The authors have substantially improved their manuscript; the methodology clarifications they’ve carried out alleviates most of my concerns. One significant concern remains, so I am recommending major revisions at this stage – I don’t think that this will take a significant amount of time for the authors to address, but needs to be rectified before I can recommend publication.
The authors have apparently removed information presented in the original manuscript, information which I suggested implied the smoke model was overestimating particulate concentrations, from the revised manuscript. This presents the appearance that the model did not create the high values shown in the original manuscript – from the responses to this Reviewer, this change was not a result of any changes to the model itself. I don’t think this is the best way to deal with the issue, nor is acceptable for GMD. Some examples, contrasting the original and revised versions:
(1) The original manuscript (line 636, section 3.3) mentioned that the distribution of PM2.5 concentrations had “a long tail extends toward extremely high values exceeding 1.E+07 ug m-3”. I notice that my attempt to copy this into the GMD on-line format in my original review made it come out there as “107” ug/m3, and in their responses the authors referred to this as 1070 ug/m3. However, the original manuscript stated this as 1E07. This mention of high values has been removed from the revised manuscript.
(2) In the authors original manuscript, their Figures 3 and 5 show on their upper scale, maximum model-generated SOD (AOD from smoke) values of 2502. On their Figure 6, maximum model values were between 1352 and 2700 (I assume here the maximum was 1352). In the revised manuscript, Figure 7 shows a scale where the maximum is given as “>=4.0” – which is certainly true, but obscures the actual upper end of the modelled values. Similarly Figure 10 in the revised manuscript has a maximum colour scale of “>4”. The maximum values of 2502 or 1352 have been removed, in comparison to the original manuscript.
(3) Their original manuscript Figure 7 PM2.5 distributions includes cases where the PM2.5 gets 7E4 ug/m3 (Figure 7(a)), 8E6 ug/m3 (7b),4E7 ug/m3 (7c) and 2E7 (7d).
The revised manuscript Figure 14 cuts off the maximum in the distributions at 1E3 (Figure 14(a)) and 1E4 (Fig 14 b,c,d). The revised manuscript Figure 15 includes the x-axis scale going up to 1E7 ug/m3, but unlike the original manuscript Figure’s logarithmic concentration scale, uses a linear scale in the vertical, with the result that only the lowest two bars in the frequency scale are visible.
In my review of the original manuscript, I noted that the upper end of model’s predictions for PM2.5 and for SOD are unrealistically high. In their response, the authors argued that a value of 1070 ug/m3 was not unreasonable. I agree that 1000 is within the range observed for actual forest fires such as the Landis et al reference I suggested. 1000 or even 10000 is possible. However, 1E6 or 1E7 I can’t accept as being realistic, without observational evidence quoted from the literature. Aside from Landis et al., there are other observation studies where high values in prescribed burns (i.e. measurements right at the burn location) reach 2526 ug/m3 (Gili et al., Env. Poll., 368, 125660, 2025 https://doi.org/10.1016/j.envpol.2025.125660), four orders of magnitude below 1E7.
The authors argue that spatial averaging may be the cause of the mismatch between the high AOD values generated by their model and the satellite observations. I agree that this is certainly a factor making the direct comparison difficult, but does not provide a justification for the very high values generated by their model. Again, turning to aircraft and ground-based observation studies of AOD, some examples that I’ve seen (references follow) quote maximum values ranging from 5 to 17.5.
Shinozouka and Redemann, ACP, 11, 8489-8495, 2011 : https://acp.copernicus.org/articles/11/8489/2011/ AOD values up to 10 observed in aircraft observations within 20km segments (their Figure 1a).
Kassianov, E., Flynn, C.J., Barnard, J.C. et al. Radiative impact of record-breaking wildfires from integrated ground-based data. Sci Rep 15, 8262 (2025). https://doi.org/10.1038/s41598-025-85103-1. Values up to 16 observed (their Figure 6b).
Eck et al., JGR Atmospheres, 2019, https://doi.org/10.1029/2018JD030182 . Values up to 14 observed (their Figure 6)
Daniels et al., Sci Tot Env, 921, 2024, . https://doi.org/10.1016/j.scitotenv.2024.171122 , max aeronet values of 5 (their Figure 3).
Paton-Walsh et al., JGR, 110, 2005, https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2005JD006202 max values of 6 based on surface observations.
Frausto-Vicencio et al., ACP 23, 4521-4543, 2023, FTIR quite close to a fire, max of about 17.5 (their Figure 3). https://acp.copernicus.org/articles/23/4521/2023/
These values would be considered unusually large in the measurement literature, yet are 100x smaller than the maximum SOD generated by the authors’ model.
My two main points here: (1) the maximum PM2.5 and SOD values generated by the authors model are unrealistically high, and (2) in the revised manuscript, these high values have been obscured by the manner in which the material has been re-presented.
In order for the manuscript to be publishable, the authors need to do the following (these are relatively easy to implement):
(1) The revised manuscript Figures 7, 9, 10 need to revert back to the original mode of presentation, which includes the model-predicted maximum value at the top of the colour bar scale, not an upward arrow with >=” as the upper number.
(2) The revised manuscript Figure 14 needs to revert back to the original x-axis scale showing all of the model values, i.e. including the very high predictions.
(3) The revised manuscript Figure 15 needs to revert back to the original logarithmic vertical axis scale. An alternative, if they want to continue to make their point about the relative performance of the model for high/medium/low PM2.5 concentrations, would be to make Figure 15 a two-panel plot, with (a) being the linear vertical axis and (b) being the logarithmic one.
(4) In addition to these changes, the authors should incorporate a new paragraph following Figure 15, acknowledging that the model occasionally predicts both PM2.5 and SOD levels that are much higher than observations (referencing the corrected figures, and also referencing the above observation references). If the authors can provide surface observation-based references which show that PM2.5 levels of up to 1E7 ug/m3 or AODs up to the 1000’s have been recorded, they should also be included in this paragraph. If the predicted maximum values can’t be justified based on observational evidence, then the authors may wish to speculate on possible reasons why the predicted maxima are so high (see my note below regarding advection mass conservation algorithms for the advection component of reaction-transport models). However, if they can provide no observational evidence for the occasional very high levels being realistic, the authors should acknowledge this may be a problem with their modelling system, and something that should be addressed in future work. They should also point out that confirmation of such high values should be considered for future measurement studies.
I have a few comments and corrections on the revised manuscript:
Lines 155-160, and line183-187: A suggestion, possibly for future work: reading this description, I wondered what procedure is in the RAMS or BRAMS models to deal with conservation of mass of advected tracers. I’ve seen cases where mass conservation can give rise to very rapid growth of advected tracer mass at individual model grid cells – often for particulate matter . The typical situation is one where a large spatial gradient in concentration is created and maintained by a “chemical” process such as emissions or deposition. The advection algorithm can generate negative numbers due to this gradient, requiring clipping of the negatives, and a mass adjustment to maintain the net field mass. I’m wondering if the authors’ extremely high PM2.5 levels at single gridpoints might be related to this issue, since it can erroneously add very large amounts of mass (many orders of magnitude) to a single gridpoint in a model domain, over the course of a few hours. It would be worth the authors doing a basic mass check of the mass in the model domain (add up the mass in the model domain in each time step, do the difference between steps, and compare the difference to the emissions flux into the domain) – is mass being conserved? Some references on mass conservation in air quality models include:
Aranami, K., Davies, T., and Wood, N. (2015). A mass restoration scheme for limited-area models with semi-lagrangian advection. Q. J. R. Meteorol. Soc., 141:1795–1803.
Bermejo, R. and Conde, J. (2002). A conservative quasi-monotone semi-Lagrangian scheme. Mon. Wea. Rev., 130:423–430.
Sørenson, B., Kaas, E., and Korsholm, U. S. (2013). A mass-conserving and multi-tracer efficient transport scheme in the online integrated Enviro-HIRLAM model. Geosci. Model Dev., 6:1029–1042.
Zerroukat, M. and Shipway, B. (2017). ZLF (Zero Lateral Flux): a simple mass conservation method for semi-lagrangian-based limited-area models. Q. J. R. Meteorol. Soc., 143:2578– 2584.
Section 2.1.2.7 Suggest instead of “what we mean by” use “definition of” in the title for the section.
Line 444, suggest adding a linking sentence like “The generated emissions are then distributed in the vertical as described in section 2.1.4.”
Line 480: suggest specifying which of the two modes is being examined in the paper, in this sentence.
Line 600. Suggest mentioning that “the complex refractive index values used will be discussed in section 2.3.8”, here.
Line 610: a range is given for the kappa value for fresh smoke – what was the number used?
Line 617-618: Rather than the range, what value was used in the model?
Line 637, section 2.3.5 title: not clear why “no circularity” appears in the title. Remove.
Line 683: I was a bit surprised at the low values for the complex portion of the refractive index used to the different wavelengths. The authors made mention elsewhere that the black carbon portion had a relatively small impact, but that may be due to the more generic complex refractive index values used by the authors. Cf. Curci et al. 2015, Table 2, BC column, where complex refractive index values for black carbon are between 0.71 to 0.44.
Line 828 to 830: The two sentences starting with “Have the advantage” are not clear (and the first sentence is incomplete). Please clarify.
Liners 940 – 1100 and Table 4. I liked this section (the stats analysis greatly improved the paper) though somewhere it needs a sentence at the start to explain that the modelled SOD values can be used to potentially explain the AOD contributions, under the assumption that the modelled SOD values are accurate. Last part is important; the model is a useful tool for attributing potential contributions… ->under the assumption that the model values are correctly representing those contributions<-.
Line 1022: sentence ending “as seen upper.” is unclear; please clarify.
Comment at the end of line 1029: To a certain extent, the better performance in the late afternoon may reflect on the strength of the “signal” from smoke: that corresponds to the usual peak in combustion in the diurnal cycle of the fires, so that’s when you’d expect to see the biggest contribution of the fires towards the AOD.
Figure 8. Include R^2 values on these plots.
Figure 11: include R^2 value on these plots.
Figure 13: include R^2 on these plots.
Line 1195: “compositional contrasts, rather than parameter uncertainty”: I’m wondering how sensitive the model results are to the complex refractive index values used, black carbon portion in particular. Ditto line 1265 to 1268; maybe it would be useful to compare typical complex refractive index values for fires versus other aerosol types in the SI and mention here. Line 1275 is a bit more clear, mentioning fixed refractive index as the issue – maybe move this up to line 1195 hence defining what’s meant by parameter uncertainty? I would think of the complex refractive indexes as a parameter.
Line 1286: reaching 1E7, that should be. See earlier comment regarding scales and mentioning full range of the model values in the text.
Line 1318: This section would benefit from a lead-in sentence to introduce the topic to the reader, e.g. “We next examine the impact of wildfires on atmospheric stability.”
Line 1318 to 1370 is good background information, but I was wondering if it would be better as a shorter summary here, with the details moved to the SI.
Figures 17-19: Figure caption should mention units of CAPE and CIN isolines. Are the times on these Figures UT.
Figures 20-23: These figures are very busy and need some corrections and/or simplifications. (a) The isoline contour interval apparently is in real numbers, but the numbers on the plots are in integers, with the result that there are multiple contour lines with the same label. I suggest that a larger contour interval and/or real numbers for the labels should be used here. Also, I think these plots would be much easier to understand if the isolines were moved to a third row of panels below the current two. It also might be worthwhile to plot the no fire temperatures on the left, and the change in temperatures when fire is included, on the right, so the viewer can more clearly see the changes resulting from the fire emissions.
Spelling, etc.:
Line 41 “terrains” should be “terrain”
Line 513. Reference has “n.d.” instead of a date. No date?
Line 514. Two periods.
Line 1421: space needed before start of new sentence. |
The manuscript by Menezes et al. proposes a modeling methodology with BRAMS to provide the simulation of fire spread, smoke injection height, fine mode aerosol concentration and radiative impacts and thermodynamics impact. In order to do so, BRAMS model was updated, including the integration of a crown fire spread behaviour model, which allows the simulation of fire propagation on standing trees, and the implementation of fire radiative power calculations in the SFIRE model. The study is interesting and it fits the journal scope. However, major revision is necessary in order to improve the final version and to clarify some aspects.
First of all, some of the results are compared with MERRA-2 reanalysis products. It is mentioned (lines 298 and 299) that monthly mean results of MERRA-2 products were compared with AERONET retrievals. Considering SSA in particular, was the MERRA-2 result also compared with AERONET retrievals? As detailed below, some inconsistencies were observed. Finally, please, consider increasing the font size of Figures 9 and 10 and verify if the isolines of CAPE in the vertical profiles of the Figure 9 are correct. I had difficulty in interpreting the results just looking at the figures.
Specific comments and technical corrections:
line 42 - GFED, CAMS-GFAS - What do the acronyms mean?
line 58 - replace “originates” by “originated”.
line 68 - include “the” in “increases the heat release…”
line 84 - replace smoke-related aerosols by smoke-related aerosol optical properties.
line 106 - Give the meaning of the acronyms CPTEC and USP.
lines 166, 167, 171 (Eq. 1). Please, verify the subscripts of I and R (initialization x inicialization, respectively). The correct answer should be initialization, unless the authors have a reason to differentiate them. If so, a brief explanation is necessary.
line 400 - please add references discussing the spectral region where OC and BC present higher absorption efficiency.
lines 408 and 411 - Use of AOD x AOT. I recommend using only AOD - aerosol optical depth. Please, check the manuscript thoroughly. Also, in lines 411-412, the authors mention that MERRA-2 estimates AOD at 500 nm, while BRAMS calculates SOD at 550 nm. Given that smoke optical depth varies spectrally, please, clarify if the comparison between these variables was made at different wavelengths (500 nm x 550 nm).
lines 452, 720 and 751 - is there any reason to include all the authors of Menezes et al. (2024) paper? Please, just refer to Menezes et al. (2024).
line 485 - the correct acronym is AOD, instead of TOA. Please, refer to the previous comment.
line 486 - what do you mean by biomass fuel models?
line 487 - it is not only during the flaming phase that aerosol particles are emitted, but during the combustion process. Please, rephrase.
line 520 - add a dot signal after wavelength.
line 521 - rephrase “Understanding the spectral behaviour of smoke aerosols is essential for interpreting optical depth measurements” to “Understanding the spectral behaviour of smoke aerosol optical properties is essential for interpreting optical depth measurements”.
line 522 - what do the authors mean by spectrally integrated SOD?
Figure 3 - Please verify the top numbers in the vertical colorbar (2502.00 and 5000.00) representing SOD values. From the maps, MERRA-2 AOD higher values were observed in the northern part of the region, while SOD highest values were observed between 39.8° and 40° N at 15:00, moving to the north later, reaching 40.3° N at 21:00, when both AOD and SOD presented similar patterns (six hours later only). From the discussion presented in lines 542 to 552, how do the authors explain the high AOD values, above 1.3, further north at 15:00? As discussed, if MERRA-2 did not fully capture the peak of the fresh fire emission, shouldn’t we expect low AOD values at 15:00 everywhere in the map? Or does that mean that MERRA-2 fire source is located further north? From the color gradient, it seems that the peak of AOD from MERRA-2 is further north outside the presented map. Maps at 15:00 in Figure 4 seems to confirm this.
lines 579-580 - Even though OC/BC ratio is higher during the smoldering combustion phase compared to the flaming, OC concentration is always higher than BC for most of the vegetation types, independently of the combustion phase. According to the review by Reid et al. (2005), the exceptions are forest debris and herbaceous fuel.
Figure 6 - The numerical scale of the vertical colorbar must be verified (Simulated single scattering albedo). SSA can vary only between 0 and 1. The top left map (from 15:00) shows lower SSA values from MERRA-2 in the southeastern portion of the map, increasing towards the northern region. Maps generated for later times also show lower SSA values in the southeastern region. Does it mean that MERRA-2 is not reproducing the smoke event accordingly? If not, maybe it is not a good reference for comparison.
lines 687-688 - Please add the references of the mentioned studies.
Figure 8 - It is not clear what the authors mean by "Fire-weighted smoke absorption", whose value can reach 60000 W/m2, according to the presented scale. Sertã time zone is UTC + 1, thus, the absorbed solar irradiance should be close to zero at 17:00 UTC and zero at 21:00 UTC (i. e., no absorbed irradiance, since no solar radiation is available), as shown in the map of No Fire - Fire change in downwelling flux. In the longwave spectrum, by contrast, please confirm if the correct variable is "Fire-weighted smoke absorption” or “Fire-weighted smoke emission", i. e. the irradiance emitted by the smoke plume due to its higher temperature compared to the surrounding environment.
line 735 - replace "shown” by "shows”.
lines 737-738- Discussing Figure 9, it is mentioned “while CAPE and CIN isolines are superimposed: dashed yellow lines indicate the fire simulation and solid orange lines represent the no-fire scenario”. But in the legend, it says: “Dashed black lines indicate the fire simulation, while dashed orange lines represent the no-fire scenario”. What do the authors mean by “superimposed” in the context? Looking at Figure 9, I couldn’t identify the superposition of CAPE and CIN, since it seems they were plotted separately. How can one distinguish the differences of Fire x No-Fire for CAPE in the profiles? Moreover, it is very difficult to read the information in Figure 9, as the font size is too small (the same for Figure 10). Please, consider increasing the font size.
line 764 - the mentioned wavelength is 400 nm, but in the legend, it says 700 nm.
lines 766 to 768 - The spectral dependency is also a result of the smoke particle size distribution, concentrated in the fine mode.