In this paper, the authors attempt to develop a new approach for retrieving bimodal lognormal stratospheric aerosol size distributions from measured and simulated extinction spectra.  However, the methodology is difficult to follow, and the novelty of the approach is not clear.  The basic idea of fitting extinction spectra to theoretical Mie cross sections is well-known, and ill-constrained given the complexity of aerosol size, composition, etc.  The main issue with this paper is that the authors do not rigorously develop the retrieval method and show the advancement beyond other methods like, for example, Knepp et al., AMT, 2024 and the associated publicly released SAGE particle size product.  The mathematical foundation for the retrieval with a clear case for robustness with respect to the retrieved parameters and discrimination of composition is required.  As part of this, the figures showing the results of the retrievals, Figures 5-10, need to be revised in a way that better shows the capability of the retrieval (with attention to figure creation, like for example in several cases the legend shows markers that correspond to a retrieval uncertainty and these are completely missing from the plot). 

General impression:

The paper deals with the aerosol products retrieved  from SAGEIII/ISS observations. The material is worthwhile to be published in ACP/AMT. The focus is on wildfire smoke observations. This seems to be the new aspect here and main motivation for the manuscript. The history of the development of techniques to analyze spaceborne multiwavelength aerosol extinction observations (with SAGE and other instruments) is long and must be better reviewed in the introduction to clearly state what are the new aspects presented in this paper. The highlight of the manuscript is the comparison with balloon borne in situ observations of stratospheric aerosols (of smoke). This should be better emphasized in the introduction.

Detailed comments: The used and carefully analyzed stratospheric  in situ aerosol observations are provided by the balloon team of the University of Wyoming. Why is there no acknowledgement statement? Furthermore, I would find it well justified if one or two members of this group would be  even co-authors. Section 5 (comparison of SAGEII/ISS products with Wyoming balloon observations) is clearly the strongest (unique) part of the manuscript.

Lines 33-39: To my opinion, the European contribution to wildfire smoke research is poorly represented. This is not acceptable for a European reviewer. Especially the contributions of the EARLINET lidar teams need to be mentioned. Besides Khaykin et al. and Hu et al. (in the references) further essential lidar publications are: Haarig et al., ACP, 2018, Baars et al., 2019, Ohneiser et al., ACP, 2022. Ozone depletion related to wildfire smoke events was more pronounced in the polar regions than at midlatitudes. The potential impact of ozone depletion by wildfire smoke was described and summarized by Ansmann et al., ACP, 2022, and mentioned already in the papers of Ohneiser, ACP, 2021 and 2022.

The paper of Baars et al., ACP,  2019 shows Canadian smoke distributions and aerosol properties over Europe (including November 2017) and should be used when discussing the SAGEII and balloon observations performed in November 2017. To that time the aged wildfire smoke was homogeneously distributed  over the northern mid latitudes so that comparisons can be carried out.

Lines 34-61: Here one should add the long-term studies of Ohneiser et al., ACP, 2022 (Australian smoke) and Baars et al., ACP, 2019 (Canadian smoke).

Lines 62-66: In the papers of Haarig et al., 2018, Ohneiser et al., ACP 2021, and especially in Ansmann et al., ACP, 2021 (on wildfire smoke) you will find particle size distributions of wildfire smoke showing a very pronounced accumulation mode. There is obviously no bi-modal size distribution with separate fine mode and coarse mode.  That should be included in the discussion, later on. Could be mentioned already here.

Line 68: ‘well suited’ is not appropriate, to my opinion! When using just forward modelling you need so many assumptions and these assumptions are significant uncertainty sources…..  Because of the potentially high uncertainties, I like the very convincing Section 5 (comparison with balloon observations). 

Line 80:  After the paragraph line 68 -79, one could provide a short review, i.e., provide a history of the methodology applied to SAGE observations to obtain aerosol products! Who was the first (reference), what were big steps forward (references), what is now your new contribution?

Line 81-89: I would appreciate if Sect. 5 is here better introduced. For me, it is the best section of the manuscript.

Section 2.2, lines 110-118: The Wyoming ballon team did a great job. Are all these balloon profiles in Figs. 11-17 simply in the mentioned data base? No specific preparation of balloon aerosol profile products for this manuscript was necessary? If the Wyoming team had to process their data, guided by your wishes, they should be co-authors!

Section 3, lines 128-177: The description of the methodology is given in this section. In Figures 11-15, so many different data analysis scenarios are shown (in Figure 11, eight different scenarios), but a clear description of all of these 8 different aerosol products is not given. It took me a long time to make me again familiar with the different scenarios. If possible, please create a small Table with all the different scenarios shown in Figure 11-17, maybe, name of the scenario and then the key information on contrasting input parameters, for example, different refractive index characteristics….

Line 141: ‘m’ is not explicitly introduced.

Lines 351-357: In the AERONET data base, the smoke shows a monomodal size distribution (Ansmann et al., ACP, 2021). And this (fresh or aged) accumulation mode belongs to the fine mode. Here, in this manuscript this mode belongs obviously to the coarse mode. Confusing!

Section 5

One should include the smoke products that were presented in the papers of Haarig et al., ACP, 2018 and of Baars et al., ACP, 2019 in the discussions in Section 5. Volume concentrations (Baars shows a table with input parameters, smoke particle density of 0.9 g/cm3) and effective radius can be compared.

Figure 14 can only be understood at the screen. Dark colors are used in the case of three size distribution. Hard to find out which color belongs to which profile. Please use different line styles (dotted, dash, solid, thick and thin lines) as well as clearly distinguishable dark and bright colors.

Almost the same in Figure 17: Please make use of thick and thin line styles.

Line 524: This is new for me that the wild fire smoke size distribution consists of two modes. This means, there are two sources of smoke particles! But in case of pyroCB there is definitely only one source. The AERONET sunphotometer-derived size distributions and the lidar-derived size distributions do not corroborate your hypothesis. You assume a bi-modal size distribution. Ok! But that does not automatically mean, that the two modes really exist.

Lines 589-591: According to Ohneiser et al, ACP, 2021 (MOSAiC observations) and  Ansmann et al, JGR, 2024  (doi: 10.1029/2022JD038080), there was not only the Raikoke event, there were also strong Siberian wild fires, feeding the UTLS height range with aerosol.

Lines 606-607: Also Baars et al., ACP, 2019 documents this long lasting stratospheric smoke event.

Line 611: There was not only aging and sedimentation, also lofting has an impact on the vertical distribution of smoke particles.

Figure 18: What about the large values measured in the second half of 2020 and in 2021 by LOPC? Are these values trustworthy?

Line 657: …display two modes within the accumulation mode … A mode is a mode, and does not consist of two modes.