the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Using OMPSLP color ratio to extract stratospheric aerosol particle median radius and concentration with application to two volcanic eruptions
Abstract. We derive stratospheric aerosol microphysical parameters from Ozone Mapping Profiler Suite Limb Profiler (OMPSLP) satellite measurements using aerosol extinction coefficient ratios at two wavelengths (the color ratio), which is sensitive to the particle radius, and concentration. We estimate various sources of uncertainty in this technique including extinction coefficient measurement error, sensitivity to the size distribution width assumption, and the OMPSLP algorithm phase function error. We apply our algorithm to extinction coefficient measurements made by the Stratospheric Aerosol and Gas Experiment on the International Space Station (SAGE III/ISS) to verify our approach and find that our results are in good agreement. Our results also compare favorably to balloon borne particle size measurements and concentrations under ambient condition and 2019 Raikoke volcanic eruption assuming a lognormal particle size distribution width of 1.6. We also estimate the changes in aerosol median radius and concentration following the 2019 Raikoke and 2022 Hunga TongaHunga Ha’apai volcanic eruptions and the result is consistent with other retrievals published in the literature.
 Preprint
(2830 KB)  Metadata XML
 BibTeX
 EndNote
Status: open (until 07 Jun 2024)

CC1: 'Comment on amt2023267', Larry Thomason, 30 Apr 2024
reply
Review of ‘Using OMPSLP color ratio to extract stratospheric aerosol particle median radius and concentration with application to two volcanic eruptions’ by Yi Wang, Mark Schoeberl, Ghassan Taha
This is a resubmission of a paper that was submitted last year and ultimately rejected. I was brought in as an additional reviewer late in the review process by the previous editor due to widely divergent reviews. I spent several days going through the manuscript and provided a detailed accounting of the deficiencies of the manuscript. For whatever reason, my review is not posted on the AMT website, so I have posted as a supplement. In the last few days, a colleague noted that this manuscript had reappeared as a new submission. I must say as a prior reviewer who spent (as I said) several days reviewing the earlier submission, I am disappointed that AMT did not inform me that it had returned as it is incumbent on AMT to be sure that all earlier reviewers were notified of a resubmission.
The new manuscript is better but still exhibits many of the same issues of the prior submission. I will not do a complete review of this document, rather I will concentrate on a couple of issues that, among others, must be corrected prior to publication.
Issue 1
The basic algorithm the authors use is exactly the same as used by Yue and Deepak (1983). Since they have only 2 channels to work with, this is about as good as they can do. Using a fixed width is ok but the authors regularly misrepresent the findings in the reference material particularly those by Rieger, Wrana, and Knepp about how ubiquitous the value of 1.6 is. They could simply note that this value is roughly in the middle of the range normally inferred for SMLN fits and reference the same papers for the range but there are (and must be) negatives associated with this limitation. They could show this with plots that show the impact of using 1.6 when a value of 1.4 or 1.8 is closer to the true value (this sort of hides in Figure 3). I’d recommend that they focus on higher order moments like surface area density or volume density since they are contained much better by the observations, and thus more robustly inferred, than low order moments like number density and number density.
The big problem begins with the basic approach that locks in a onetoone relationship between color ratio and the inferred radius as shown in figure 1. There can be no variation in the inferred radius for the same color ratio wherever it comes from. This is why I am dumbfounded by the authors use of SAGE III as ‘validation’ for their number density/radius retrievals. Since they are employing the exact same algorithm on the SAGE III/ISS data, the only possibility for differences between the inferred SAGE III values and the OMPS values using their algorithm is if their color ratios are different. Since this is an attribute of the extinction coefficient measurements by the two instruments, comparisons of radius and number density are meaningless and have nothing to do with the algorithm and absolutely do not provide validation for their OMPS product. This was specifically called out in my previous review yet remains in this manuscript and, in fact, it is expanded upon. I see no option other than to remove this part of the discussion in its entirety. It could easily be misinterpreted by readers as actual validation. (The authors show a comparison Wyoming OPC during a fairly benign period when comparisons in perturbed periods must be available. These are periods more challenging to their algorithms and the nominal focus of their entire paper.)
Issue 2
In addition to the referencing issues mentioned above, the authors too often misrepresent the content of papers or use irrelevant references when much better ones are available. For instance, the authors state that ‘Our algorithm is similar to the color ratio method developed by Thomason and Vernier’s (2013) for SAGE II that has been also used for cloud identification in SAGE III/ISS data (Schoeberl et al., 2021; Kovilakam et al., 2023).’ None of these papers use the color ratio to infer attributes of the aerosol size distribution but solely use color ratio as an aid to the detection of cloud presence. There are literally dozens of papers which use color ratio to do this (any approach must use color ratios or the aerosol extinction coefficient spectral dependence), these should be used here. I don’t care which ones just use proper ones. When the authors discuss cloud clearing, they reference one of these papers but apparently did not absorb the nuances of cloud detection in limb data nor do they show that their method works well and basically just assert that it works. There are many further such issues like referring to Thomason et al. (2021) as inferring aerosol size distributions (it does not) and so on. I do not normally spend a lot of time on references but it was painfully obvious that they need to scrub their references carefully to ensure relevance and that they are properly represented. Their use of reference material is at best sloppy.
Issue 3
The authors still do not discuss the impact of known deficiencies of the OMPS aerosol products particularly at short wavelengths and after virtually any significant aerosol event in the stratosphere (e.g., volcanoes). The authors must be aware of the ongoing efforts (that they are themselves are doing and/or contributing to) to improve these products but these issues are barely addressed herein. It is well known that there are large differences between SAGE III/ISS and OMPS aerosol products during the HTHH period and other volcanic and smoke events.

CC2: 'Reply on CC1', Travis N. Knepp, 01 May 2024
reply
I agree with much of Larry's comments, but I disagree with the notion that a static distribution width of 1.6 is acceptable. I understand the limited information content of the OMPS measurement forces assumptions, but the value of 1.6 is only reasonable when we are at background conditions, which are quite boring. When the atmosphere is scientifically interesting (post eruptions or pyroCbs) is when this assumption introduces significant bias in this method as can be seen in the Fig. 10. I'll post a separate review of this paper shortly, but I wanted to comment on this aspect of Larry's review that caught my eye.
Citation: https://doi.org/10.5194/amt2023267CC2 
CC3: 'Reply on CC1', Mark Schoeberl, 04 May 2024
reply
While we appreciate Dr. Thomason reposting his older review for discussion, the resubmitted paper is substantially different from the manuscript he previously reviewed. We tried hard to address his and other reviewer’s comments in this version.
Comment 1. We absolutely agree the algorithm we used has been used before by solar occultation instruments, but we are applying here it to a limb scattering instrument. Yue and Deepak (1983) are referenced as are more recent papers using color ratio, for example Bourassa et al. (2008b) with OSIRIS measurements. We are aware of the threewavelength techniques developed by Wrana et al. (2021) and Knepp et al. (2024) can produce median radius and width of the lognormal size distribution. The three channel techniques make use of the SAGE 1.543 µm channel, not available on OMPS. That said, the twochannel techniques applied to limb scattering appear to be robust to the measurement uncertainty and less likely to filter out many pixels beyond the lookup table ranges. On the downside, two channel methods are sensitive to assumptions about size distribution width as we show.
We agree that the SAGE and OMPSLP color ratio retrievals would show no difference when the color ratios are the same. SAGE occultation measurements are the benchmark. We are careful not to refer to our SAGE comparison as validation, but we note that SAGE does provides a check on our method. We refer to the section as ‘Comparison of OMPSLP with SAGE’ and refer to this as a verification of our technique. Differences occur when the extinction values between the instruments differ. For actual validation, we did comparisons to the Wyoming balloon OPC, before the Raikoke eruption and after.
Comment 2. We will relook at our references again. Note that we have included more relevant recent papers by Wrana et al., (2021) and Knepp et al., (2022), etc.
Comment 3. The width of the distribution is fixed in the OMPS retrievals, and using SAGE, with more wavelengths than is available from OMPS, Wrana et al. (2021) showed that ~1.6 is reasonable for background aerosol concentration. This result is in agreement with studies by Rieger et al. (2018), and in situ balloon measurements. For volcanic distributions, 1.6 is too large for the initial eruption, and we used 1.2 from Duchamp et al. (2023, Fig. 2) for Hunga. The sensitivity of the retrieved median radius to the width is shown in the paper.
Unfortunately, color ratio methods using OMPS, cannot tell us much more than the median radius. SAGE with its 1.543 µm wavelength can produce some estimate of the distribution width (Wrana et al., 2021, 2023 and Knepp et al., 2024). If the distribution is bimodal, as occurs after a volcanic eruption, a ‘wide’ distribution and larger median radius will be inferred rather than a small particle mode and a large particle mode as seen in balloon measurements (Wrana et al., 2023). The Knepp et al. (2024) study suggests that even SAGE probably does not have large enough wavelength range to characterize a bimodal distribution.
Our paper tries to characterize the uncertainty in assuming a fixed width, and our results show that that uncertainty is large if the distribution width is significantly different from ~1.6. It might be useful for future OMPS retrievals to use SAGE estimates of the distribution width when considering volcanic eruptions or pyroCb events.
Yi Wang and Mark Schoeberl
Citation: https://doi.org/10.5194/amt2023267CC3 
CC4: 'Reply on CC3', Larry Thomason, 06 May 2024
reply
I thank Mark Schoeberl for replying to my comments. Reading these, I feel that my comments about the SAGE comparisons and the replies to it need further discussion.
The algorithm used in this paper dates to Yue and Deepak (1983) which is the first paper (that I am aware of) to attempt to infer aerosol size distribution from limb/occultation data. In the following 40 years, there have been literally dozens (hundreds?) of papers to attempt this using a variety of instruments with varying numbers of measurements of extinction coefficient (and similar quantities) and approaches. I've written a couple myself. All of these apply constraints to the solution space (the outcomes_ because none of these measurement sets contain a great deal of information about the underlying aerosol size distribution. The type and strength of the constraint has an important impact on retrieved values depending on characteristics of the data set to which it is applied and the constraints can even dominate the solutions. Low information content on aerosol size distributions is a feature of with SAGE II measurements of aerosol extinction coefficient (at 34 wavelengths) and to SAGE III with nominally 9 measurement wavelengths. It is also true with OMPS as the authors employ only 2 measurement wavelengths. It is exacerbated by measurement uncertainty and correlation among the measurement wavelengths. These retrievals require care and will always be limited in how well they can ever infer the size distribution and describing any of these processes as 'robust' is optimistic particularly without carefully evaluating the impact of the constraints applied.
That said, my concerns about the SAGE/OMPS comparisons isn't actually dependent on the algorithm applied. For the purposes of this comment, the algorithm can be considered a black box into which data from an instrument is injected and size distribution parameters are output. In any algorithm I've seen applied to these sorts of data, injecting similar data yields similar outputs. The outputs are not even required to make sense. Certainly, for the algorithm applied in this paper, the idea that similar data inputs yield similar output data must be true. It doesn't matter where this similar data comes from. Similar OMPS sets of measurements must yield similar output values. A set of measurements from SAGE III must yield an output that is similar to that produced by similar measurements from OMPS. Unless there's something very unusual in how the algorithm is implemented, I don't see how this can be any other way. We also know that at least in benign conditions (as in the example in Figure 7), OMPS aerosol and SAGE III aerosol are in rough agreement at some wavelengths. As a result, as I stated in my original comment, the only thing that the SAGE III comparison shows (e.g., Figure 7) is that the data from the two systems going into the algorithm black box in the spatial/temporal range shown is similar. It says nothing about the robustness of the algorithm being applied and it simply can't. I do not understand what the authors think these comparisons show. And despite carefully not referring to these comparisons as validation, it is very likely that the average reader will interpret comparisons which nominally show decent agreement as just that. Given that it is really relatively the SAGE comparisons are pretty meaningless except as an indirect assessment of how well the extinctions coefficient measurements agree, it really must be removed.
If the authors really want to use SAGE III data in this paper, a suggestion is to use SAGE III to test how sensitive the inferred mode radius and number density (SAD, etc.) from this algorithm are to wavelength pairs. With SAGE III, you could potentially test any number of such wavelength pairs with wavelengths as far apart as 449 and 1540 nm. The results could be interesting. I think Wrana did work similar to this.
For a paper that is focused on the Raikoke and HTHH eruptions, it is kind of surprising that the comparisons between SAGE and OMPS is for a very benign period (Figure 7), a partial comparison for Raikoke (Figure 8 with no number density) and nothing for HTHH. Why is that? There is plenty of overlap between the instruments throughout their lifetimes. It seems necessary that any comparison of SAGE III and OMPS retrievals are shown for nominal time period on which the paper is focused.
The OPC comparisons are Ok as validation. Though only for Raikoke. Certainly OPC data from HTHH exists (see work by Corinna Kloss, Vernier, and the POPS team). Why not extend this?
Citation: https://doi.org/10.5194/amt2023267CC4

CC4: 'Reply on CC3', Larry Thomason, 06 May 2024
reply

CC2: 'Reply on CC1', Travis N. Knepp, 01 May 2024
reply

CC5: 'Comment on amt2023267', Travis N. Knepp, 06 May 2024
reply
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt2023267/amt2023267CC5supplement.pdf

CC6: 'Reply on CC5', Yi Wang, 07 May 2024
reply
We greatly appreciate the thorough and detailed review by Dr. Knepp. Clearly he spent a lot of time on our paper and his remarks and suggestions will be taken very seriously in our revisions. Reviewing is a thankless job, and this review no doubt took some time to develop. It is apparent to us that some of our arguments are not clear, and that is on us. Again, we thank him for this review.
Citation: https://doi.org/10.5194/amt2023267CC6 
CC7: 'Reply on CC6', Travis N. Knepp, 07 May 2024
reply
Hello Dr. Wang! Thank you for the kind words. Indeed, reviewing can be a thankless job.
You commented that "It is apparent to us that some of our arguments are not clear, and that is on us." I am afraid it is worse than that; this is not a simple misunderstanding. I dare say that I understand your method quite well and it is wrong for the reasons I posted in my review. I raised several major concerns with your manuscript (starting with the original submission) and to date they have not been addressed.
Citation: https://doi.org/10.5194/amt2023267CC7

CC7: 'Reply on CC6', Travis N. Knepp, 07 May 2024
reply

CC6: 'Reply on CC5', Yi Wang, 07 May 2024
reply

CC8: 'Comment on amt2023267', Mahesh Kovilakam, 09 May 2024
reply
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt2023267/amt2023267CC8supplement.pdf
Viewed
HTML  XML  Total  BibTeX  EndNote  

236  80  24  340  11  11 
 HTML: 236
 PDF: 80
 XML: 24
 Total: 340
 BibTeX: 11
 EndNote: 11
Viewed (geographical distribution)
Country  #  Views  % 

Total:  0 
HTML:  0 
PDF:  0 
XML:  0 
 1