ reply: Thank you for your valuable suggestions and positive evaluation of our work. We have checked the English proofreading carefully, and rewritten parts of this manuscript. Please see the revised manuscript highlighted in blue for more details.
In this study, the atmospheric and aerosol types are considered as input parameters for 6S model. In fact, the aerosol lookup table can be simulated for a diversity of atmospheric and aerosol types (i.e., Continental, Maritime, Urban, Desert, Biomass, and Stratospheric). In this study, we just took the mid-latitude summer, continental aerosol type as an example for demonstration the ability and validity of algorithm with the measurements of AERONET sites in Beijing. For making it much clearer to understand, we have added a description about it in Section 2.4. Please see Page 6, Lines 174-175 for more details.
Reviewer reply: I realized that this specific question is not really a weakness. Using a fixed reasonable aerosol model such as Summer Mid Lattitude is a suitable assumption and is consistent with all major algorithms including MODIS for China. Most important, as seen in their statistical comparisons, the number of days (cases) analyzed is sufficient enough to assess the algorithm over the natural diversity of aerosol micro-physics.
2) Based on the general structure of the algorithm, the authors through figure 4 are left with a situation where different AOD retrieval values will occur based on the choice of the 'constant' regression coefficient which is hard to lock down on physical grounds so the authors allow different offset values indexed by counting parameter {i} so that the retrieval at this stage allows for a set of AOD{i} and the specific value obtained must make use of additional constraints that are not part of the used blue / green channels used in the algorithm.
For this purpose, the authors introduce 2 spectral ratio type constraints including an ACI constraint and an NDVI constraint which are supposed to pin down the offset parameter {i}, To that end, the authors plot in Fig 4 the ACI vs offset {i} and provide some very loose arguments on how the mnature of the plot can be used to constrain the {i} but it is not at all clear.
The second constraint uses the NDVI obtained from existing products as a contraint taking into account of the seasonal changes that can be expected.
My "guess" is that the value of {i} and the subsequent tau{i} are then used for atmospheric correction to quantify NDVI{i} which can be compared to constain the {i} parameter.
Unfortunately, this is conjecture since the authors do not provide enough detail and show this approach from beginning to end on a specific pixel. This needs to be improved.
Author’s reply: Thank you for your valuable suggestion. We have reorganized and rewritten parts of this section (Section 2.2-2.5 in the revised manuscript). Please see Pages 3-8 for more details.
Reviewers reply: Unfortunately, nothing in the reply or in the revised manuscript improves the situation regarding how the authors actually employ the ACI and the NDVI constraints to estimate the offset. Almost no changes were made to the discussion.
Again my points are
a) Based on their figure 5, the authors seem to "hint" that the optimal offset condition occurs when the blue channel reflectance ~ 0 which is when the ACI reaches its peak. If so, based on figure 5, are the authors saying that the optimal offset index is i=20 which corresponds to AOD ~ .92. If so, the authors should make this very very clear.
b) It is not clear how the authors use the 2 constraints together. Suppose the NDVI constraint gives an offset different than the ACI offset. What is done? Average the 2? Assume one constaint has higher priority than the other? What is the protocol? Any user would need to know this.
3) I am very dubious that retrievals can be done at 30m (single pixel) retrieval in any meaningful way. My prior experience in applying single pixel retrievals to Landsat 30m using MODID like algorithms result in strong 'noisy' variability on any AOD retrieval and significant averaging of the radiances are needed reducing the resolution. To convince the community, the authors should provide 'zoomed' in retrievals at 30 meter resolution where the AOD as well as the variability of the intermediate products such as the regression coefficients and the offset parameter can be seen and interpreted by the reader.
Author’s reply: A very good question. In this study, for avoiding the effects of sensor noises, the HJ-1 A/B CCD data was firstly resampled to 300 m, and then the AOD was retrieved from the low spatial resolution data. Finally, the AOD was resampled to 30 m for obtaining a high spatial resolution dataset. We have to admit that this processing procedure reduced the real spatial resolution of retrieved AOD dataset. However, the image quality of HJ-1 A/B CCD AOD is still superior than that of MODIS 3 km aerosol product. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 370-375 for more details.
Reviewers reply.
The authors provide more information on their procedure which is appreciated so the true spatial resolution is not single pixel 30 meters but 10 x 10 aggregated retrievals (or 300 meters).
The authors also provide useful intermediate regression products 1d and 1e which seem reasonable and reinforce that the offset is a strong function of view geometry. My choice would be to include this in the revised manuscript to help the reader but the authors chose to include only the figure 1a-c into their figure 13 which is better than before.
One issue I do not like is that the after the authors admit that the true retrieval is 300m resolution, they state that they resample back to 30 meters which I imagine is some simple interpolation. I find no value in interpolating a data set to resolution higher than the information content so the authors would do better by leaving. everything as 300 meters
4) It seems difficult to me that a single AOD 'calibration' allows for the offset correction which is then used for all pixels of the image. Based on that, when doing an un-calibrated image, that should mean the for the entire image, the offset parameter is the same even for different sun-view angles and land types.
Author’s reply: A very good question. The AOD retrieval accuracy can be significant improved once the measurements of an AERONET site were used as prior knowledge. The main reason for this phenomenon is that the estimation of offset step i of the regression intercept can be seriously affected by the issues of sensor calibration. Thus, the in situ measured AOD can be used as a calibration dataset for various of satellite imageries, get the optimal estimation of offset step i directly, and improve the estimation accuracy of AOD derived from satellite observations. We have added a discussion for it in Section 3.4. Please see Page 16, Lines 364-369 for more details.
Reviewer reply: The response and the authors reply does not remove the question of this procedure from my mind. Yes, the calibration differences of the instrument and the data used to build up the radiative transfer based band correlations means that every pixel should in principle be processed for an offset correction and as the authors demonstrate, the geometric view angle affects the offset parameter.
When a single AOD station is available to find for that particular pixel the correct offset to be used.
i_true, the authors seem to imply that that value i_true needs to be used for every pixel. But clearly, the i value is changed based on the geometric view angle (at least) so I do not understand that a single i_best fixes the biases. This would also imply a dramatic decrease in the processign time of the algorithm I imagine,
Perhaps the authors imply that the AERONET AOD calibration allows us to create an adjustment to the b coefficient and that adjustment is applied to the LUT directly and per pixel offset correction still needs to be applied. However, I do not see why that would be an improvement over the original scheme since the search for an offset should allow that correction to be made pixel by pixel.
5) Perhaps the authors may provide some expertise and comments on whether this approach can be applied to Landsat.
Author’s reply: Thank you for your suggestion. The method proposed by this study can be applied to various remote sensing sensors and application scenarios. Once the sensor has visible and NIR bands, the AOD can be used for retrieved. It can be applied to commonly used satellite data, such as Landsat TM/ETM/ETM+, MODIS and POLDER etc., and also has great advantage for retrieving AOD from the sensors without SWIR bands, i.e., Quickbird CCD, Chinese Gaofen series satellites sensors, and low cost cameras onboard of unmanned aerial vehicle (UAV).We have added a discussion for it in Section 3.4. Please see Page 15, Lines 381-386 for more details.
Reviewer reply: This is helpful and addresses the main issues of what the authors believe this algorithm can do on other platforms.
In summary, the paper has been improved in readability and construction but I still have issues with some key points that were not sufficiently explained to allow me to deeply understand the approach in certain details
Author Response
Responses to Reviewer#1
We would like to thank the anonymous reviewer for his/her valuable comments that helped us revise and improve the presentation and the technical context of our paper. In the following, we addressed the comments and suggestions made by the reviewer. The corrections have been made in this revision were highlighted in blue. For convenience, the comments for the leaving issues are repeated below in italics.
In general, the revised paper has been sufficiently improved with respect to English and grammatical structure. The rest of my response will be focused on how well the authors addressed my major concerns.
2) Reviewers reply: Unfortunately, nothing in the reply or in the revised manuscript improves the situation regarding how the authors actually employ the ACI and the NDVI constraints to estimate the offset. Almost no changes were made to the discussion.
Again my points are
a) Based on their figure 5, the authors seem to "hint" that the optimal offset condition occurs when the blue channel reflectance ~ 0 which is when the ACI reaches its peak. If so, based on figure 5, are the authors saying that the optimal offset index is i=20 which corresponds to AOD ~ .92. If so, the authors should make this very very clear.
b) It is not clear how the authors use the 2 constraints together. Suppose the NDVI constraint gives an offset different than the ACI offset. What is done? Average the 2? Assume one constraint has higher priority than the other? What is the protocol? Any user would need to know this.
Author’s reply: Thank you for your valuable suggestions. We have provided a detailed description about how the two constraints were used in the revised manuscript:
There are two approach for determination of the offset step i: (1) If the AOD measurements can be used as prior knowledge, the offset step i for an entire image can be determined directly by Eq.(13) in the revised manuscript; (2) If there are no available AOD measurements, optimal estimations of AOD and offset step (i) can be retrieved with the constraints of ACI and seasonal variations of NDVI. In this approach, the ACI was first used for narrowing the searching range of i, and then the seasonal variations of NDVI were used as references for getting the optimal AOD estimations.
In this study, the ACI was used for narrowing the search range of i, and avoiding the excessive atmospheric correction occurs. The value of offset step which indicates excessive atmospheric correction occurs were screen out.
The seasonal variation of NDVI was used as a constraint to further determine the AOD estimation. The NDVI was firstly calculated with the atmospheric corrected reflectance over the dense vegetation covered area. Then, the temporal variation of NDVI was fitted with a quadratic function. In the retrieval procedure, the optimal offset step i and AOD estimations can be obtained when the objection function reaches the minimum value.
We have added the description for it in Section 2.5. Please see Pages 7-10, Lines 207-245 for more details.
3) Reviewers reply: The authors provide more information on their procedure which is appreciated so the true spatial resolution is not single pixel 30 meters but 10 x 10 aggregated retrievals (or 300 meters).
The authors also provide useful intermediate regression products 1d and 1e which seem reasonable and reinforce that the offset is a strong function of view geometry. My choice would be to include this in the revised manuscript to help the reader but the authors chose to include only the figure 1a-c into their figure 13 which is better than before.
One issue I do not like is that the after the authors admit that the true retrieval is 300m resolution, they state that they resample back to 30 meters which I imagine is some simple interpolation. I find no value in interpolating a data set to resolution higher than the information content so the authors would do better by leaving everything as 300 meters
Author’s reply: Thank you for your valuable suggestions. Following revision were carried out in the revised manuscript: (1) We have decided to leave the final AOD estimation results with a spatial resolution of 300 meter. We have added a description about the resampling procedure in Section 2.1 and the flowchart (Figure 1). Please see Page 3, Lines 92-94 for more details. (2) We have also provided the spatial variations of inter-band regression coefficients (Figures 13e and 13f) in the previous responses as the reviewer’s recommendation. Please see Page 16, Line 388 for more details.
4) Reviewer reply: The response and the authors reply does not remove the question of this procedure from my mind. Yes, the calibration differences of the instrument and the data used to build up the radiative transfer based band correlations means that every pixel should in principle be processed for an offset correction and as the authors demonstrate, the geometric view angle affects the offset parameter.
When a single AOD station is available to find for that particular pixel the correct offset to be used.
i_true, the authors seem to imply that that value i_true needs to be used for every pixel. But clearly, the i value is changed based on the geometric view angle (at least) so I do not understand that a single i_best fixes the biases. This would also imply a dramatic decrease in the processing time of the algorithm I imagine, Perhaps the authors imply that the AERONET AOD calibration allows us to create an adjustment to the b coefficient and that adjustment is applied to the LUT directly and per pixel offset correction still needs to be applied. However, I do not see why that would be an improvement over the original scheme since the search for an offset should allow that correction to be made pixel by pixel.
Author’s reply: Thank you for your valuable comments. We are sorry for the expression about offset step i is not quite clear in the previous manuscript. For improving the presentation of this part, we have added description about why the offset step is a unique value for an entire image.
Based on the inter-band regression coefficients and the aerosol lookup table, the AOD can be retrieved from the apparent reflectance of HJ-1 A/B CCD. For a specific pixel, the atmospheric parameters at a prescribed solar/view geometry and different AOD levels can be obtained from the aerosol lookup table. Thus, the atmospheric correction procedures were performed with assuming AOD levels. The optimal AOD estimation can be obtained by search the minimum value of the following object function.
However, we found it difficult to get the optimal AOD estimation due to the systematic errors of the imageries, such as the calibration issues of HJ-1 A/B CCD data. Since the regression intercept b is not robust due to the systematic error, a correction for the intercept b is request by this study. The results of sensitivity analysis show that a change of 0.05 in the regression intercept (b) can lead to a change of about 0.2–0.3 in the retrieved AOD Therefore, an offset step i for an entire image was added to the regression intercept b for each pixel (with 0.005 as the interval, bi=b+(i×0.005−0.10), i=0~41) .
In this procedure, the offset step i should be predetermined for obtaining the optimal AOD estimations. There are two approach for determination of the offset step i: (1) If the AOD measurements can be used as prior knowledge, the offset step i for an entire image can be determined directly by Eq.(13); (2) If there are no available AOD measurements, optimal estimations of AOD and offset step (i) can be retrieved with constraints. Please see Pages 6-7, Lines 188-212 for more details.
We would like to thank the reviewer again for his/her detailed remarks that helped us improve the presentation of our work.
Author Response File: Author Response.pdf
Reviewer 3 Report
I would like to thanks the authors for providing responses to my comments. In my previous review, the authors were asked what is meant by “if the AOD of a specific point…” and the authors replied that “they have used AOD from an AERONET station as prior knowledge to improve the quality of the retrievals.” I have the following comments here:
1. It is recommended to clearly mentioned in the abstract that “if AOD is used from an AERONET site in the inversion method…”.
2. This approach, “the use of AOD from the AERONET site and validated with the other available AERONET sites”, is developed by Bilal et al. 2013 and it is known as “Simplified Aerosol Retrieval Algorithm (SARA)” and the authors have not acknowledged and cited it. Therefore, I would encourage the author clearly mention in the Sections 3.2 and 3.4 that the “the Simplified Aerosol Retrieval Algorithm (SARA) approach, i.e., to use the AERONET AOD as an input in the inversion method as a prior knowledge, was adopted (Bilal et al. 2013; 2014, and Bilal and Nichol, 2015) to improve the quality of the retrievals.
I. Bilal, M.; Nichol, J.E.; Bleiweiss, M.P.; Dubois, D. A Simplified high resolution MODIS Aerosol Retrieval Algorithm (SARA) for use over mixed surfaces. Remote Sens. Environ. 2013, 136, 135–145.
3. It should be mentioned in Section 3.2 that which AERONET site was used as an input and which were used for validation.
In response to another comment the authors replied that they have (i) resampled the data into 300 m resolution, (ii) AOD was retrieved, and (iii) then resampled back to 30 m resolution for validation. It is not suitable to discuss it only in Section 3.4, it should be mentioned in Section 2 where the methods are described.
In the previous review, the authors were asked to explain the expressions of S and T (Eq. 6) and the replied that see 168 for more details – But unfortunately, I did not find any details how the Eq. 6 was derived. How the mentioned coefficients were derived? More details about these expressions should be mentioned.
Author Response
Responses to Reviewer#3
We would like to thank the anonymous reviewer for his/her valuable comments that helped us revise and improve the presentation and the technical context of our paper. In the following, we addressed the comments and suggestions made by the reviewer. The revisions in the manuscript were highlighted in blue. All numbered items (pages, equations, figures, and references) are consistent with those in the revised manuscript, excepted if otherwise stated. For convenience, the comments of the reviewer are repeated below in italics.
I would like to thanks the authors for providing responses to my comments. In my previous review, the authors were asked what is meant by “if the AOD of a specific point…” and the authors replied that “they have used AOD from an AERONET station as prior knowledge to improve the quality of the retrievals.” I have the following comments here:
1. It is recommended to clearly mentioned in the abstract that “if AOD is used from an AERONET site in the inversion method…”.
Author’s reply: Thank you for your suggestion. We have changed the expression in the abstract to “If the measurements of an AERONET site were used as prior knowledge, the AOD retrieval results can be much more accurately obtained…”. Please see Page 1, Lines 21-22 for more details.
2. This approach, “the use of AOD from the AERONET site and validated with the other available AERONET sites”, is developed by Bilal et al. 2013 and it is known as “Simplified Aerosol Retrieval Algorithm (SARA)” and the authors have not acknowledged and cited it. Therefore, I would encourage the author clearly mention in the Sections 3.2 and 3.4 that the “the Simplified Aerosol Retrieval Algorithm (SARA) approach, i.e., to use the AERONET AOD as an input in the inversion method as a prior knowledge, was adopted (Bilal et al. 2013; 2014, and Bilal and Nichol, 2015) to improve the quality of the retrievals.
I. Bilal, M.; Nichol, J.E.; Bleiweiss, M.P.; Dubois, D. A Simplified high resolution MODIS Aerosol Retrieval Algorithm (SARA) for use over mixed surfaces. Remote Sens. Environ. 2013, 136, 135–145.
Author’s reply: Thank you for your valuable comments. We have added following sentences in Section 2.6: “In the validation approach Ⅱ, the Simplified Aerosol Retrieval Algorithm (SARA) approach (Bilal et al., 2013; Bilal et al., 2014; Bilal and Nichol, 2015) was adopted to improve the quality of the retrievals. In this approach, parts of the AERONET measurement data was used as an input for obtaining an optimal offset step i, the other measurements were used for validation.” Please see Page 10, Lines 277-282 for more details.
We have also added these references in Section 3.4: “The AOD retrieval accuracy can be significant improved once the measurements of an AERONET site were used as prior knowledge (Bilal et al., 2013; Bilal et al., 2014; Bilal and Nichol, 2015)”. Please see Page 16, Lines 402-403 for more details.
3. It should be mentioned in Section 3.2 that which AERONET site was used as an input and which were used for validation.
In response to another comment the authors replied that they have (i) resampled the data into 300 m resolution, (ii) AOD was retrieved, and (iii) then resampled back to 30 m resolution for validation. It is not suitable to discuss it only in Section 3.4, it should be mentioned in Section 2 where the methods are described.
Author’s reply: Thank you for your valuable comments. The responses for the two sub-questions were listed as follows,
(1) The purposes of the AERONET sites used in the validation approach were added in the revised manuscript (Table 4): In the year of 2012, Beijing_CAMS site was used as an input, Beijing, Beijing_RADI, and XiangHe sites were used for validation; in the year of 2013, XiangHe site was used as an input, Beijing, Beijing_RADI, and Beijing_CAMS sites were used for validation; in the year of 2014, Beijing site was used as an input, XiangHe, Beijing_RADI, and Beijing_CAMS sites were used for validation; in the year of 2015, Beijing_CAMS site was used as an input, Beijing, Beijing_RADI, and XiangHe sites were used for validation. In this procedure, Beijing_RADI site was not used as an input due to the limited number of valid observations. We have added the description for it in the revised manuscript. Please see Page 10, Lines 275-288 for more details.
(2) We have added a description for the resampling procedure in Section 2.1 and the flowchart (Figure 1). “In this procedure, the original HJ-1 A/B CCD data with a spatial resolution of 30 m were resampled to 300 m for minimizing the effects of sensor noises.” Please see Page 3, Lines 92-94 for more details. Because we finally decided to leave the retrieved AOD with a spatial resolution of 300 m, the procedure for resampling the results to 30 m was not performed any more.
4. In the previous review, the authors were asked to explain the expressions of S and T (Eq. 6) and the replied that see 168 for more details – But unfortunately, I did not find any details how the Eq. 6 was derived. How the mentioned coefficients were derived? More details about these expressions should be mentioned.
Author’s reply: Please see the attached pdf file. We have also added descriptions for solving these equations in the revised manuscript. Please see Pages 5-6, Lines 166-177 for more details.
We would like to thank the reviewer again for his/her detailed remarks that helped us improve the presentation of our work.
Author Response File: Author Response.pdf
