Referee Report on “Identification of Smoke and Sulfuric Acid Aerosol in SAGE III/ISS Extinction Spectra Following the 2019 Raikoke Eruption”, by Knepp et al.

General Comments

This paper presents a method to distinguish smoke particles and sulfuric acid aerosols based on the analysis of the extinction coefficient at 1020 nm and of the slope of the extinction spectral dependence between 450 and 1550 nm, estimated by linear regression. The method is applied to two recent volcanic eruptions (Ambae and Ulawun) and two wildfires (2017 Canadian and 2020 Australian wildfires), as well as to the more complex case of Raikoke which was coincident with large wildfires in Russia and Canada.

The proposed method seems to be effective and provides interesting perspectives in the challenging problem of aerosol type identification. Its limits of applicability and limitations are clearly mentioned. The introduction is particularly well-written and well-documented. On the other side, the spectral behaviour of sulfuric acid, back (BC) and brown carbon (BrC) brings quite a lot of questions and sometimes some confusion related to the fact that sulfuric acid and BrC seem to show quite similar spectral behaviour. This aspect is not discussed in the paper, and it would be important that the authors take them into consideration while revising their manuscript.

Specific comments

Title: It is surprizing that the authors put the focus of their paper on the Raikoke eruption, since this event is only one of five cases under investigation and the most difficult case with the less successful results due to the complexity of the situation. They could consider changing it.

L. 49, p.2: The notation “VEI-7-8” is unclear. Please clarify.

L. 118, p.4: Does the removal of all events with errors higher than 20% have a significant effect on specific time/latitude ranges by decreasing the amount of data down to a very small number of events specifically in these time/latitude intervals ? If it is the case, could this possibly induce some kind of bias in the results shown later in the paper ?

L. 133-135, p.5: The users are probably using Level 2 CALIPSO data in the version 4.2. On the contrary, the Level 3 data are monthly averaged. This important difference between the data level as well as the time duration used for the averaging should be specified.

L. 142, p.5: It would be useful to define the concept of “attenuated scattering ratio” or to refer to some paper where it is defined.

L. 149, p.5: Do the authors mean “as early as September” ? It is hard to see any trace of any secundary plume in figure 2(c) and the indication “secundary plume” is moreover shown in Figure 2(d).

L. 185, p.7: The authors should justify their statement that BrC is more likely to be present in the stratosphere than BC, or provide a reference in this sense.

L. 187-189, p.8 and Figure 3, p.9: For the sake of clarity, the same notation notation (“1.”, “2.” or “(a)”, “(b)”) should be used here and in Figure 3. Also, the text and figure captions should be clarified. In the text: Is this simulation made for the 3 situations (sufate, BC and BrC) ? In the caption: What is the reference used for the normalization ? In figure (a), what is the mode radius ? In figure (b), what is the reference wavelength for the calculation of the spectral slope ?

Figure 3, p.3: It might look surprizing, also with respect to the data in Table 1, that the sulfuric and BrC behave quite similarly (although scattering dominates for the first one, and absorption for the second one) while the BC curve have a significantly different behaviour ? Do the authors confirm that there is no confusion between some curves ?

L. 204, p.8 - L.207, p. 9: If the attribution of the considered species in the 3 curves is correct, in view of the large uncertainty on/variability of the mode radius and taking into account the fact that sulfuric acid droplets have no reason to have similar sizes to BC particles, one has to imagine a large uncertainty around each curve, and the difference between sulfuric acid and BrC is unlikely to be really detectable using this method. This might be an additional reason in the presence of “false positives” in both volcanic eruption and wildfire cases.

Caption Figure 4, p. 11: Please repeat in the caption the relevant information provided in the figures. Character size is quite small in the figure and the information mentioned in it is difficult to read.

L. 214-216, p.9: The authors’ argument is not clear. In the four cases, the light blue points corresponding to the extinction ratio with the 1020-nm channel provides flat curves, in both volcanic (with dominating sulfuric acid) and wildfire (with dominating carbonaceous aerosols) cases. Their composition is thus quite different from the background in at least one case ! On the other hand, extinction ratio values change quite strongly with the 1020-extinction coefficient in the case of the 520:1550 ratio, for both volcanic and wildfire cases. What do they mean by this sentence ?

L. 223-227, p.10: The fact that a Pinatubo-like eruption cannot be assessed by the present method is not related to a difference of process (in all cases, SO2 has to be converted in sulfuric acid using the available water vapour and within some characteristic formation time), but to the size of the resulting particles: if the resulting particles are very large, the spectral dependence is flat and the extinction ratio is close to 1; if the resulting particles are not large with respect to the wavelength, a varying spectral dependence is found. Therefore, the explanation provided in L. 223-224 seems not the right one. Also, the role of ashes is not taken into account in the present discussion.

L. 240, 245-256, p.10: In view of the case illustrated in Figure 3, the linear regression is much more reliable if you don’t consider the 1550-nm channel. Why do the authors conserve this 1550-nm channel ? Starting again from the case of Figure 3, the value of the slope is likely to be very similar in the sulfuric acid and BrC cases.

L. 268-272, p.12: The authors should explain or show on z figure why the slope is more negative / flatter than the background slope for sulfuric acid / smoke.

L. 286, p.12: I suggest that the authors add the corresponding value of the depolarization ratio after “do not depolarize”.

Caption Table 3: The authors should specify what they mean by “Raikoke Primary” and “Raikoke Secondary”, or refer to the explanation given in Section 7.2.

L. 301-302, p.14: Where are these number coming from ? In Table 3, the fraction of misclassified events reaches a maximum of 62% at 24 km height for Ambae and 100% at 15 km height for Ulawun. Please clarify.

Caption Figure 7: Please complete the caption and describe all panels to make the figure self-explanatory.

L. 352, p.17: Where is the estimate “>81%” coming from ? From Table 3, the fraction of identified smoke events is >60% if all altitudes considered, and >86% up to 24 km height.

Figure 13: I suggest that the authors use another colour for the indication “LB” and “R”, which are poorly visible.

L. 404, p.21: Large particle have to grow from condensation nuclei to large particle by all successive microphysical processes (condensation, nucleation, coagulation). They are thus likely to need several weeks (up to one month) to become large particles. Per se, they are expected to be short-lived, but to appear later. The case shown in Figure 14 was measured on 30 June 2019, about one week after the eruption. Hence, isn’t it likely that these particles rather concern ash ?

L. 420, p.22: Do the authors mean: “either a mixture of sulfuric acid and ash or smoke” ?

L. 422-423, p.22: The authors try to distinguish sulfuric acid and smoke, but do not discuss the distinction between BC from BrC, although their respective spectral behaviours illustrated in Figure 3 look quite different. Actually, in view of the relative similarity between the cases of sulfuric acid and BrC, wouldn’t it provide a plausible explanation for many “false positive” cases in all cases where wildfires take place (Australian and Canadian pyroCb and Raikoke) ? It is noticeable that all these cases show a significant amount of “false positive” (see Figure 9, 10, 15, and 16) while both purely volcanic cases show only very few ones (see Figures 5-6).

L. 425, p.22: Citing altitudes of 19 and 20 km could be even more convincing.

L. 453, p.24: The statement is different here from above in the text (L. 199-202, p.8). The authors should replace “there is a chance for” by “the result is most likely to be”, or just repeat that the method is not applicable in this case.

Technical corrections

L. 58, p.3: Duplicated “has”.

L. 130, p.5: “which“ should probably be removed.

Caption Figure 1: I suggest that the authors reproduce the time and geolocation of the four events in the caption for the safe of readability.

This paper presents a technique to use SAGE III/ISS solar occultation measurements of the spectrum of stratospheric aerosol extinction coefficient to classify enhanced aerosol cases as smoke or sulfuric acid.   The main idea is to use the spectral slope of the extinction coefficient given that, under a set of assumptions, smoke generates a flatter spectral dependence than sulfuric acid. Four case study events are presented and then the technique is used to analyze measurements in the time period following the 2019 Raikoke volcanic eruption.

While this is an enticing proposition, the main concern already raised by Reviewer #2 (Mike Fromm) is valid.  Figure 3, which plots the Mie theory dependencies of spectral slope on particle size, shows that BrC and aerosol in fact have very similar spectral slope characteristics.  A shift of only about 50 nm in mode radius brings the two curves on top of each other.  (Not even accounting for changes is distribution width or multi-modal, gamma, etc., shaped distributions)  This means that BrC is essentially indistinguishable from slightly larger sulfuric acid droplets in terms of spectral slope when allowing for uncertainly in particle size. So while, yes, the Raikoke spectra have flatter slope than background this is consistent with both larger particles and smoke.  The authors are aware of this and state briefly regarding the Raikoke case: “Given the magnitude of this eruption, the spectra identified as smoke here may be the product of both ash and large particle formation.” The authors explain they are not suggesting the data be interpreted as either sulfuric acid or smoke but rather a mixture, but it is not clear how to untangle the effects of changing particle size.  I appreciate the context of the case studies provided by the authors and the interesting consistency of the results with the hypothesis that the higher altitude plume is contaminated by smoke. However, figures 15 and 16 do not clearly show a sulfuric acid main (lower) peak and a smoke dominated secondary peak (unless I am missing something).

Overall, I am not convinced that the results merit the main conclusion of the study that the method can discriminate between sulfuric acid aerosol and smoke even under the “applicable” scenarios in the SAGE III/ISS record.  If a revision is considered, it should show the technique is valid given realistic uncertainty about changing particle size. 

The authors present new means of classifying SAGE 3 iss data to separate between volcanic and smoke. The method is applied to data after the Raikoke eruption, an event that was accompanied by large wildfires in Siberia and North America. A good deal of the aerosol signal came from smoke according to the manuscript. I imagine that a layer algorithm like this could be very useful, but the authors do not present what the method is intended to be used for, and I find their method interesting, but am not yet convinced that it has enough precision. This has to do with a few issues e.g. represented in their classification. I therefore suggest a revision.

The authors did not discuss what the results of the classification method will, or could be, used for. Large wildfire smoke events may be possible to identify in periods of moderately volcanically elevated sAODs, but OMPS (UVAI, ext coeff) and Calipso (dep ratios, col ratios) already does this.  Are there any advantage of using SAGE rather than other platforms?

Can you be sure that it is only wildfire smoke and not something else? Carbonaceous components have been found in volcanic aerosol. Could it not be that one of the eruptions of Raikoke contained soot and organics, or that smoke in the area of Raikoke was entrained in the volcanic cloud?

Further, I found no difference in depolarization ratio for the assumed wildfire smoke and volcanic aerosol (after Raikoke). Why does it not show up as wildfire smoke in calipso’s depolarization ratio? You show information on particle sizes, but not on other particle properties. I think that the depolarization ratio should be shown here since it is a very strong indicator of smoke.

Section 5: In the beginning of this section, it reads that the slope was computed via linear regression. How did these regressions look and how well did they fit to the data? From the figures, e.g. Fig4, it looks like there is a large variance in the data. It is difficult for the reader to grasp this without some type of illustration of these regressions. Aren’t the widths of these distributions rather important for your classification? The standard deviations of these linear regression models could be used to distinguish between cases where the identification is more or less 100% indicative of one class, and cases where the data points are mixtures of smoke and sulfate. I think that this could be a means of telling whether the rising stratospheric aerosol after Raikoke/Ulawun/Fires are a mixture of smoke and sulfate or only smoke.

I think that the figures showing the altitude dependent slopes are really good illustrations to highlight where the different aerosol layers are located. Does this work well when separating background aerosol from low volcanic impact? In the analysis you had to divide data into altitude segments (since extinction coefficients in rising or descending air masses becomes pressure dependent). Would it be possible to normalize the data with pressure to get an altitude independent slope for each class (backgr, volc, smoke)?

Table 3: It looks to me that there are quite some misclassifications even at times and altitudes with large sample sizes. Starting with the Canadian fires 2017, 62% of the data at 14 km altitude were classified as sulfuric acid, and at the highest altitudes (23-25 km) 57-99% are misclassified as sulfuric acid. What would be the source of this sulfuric acid? I don’t know of any potential eruptions occurring in the first half of 2017. To me this indicates big issues with the assumptions used for the classification algorithm. The same issue occurs after the Australian fires 2019/2020, but only at the highest altitude shown (25 km). I would like to see how well the algorithm does above 25 km. It is evident in the SAGE 3 iss data that the smoke rose to >30 km, and some dense smoke layers in the v5.10 data lacked data below 27 km indicating too high optical depth in the line of sight to quantify the extinction. So these are not faint layers. It is difficult to interpret the classification results after Raikoke if these issues occur in the periods of known sources.

After Raikoke all the highest extinction coefficients (Fig 15) were classified as smoke. I find this surprising. Does this mean that a large fraction of the AOD elevation after Raikoke was actually caused by fires?

I wonder to what degree the small difference in refractive index affects the classification. The refractive index for black carbon differs quite from that of H2SO4. However, brown carbon and sulfuric acid has rather similar values in refractive index, indication that it is difficult to separate between the two.

Why did you not include a spectral slope for ash (Fig. 3), and in what way may this impact your classification?

The numbers in Table 3 don’t add up. I did not check them all but noticed the issue at Australia @25 km altitude (0.30 + 0.60).

L325-328, regarding Fig 7&8: You claim a rapid decrease in the slopes. I see a slope changing value over several kilometers. Isn’t this an indication of mixed sources?

Regarding the figures with calypso curtains, I suggest that you add curtains of the beta-532 signal as it is difficult to understand why there is a yellow feature in Fig 7b (same in Fig 8). The volcanic layers should be visible in beta-532.

L358: You write about an aerosol layer at 19 km altitude. It is actually visible in calypso images, but it is classified/misclassified as clouds. No cirrus should be present so far (~7-8 km) above the TP (even above the ExTL).

Section 2.1: Why was the data limited to 2 km above the TP? Was it to minimize cloud interference? And why not 1 or 3 km? Are there any risk of cloud interference that may disturb the classification?

Section 2.2: The lvl3 sAOD product has strong bias in the extratropics (Kar et al. 2019). Does this have any impact on the comparison with SAGE?

Caption Figure 3: I think that you should add the word ‘normalized’ to the ylabel.

There is strong gradient in the slope in Fig. 12d at 10-11 km altitude. Could it be clouds interfering? Also, no TP was marked in Fig.7&8. Is the TP height lower than what’s shown in the graphs?

The particle size distribution evolves with time, especially in the first month or two after eruption (or smoke injection). This should lead to increased variance in your data.