Closed-Form Method for Atmospheric Correction (CMAC) of Smallsat Data Using Scene Statistics
Round 1
Reviewer 1 Report
Review for applsci-2342003
The authors describe a statistical method to account for the effect of aerosol light scattering in visible satellite imagery, or atmospheric correction, as they put it.
The proposed method is aimed at being fast and simple and specifically targeted at smallsats.
Comments:
- The conventional approach for this problem would be to application of a radiative transfer solver.
- Besides aerosols, visible and NIR light is also affected by trace gas absorption while traveling through the terrestrial atmosphere.
- line 35f: ice crystals and droplets are not aerosols. They form separate categories.
- line 59f: This is a good point. Accurate calibration of smallsat sensors is definitely a problem.
- line 100: The workflow definitely does add uncertainty in all cases.
Author Response
Responding to “Can be improved” relative to adequate method description. Significant additional explanation is provided for CMAC in the Methods and Materials section. This includes the mathematics for reversing atmospheric effects to deliver surface reflectance and a step by step reconstruction for generating the atmospheric index (Atm-I) model.
“Besides aerosols, visible and NIR light is also affected by trace gas absorption while traveling through the terrestrial atmosphere” True, however, through CMAC this is dealt with as a lump-sum as an atmospheric index model (Atm-I). The CMAC correction provides a “see it (i.e., Atm-I magnitude), correct it“ approach. Also, the narrow NIR band of Sentinel 2 8a is used that is resistant to water vapor absorption. Variability of trace gases does not appear to be a problem in this approach because CMAC is proving more accurate in comparison to methods that attempt to take trace gases and water vapor into consideration. Since no smallsats can presently measure such trace gases and even research grade satellite have problems with these measurements, at least for now these influences can be ignored.
line 35f: ice crystals and droplets are not aerosols. They form separate categories. True, however, while water droplets (thin clouds and fog) and ice crystals (cirrus) may be separate categories, other than diffractive effects, CMAC can remove those effects. Their interaction with transmitted light is very similar to aerosol particles. These have been removed from the sentence you reference.
Reviewer 2 Report
The manuscript entitled “Closed-form Method for Atmospheric Correction (CMAC) of Smallsat Data Using Scene Statistics” by Groeneveld et al. aims at proposing a new scheme for working in near real-time to convert top-of-atmosphere reflectance (TOAR) directly to surface reflectance based on image statistics specifically applied to smallsats. The topic of the manuscript is of interest to the smallsat processing as issues still remain in automated atmospheric correction without knowledge of sensor radiometry.
General comments: The manuscript is well written, the figures are good, there are few typos (see specific comments). I believe the references could be really enhanced including more work on atmospheric correction algorithms and validation, especially articles about application for smallsat.
For all those reasons, I recommend major revision.
I think English should be checked carefully by native speaker.
The analysis if existing AC algorithm for retrieving land surface reflectance could be more comprehensive. The authors have summarized the existing methods for retrieval of land surface reflectance, containing radiative transfer modeling, image-based dark object subtraction methods and using multi-temporal data sets for aerosol and surface reflectance retrieval, which illustrate the essential unreliability of the method of using multi-temporal data sets. However, the authors are hasty to conclude that no algorithm can work in near real-time without analyzing the uncertainties of algorithms radiative transfer modeling and image-based dark object subtraction methods. It is suggested that the authors further analyze the inapplicability of above two methods.
It is suggested that the authors add a brief summary of the other Sections for the manuscript at the end of Section 1.
How the atmospheric index (Atm-I) is calculated does not seem to appear in the manuscript. It is suggested that the authors describe the relevant calculation process and formulas in detail in the manuscript.
The description of the method could be improved. For instance, there are no explanation for some assumptions such as the diameter of the area of interest being five kilometers, the given AOI including clear and hazy conditions and the same surface reflectance for a given AOI. I would have like to see sensitivities studies to these assumptions.
The quality of figures within the manuscript should be improved. For instance, the labels in Figs. 9-14 are somewhat blurred.
The authors demonstrate the applicability of the new algorithm by comparing it with the standard algorithm Sen2Cor using Sentinel 2 images as an example. However, the authors target smallsats in the title, but the applicability of the new algorithm in smallsats is not proved in the manuscript. It is suggested that the authors add an example of the application of the new algorithm in smallsats.
Author Response
Relative to greater description of the existing methods and their applicability (the gist of several comments). Our intent for this paper was to concentrate on introducing the CMAC method and acknowledge the existing methods. Because this is our intent, a deeper dive into how RadTran works is off-mission, particularly because there is virtually no commonality between CMAC and the RadTran methods. Zhang [cited as 6 in the introduction] provides an in-depth look at the various radiative transfer methods, so if the reader is truly interested materials for additional study are readily available. Also, this paper has become very involved with lots of statistical analyses and including focus on the comparison to RadTran methods diminish the message and make the paper unwieldy. We do mention a series of well-known limitations for RadTran in a number of places in the text such as requiring knowledge of sensor radiometry and requirement for ancillary data. As is provided through statistical analysis and image examples, RadTran methods (Sen2Cor and LaSRC) are not accurately retrieving surface reflectance.
Note, however, that we have taken to task dark object subtraction through the illustration provided in an additional figure (Fig 5) that describes how the reflectance distribution changes as aerosol content increases. This new figure provides greater insight for the scaling aspect for data correction and why dark object subtraction is an inappropriate fix.
“The Authors are hasty to conclude no algorithm can work in near real-time” We agree and have revised the language.
I think English should be checked carefully by native speaker.” Done.
“How the atmospheric index (Atm-I) is calculated does not seem to appear in the manuscript. It is suggested that the authors describe the relevant calculation process and formulas in detail in the manuscript. We have added a great deal of detail to the description of the Atm-I model development that is provided in a step by step description. Also, we have added the equation that is derived from the conceptual model and used for surface reflectance retrieval. Actually, because of the method, “reversal” is perhaps a better descriptor since CMAC equation simply reverses atmospheric effects measured by the Atm-I model to retrieve surface reflectance.
“the authors target smallsats in the title, but the applicability of the new algorithm in smallsats is not proved in the manuscript. It is suggested that the authors add an example of the application of the new algorithm in smallsats.” We agree and have added language and figures in the Discussion that underscores that CMAC is appropriate for all satellites that have 4-band VNIR. We then provide an example of images corrected for Landsat 8 and for a Planet Dove satellite image. This allowed us to make the point that CMAC is readily transferred to any VNIR satellite. Incidentally, CMAC will work for any VNIR band, red edge band or the extra green and yellow bands planet is using on their newer smallsats. These additional bands offer the exciting potential to be a step closer to working with surface reflectance correction for hyperspectral satellites.
“It is suggested that the authors add a brief summary of the other Sections for the manuscript at the end of Section 1.” Done.
”The quality of figures should be improved” The figures are revised to enhance resolution.
Round 2
Reviewer 2 Report
The authors responded to all points raised by the reviewer and tried to revise the manuscript accordingly. The revision has been significantly improved and now appears to be suitable for publication.
