the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Parameterizing spectral surface reflectance relationships for the Dark Target aerosol algorithm applied to a geostationary imager
Mijin Kim
Robert C. Levy
Lorraine A. Remer
Shana Mattoo
Pawan Gupta
Abstract. Originally developed for the Moderate Resolution Imaging Spectroradiometer (MODIS) in polar, sun-synchronous low-earth orbit (LEO), the Dark Target (DT) aerosol retrieval algorithm relies on the assumption of a Surface Reflectance Parameterization (SRP) over land surfaces. Specifically for vegetated and dark-soiled surfaces, values of surface reflectance in blue and red visible-wavelength bands are assumed to be nearly linearly related to each other and to the value in a shortwave infrared (SWIR) wavelength band. This SRP also includes dependencies on scattering angle and a normalized difference vegetation index computed from two SWIR bands (NDVISWIR). As the DT retrieval algorithm is being ported to new sensors to continue and expand the aerosol data record, we assess whether the MODIS-assumed SRP can be used for these sensors. Here, we specifically assess SRP for the Advanced Baseline Imager (ABI) aboard, the Geostationary Operational Environmental Satellite (GOES)-16/East (ABIE). First, we find that using MODIS-based SRP leads to higher biases and artificial diurnal signatures in aerosol optical depth (AOD) retrievals from ABIE. The primary reason appears to be that geostationary orbit (GEO) encounters an entirely different set of observation geometry than does LEO, primarily with regards to solar angles coupled with fixed view angles. Therefore, we have developed a new SRP for GEO that draws the angular shape of the surface bidirectional reflectance. We also introduce modifications to the parametrization of both red-SWIR and blue-red spectral relationships to include additional information. The revised Red-SWIR SRP includes solar zenith angle, NDVISWIR, and land-type percentage from an ancillary database. The blue-red SRP adds dependencies on the scattering angle and NDVISWIR. The new SRPs improve the AOD retrieval of ABIE in terms of overall less bias and mitigation of the overestimation around local noon. The average bias of DT AOD compared to AERONET AOD shows a reduction from 0.082 to 0.025, while the bias of local solar noon decreases from 0.118 to 0.029.
- Preprint
(8877 KB) - Metadata XML
- BibTeX
- EndNote
Mijin Kim et al.
Status: final response (author comments only)
-
RC1: 'Comment on amt-2023-128', Anonymous Referee #1, 22 Jul 2023
This study presents an adaptive improvement to apply the DT method to ABI/GEOS sensors, which follows the general scientific method and principles, and the results are reasonable and scientifically significant. The study specifically points out and discusses the differences in observation geometry between geostationary and polar-orbiting satellites when applying the DT method. It further subdivides the impacts caused by underlying surface types, which is of great significance. These contributions can also assist future satellite aerosol retrieval algorithms and match with the journal's objectives. However, the existing article structure lacks clarity, and paragraph coherence is weak, requiring improvements in the manuscript's organization. Additionally, there are some technical issues that need clarification.
Specific Comments:
- In the end of Abstract, I recommend to add a sentence to describe the final validation results of the DT/ABI method, such as several evaluation indicators: R, bias, RMSE, and %EE.
- It makes me slightly confused that the authors use quiet large text to describe existing DT/MODIS method in section 2.1 from Line 91 to 191, including strategies, equations, and product details. Though the research is based on the DT method, too many reproductions of existing literatures will make the manuscript look like a student technical report rather than a scientific study.
- Line 220: How to screen cloud from red band? Is there any product or existing method? Please clarify or add references, since cloud will greatly impact on aerosol retrieval.
- Line 225. Using 0.86 to replace 1.24 is a good choose. I prefer to a similar indicator named ‘AFRI’ (10.1016/s0034-4257(01)00190-0), and it shows a good consistency (slope close to 1) with MODIS NDVISWIR and shows relatively good atmospheric resistance (10.1109/TGRS.2020.3021021). The discussion of the NDVI index selecting should be extended here because it is one of the key steps of the DT method, and this problem also usually appears when applying the DT method to other sensors.
- Line 250: ‘…. is greater than’ what?
- Line 255: why the spatiotemporal criteria are different between AC and validation? The motivation is unclear for some operational details.
- Section 3.1: From figure 2 (g-h), the distribution of scattering angle is quite similar between ABIE and MODIS, though the solar and sensor geometries are different. In the DT method, the estimation of SPR is based on the scattering angle, not the SZA, RRA, or VZA. Therefore, the potential impacts of these differences in observational geometry in retrieval are still not clear. In addition, (Line 290) I prefer to say that the MODIS SZA is relatively evenly distributed between 20 and 70; And there is a consensus that a high SZA is not conducive to the retrieval.
- Line 305-314: Yes, the discussion here is meaningful. This means we need to pay more attention to the details of the observational geometry than just the scattering angle. In addition, considering the particularity of the DT method, this probably also be only related to the instrument itself. For example, the AOD also has nonnegligible difference between MODIS/Terra and MODIS/Aqua. And in your study, the ABIE and ABIW also show difference in distributions of AOD bias as Figure 3. When applying the DT method to the AHI/Himawari (10.5194/amt-12-6557-2019), the dependence of AOD bias with time is not emphasized. So, the differences between different hardware are also large, and comparing with the time dependence, the overestimation of AOD (positive bias) may be more noticeable. In this case, it is necessary to re-formulate the SRP.
- Line 345: Using extrapolation should be cautious and the data availability needs to be checked (such as excluding negative values). In addition, from a statistical point of view, the use of MODIS AOD products is also reasonable, which can greatly expand the AC sample number.
- Line 355: This is partially due to relatively large uncertainties of radiative transfer code when SZA is large.
- Line360: Here is using the new SRP from AC-ref instead of the original MODIS SRP? Yes, AOD bias (overestimation) is suppressed but it still shows the time dependence.
- Figure 9: I find the points in ABIE are more discrete, especially between SWIR and Red bands. This is possibly the impact from large SZA.
- I wonder whether the relationship between Blue, Red and SWIR bands changes with SZA (similar to the representation in Figure 8-9). This is meaningful to study the time dependence of AOD bias.
- Line 438: The title should be bold.
- Figure 12: This figure is interesting and shows the difference of SRP dependence between OV and CV. But what worries me is that the divergence between the data points is so large that it seems to outpace the trend itself. In addition, the SRP dependence does not show a clear change with percent of OV or CV. This is possibly owing to the features of OV and CV, which are not significant enough (IGBP is a climatological classification product). By contrast, the urban percent is an opposite.
- Do authors calculate ‘R’ using Pearson’s coefficient? I recommend to try Spearman’s coefficient. Because the latter is more sensitive to rank and this may help to determine the SRP's dependency parameters (For discussion only, the author does not need to revise this point).
- I recommend to describe the new SRP estimation method (as Table 1) with equations in detail.
- Line 547: Why not using Land type data with higher spatial resolution? Because the original L1B data is ~1km. If you remove some original pixels when aggregating retrieval pixel for DT (~10km), the removed pixels should not be used to calculate land type percent.
- Figure 14: Yes, the key blue band surface reflectance underestimation leads to the overestimation of the overall AOD in the baseline test. But the time dependence of AOD bias still partially remains as Figure 16.
- Although figures 15-17 are useful to describe the results of this new method. However, there is still a lack of a total validation scatter plot to represent the overall retrieval results, error distributions, and before and after improvements, as many other studies have done.
Â
Citation: https://doi.org/10.5194/amt-2023-128-RC1 -
RC2: 'Comment on amt-2023-128', Anonymous Referee #2, 11 Aug 2023
This is a review for "Parameterizing spectral surface reflectance relationships for the Dark Target aerosol algorithm applied to a geostationary imager" from Kim et al. This manuscript extends the well known Dark Target (DT) aerosol retrieval algorithm to the geostationary (GEO) imager ABI on GOES-16 by modifying the so-called Surface Reflectance Parameterization (SRP). This key processing step of the DT algorithm allows it to constrain the spectral variation of surface reflectance as a previous step to the estimation of aerosol properties over dark surfaces. The topic of the manuscript is of high interest for the atmospheric community and very appropriate to AMT because GEO imagers such as ABI are becoming increasingly compelling for aerosol remote sensing thanks to their high revisit time and their increasingly advanced sensing performances. The experiments described in the manuscript are relevant and most of results are convincing. However, and although it reads quite well most of the time, the manuscript would benefit of a careful revision, especially in some sections that are a bit too long (with redundant information) and some others that despite being key are a bit unclear currently (e.g. Sect. 5.2.1). Furthermore, I am a bit doubtful on the application of the proposed solution to other GEO imagers (see major comment below), as I understand that the future goal is to apply DT to the current ring of GEO satellites in conjunction with the LEO heritage spacecraft. Please find some comments and suggestions to improve the manuscript before its acceptance in AMT.
Major comment:
- I agree that land surface reflectance in some VIS and SWIR wavelengths present some correlation with each other for vegetation and dark-soiled surfaces as described in Kaufman and Remer (1994). However, this correlation is known to be imperfect and although the current MODIS DT algorithm includes modifications of the original technique to account for the residual angular-dependent biases coming BRDF anisotropy (Levy et al., 2007) these inaccuracies seem to be much more complex to model for GEO data due to their specific geometric features. I acknowledge the effort of the authors to expand their SRP method to GEO data by adding dependencies to other variables, as this will make possible to have a consistent GEO-LEO aerosol product from DT. However, I have the feeling that this manuscript is showing the algorithmic limitations of SRP, as its adaptation to GEO seems to become a bit too much complex which may in turn compromise the robustness of DT. For example, I wonder how this technique will work for other GEO sensors (seeing land surfaces with other geometries and corresponding to other cover types) giving its dependency on the GOES-16 characteristics (already discussed by the authors in the "Conclusion" section). I would like the authors to elaborate on this point and on the potential ways of using a more general, less tailor-made methodology in the future. One idea would be to consider a full characterization of the spectral surface BRDF, which should be possible from GEO and especially with joint GEO-LEO data. One way forward could be to think of a "surface BRDF parameterization" instead of the current "surface reflectance parameterization". The consideration of the anisotropy of the surface reflectance might prevent the authors to consider an empirical method with so many dependencies as they are doing in their study. In this regard, I find interesting the discussion of the authors in P14 (L430-438) for example and think that this point would be worth discussing further.
Other comments (of varying importance):
-P2, L58: HIMAWARI is not written in capital letters and please do not forget Himawari-9, which is the operational satellite since some months now
-L60-62: it would be great if you could explain why we want to retrieve aerosols from the GEO orbit. Which is the advantage? Did some studies put in evidence that LEO-derived aerosol observation are not sufficient?
-Please consider mentioning other past and recent aerosol retrieval algorithms applied to GEO data, not only for ABI but also AHI, GOCI, SEVIRI, etc. GEO aerosol remote sensing is not something new!
-P3, L71-73: Do "NOAA’s aerosol products created from ABI" use SRP too? Why does this sentence appear here otherwise?
-L76: Raleigh -> Rayleigh
-P5, L138: new line is missing here
-Eq. 4 is missing "if NDVI..." in the three cases
-P6, L163: "a" radiative transfer (RT) codeÂ
-L179: ".-The algorithm" -> Replace "-" by " "
-L193: What happens with SRP performances for cover types that are not represented by AERONET sites?
-P7, L198: "which better matches the AERONET coverage observed by each ABI during 2019" Why 2019? You did not talk about the period of your study yet if I am not wrong.
-L206 and L210: ABI red band is at 0.64 or 0.63 um? And SWIR band, is it at 2.24 or 2.26 um?
-P8, L250: "AOD at 0.55 μm is less than 0.2" -> what happens with sites with generally higher AOD values? Do you have enough data to characterize SRP properly at these sites? Won't be these sites underrepresented in the final regression?
-L250: "Ångström exponent (AE) (0.44 – 0.675 μm) is greater than" -> please complete and explain why you filter AERONET data according to the particle size.
-L252: "The 193 AERONET site" -> sites. Where does the 193 figure come from? Is it the total of available sites? Or some selected sites? Showing a map of AERONET sites would be of interest here and would definitely help to better analyze Fig. 17 for which some stations seem to be missing.
-L255 and L256: you talk first of ±0.3° and then of ±0.2° -> this is unclear
-P9, L278: did you describe "FD" previously?
-Fig. 2: you do not specify the period of time covered by the data. Based on this figure, could you comment which geometry, LEO or GEO, is more suited to derive aerosol properties in principle?
-P10, 294: "Although both scattering angle distribution in Fig. 2(g) and Fig. 2(h) cover the same angular range" -> I am surprised with this comment. I have always had the feeling that GEO cover a broader range than LEO during the year. Are bins too wide to be able to see that perhaps?
-P10, L302: "We must remember that for GEO sensors, a particular ground site is always observed with the same viewing angle while the sun angles change throughout the day." -> redundant
-L304 "... and therefore each VZA matches up to a specific land cover type according to location" -> This is not true. VZA varies with distance to sub-satellite point, making concentric lines of equal VZA values.
-Fig. 3: the interpretation of the diurnal biases from the authors is arguable. Authors should try to explain why bias is greater at noon and dawn/dusk. And why the bias peak is shifted to the early afternoon for GOES-17? I recommend the authors to link these results, as well as others presented in their manuscript, to the well-known diurnal variation of aerosol information content in the case of GEO satellites (please see the works of Luffarelli et al., 2019 and Ceamanos et al., 2023; both in AMT).
-P11, L338: which is the impact on SRP of considering a continental aerosol model only? Why not trying to use the DT models instead? This seems like a critical point to me.
-P12: What about using ground-based reflectance data to validate the retrieved AC-reflectance over a few AERONET sites?
-Fig. 5: How do you explain the variable precision of your AC-method according to SZA? Is higher SZA affected by higher multiple scattering, thus giving a lower precision?
-P12, text on Fig.6 saying "modification of SRPs varying with SZA mitigates the bias peak at noon" -> this is right but we also observe a degraded accuracy at dusk/dawn. Diurnal evolution of bias stays similar, which may give you a spurious diurnal aerosol cycle at the end!
-P13, L375-378: already said beforeÂ
-L386: "there is increased scattering" -> do you mean "dispersion" or "noise"? How do you justify this increase?
-L386: "with significantly reduced correlation in the blue/red relationship" -> Fig. 7 only seems to assess blue/SWIR and red/SWIR, so how do you conclude on blue/red correlation?
-L390: "newer MODIS data" and "the Western Hemisphere" -> unclear
-P14, Fig.9: Would plots against SZA be clearer than against xi?
-L438: Sect. 5.2 title has not the right format
-P15, L450: "In fact, the definition of BRDF requires a fixed SZA, and that SZA varies throughout every day.". Did you mean "that VZA varies..."? And why is that? Can you provide a reference on this affirmation regarding BRDF definition?
-Sect. 5.2.1 is key but not easy to follow, especially P16. Please consider improving/clarifying it.
-General comment on the LC-based methodology: What will happen in front of spatial changes of LC? Is there a risk to have spurious AOD boundaries?
-Fig. 16 is convincing but I would like to see the same plots for a few sites with different information content on aerosols (cad. different surface reflectance and location/geometry). For example, what happens over the GSFC site with respect to Fig. 6b? A site in a more arid area (Midwest?) would be great too. Also, please detail the period of time that was considered to generate this plot.
-Generally on the results, it would be interesting to investigate the link between surface reflectance change and AOD change (between baseline DT and new DT), and all this according to scattering angle. I would not be surprised to see that AOD improves more for angles associated to a lower information content despite experiencing similar reflectance changes.
-Sect. 6.3: I would like to see some AERONET/DT density scatter plots, with some average statistics.
-Fig. 17: why some sites along the US East coast are not there while they are seen in the plots for GOES-17? Please comment on the missing sites and consider adding a map with all the considered sites to be able to see which are missing in each case.
Citation: https://doi.org/10.5194/amt-2023-128-RC2
Mijin Kim et al.
Mijin Kim et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 254 | 72 | 13 | 339 | 9 | 8 |
- HTML: 254
- PDF: 72
- XML: 13
- Total: 339
- BibTeX: 9
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1