The article addresses a topic of interest for the dust and climate communities. It presents the improvements included in the COSMO-MUSCAT model to better characterize the size distribution of airborne minerals from arid regions and assesses their impact as compared to previous model versions. This will inform the scientific community on the ability of state-of-the-art models to reproduce a critical characteristic that shapes dust – Earth system interactions, its mineralogy. The article also presents two recently conducted campaigns that provide mineral fractions and elemental composition data, which constitute valuable assets for model evaluation. However, the mineral composition derived from one of the campaigns shows large uncertainty ranges for some minerals and sizes. While this aspect is acknowledged, it is not thoroughly considered in the model-observation comparison, hence weakening the validity of the conclusions. Additionally, the suggested reasons for model-observation discrepancies neglect in some cases relevant aspects (e.g. the assumptions made to derive the elemental composition, or the ability of the model to represent the dust cycle processes). Finally, some methodological aspects are not clear, which makes it difficult to properly interpret the results.
In my view, the article presents valuable information, but it still requires major revisions prior to its publication. I summarize below some of the suggested changes.
*** General comments ***
1. Clarification of methodological aspects:
The method followed to calculate and compare modelled and observed mineral and elemental fractions needs clarification, particularly regarding:
The mass fraction calculation: How do the authors compute the modelled mass fraction of different minerals to compare with observations? Is this the mass fraction over total dust? How do the authors handle scenarios where certain minerals are measured in the field but not simulated in the model?
The size range included: For elemental composition, BIN26 is explicitly mentioned as the upper size range for evaluation. Is this also the case for the mineralogy comparison? If not, how is the unspecified mineral fraction in BIN80 treated when calculating mineral mass fractions?
The assumptions behind the elemental - mineral conversions. The authors apply different values for the elemental composition of dust minerals when deriving mineral composition from the SEM analyses (Table A) versus determining elemental abundances from the modelled minerals (Table 3). These choices impact the evaluation results, and the underlying reasoning must be clearly justified.
2. Mineralogy from JATAC 2002 and treatment of observational uncertainty
The mineral fractions obtained from the JATAC 2002 campaign hold for some minerals and sizes large uncertainties. How is this information incorporated into the analysis?
For instance, several measurements in the JATAC campaign show a BIN26 clay fraction of nearly 100%. Are these values physically realistic? What is the corresponding amount of other minerals, e.g., quartz, at these specific measurement points?
For illite, the coarser bins show amounts up to 70%. In addition, the variations of illite fraction with size seem small (Figure 6), and they seem to tend toward higher mass fractions in larger particles, which is unexpected for a typically clay-sized mineral. The authors recognize challenges to identify illite in the SEM-EDX measurements (L662-664), however, the associated measurement uncertainty is not considered in the evaluation beyond this comment. Do the estimated mineral fractions add to 100% in each observational point? Does this limitation in illite detection affect the mass fractions of other minerals?
3. Evaluation procedure and interpretation
Why is the comparison with the mineral measurement compilation restricted only to the winter period (Section 4.2)? According to Table 2, both model versions were run for both winter and summer, and then the reason given in lines 488–491 does not hold.
Comparing mean biases and correlation values across experiments that rely on different samples does not seem fair (Table 4, L609-620). I would recommend comparing metrics that are calculated over the same measurement points.
In some sections, potential causes for discrepancies between model and observations are given, but these point many times to the soil mineralogy or the particle size distribution. However, the actual ability of the model to represent the specific regions where the dust plumes originated, their transport and deposition processes influence the modelled minerals (and elements) at the time-and-location of the measurements.
*** Detailed comments ***
Some of the sentences are vague and require further clarification. Examples are provided below, but I would recommend reviewing carefully the text, if possible, with the objective of synthetising it more and being more specific.
L14–15 (Abstract): “The elemental validation approach provides complementary constraints that expose discrepancies in internal mixing assumptions and reveal limitations invisible to mineral-only comparisons”. This is very difficult to follow unless the reader has already gone through the paper. I would recommend rephrasing for clarity.
L74-76: “The performance of the two schemes is assessed against a multi-level observational dataset. We first utilized a compilation of regional North African mineralogical measurements of in-situ aerosol samples to evaluate broad mineralogical patterns and enable a direct comparison between the implemented schemes. This is followed by a high-resolution comparison with concurrent in-situ observations from two campaigns in Cabo Verde [...]” The authors use three different datasets, why are these qualified as multi-level?. What do “broad mineralogical patterns” refer to? Are these geographical patterns? Does this dataset allow direct comparison between the two schemes but not the data from the campaigns? What does the “high-resolution” refer to? High temporal frequency? High spatial resolution?
L84–85: The sentence "The validation results are presented through a tiered approach categorized by analytical scale" is vague and only becomes clear later in the text.
L388-391: “In this study, one of the main objectives is to accurately represent the aerosol dust mineralogy in future modeling studies using measurement data. Hence, we adopt a classification scheme for mineral dust aerosols representative of the soil database as used in the model.” This is difficult to follow. How do the authors plan to use the measurements to accurately represent dust mineralogy? What does the classification scheme refer to? Please, rephrase and clarify.
L631-635: The authors describe measurements taken close to dust sources as having an "observational bias." If the model's resolution is too coarse to capture localized features, this is a model limitation, not a measurement bias.
L48–49 defines 2.5 µm as the size boundary between clay and silt, but in L254 this is set at to 2 µm. Please remain consistent.
L208–210: Dc needs to be explicitly defined within the main text.
L256: Typo; change "thse" to "these".
Section 2.4: This section contains redundant details regarding soil texture and soil mineral data sources that were already detailed in Section 2.3.
L509: The text mentions Panta et al. (2023), but the compilation described up to this point has only mentioned data from Perlwitz et al. (2015), which did not include Panta’s results. I suggest reorganizing this section by moving the details currently in L528–530 to an earlier position.
L611–612: The comment regarding the scope of feldspar fails to clarify what is actually occurring in Figure 5, which depicts mass fractions without size resolution. If the evaluation is done collocating model and observations in size, please, detail it in the methods.
L615: Does the underlying soil mineralogy distribution actually change here? This information should theoretically come from the soil mineralogy map, which is supposed to be identical for both modeling approaches. Please clarify.
L662–664: What is meant by a "localized observational trend"?
L675: “This suggests that shifts in the mineral PSD do not necessarily align with the composition of the parent soil.” The statement here is unclear. Why would be the changes in the PSD aligned with the composition of the parent soil? Please elaborate or rephrase.
L670–680 What do the correlation values represent? The value of r could be very high even if the bias or the slope of the linear fit is profoundly off. The low correlation could stem from source soil mineralogy, but it could also reflect the model’s inability to capture specific transport and deposition patterns. How well does the model simulate dust itself? While these transport factors might average out in a climatological comparison, they gain relevance here since the authors are attempting to simulate hourly variations in mineral composition during an airborne campaign.
L735: Consider filtering the data to remove contamination from non-dust sources.
L763–764: “This suggests that while the mass balance is improved, the inherently weak correlation, likely driven by the influence of non-simulated minerals, is further exacerbated by the proposed mass redistribution.” This sentence is difficult to understand. Please rewrite.
L765-773: This discussion is sensitive to the chosen assumptions regarding the iron content within the simulated minerals.
L770: Note that not all iron present in minerals exists in the form of iron oxides. This distinction should be accurately reflected in the text.
L877–L896: I think this discussion would fit best in the Conclusions section. |
Dust aerosols are key players in the Earth system, exerting a variety of impacts that are many times shaped by their mineralogical composition. This manuscript presents advances in the characterization of the dust mineralogy from North African sources and approaches one fundamental problem that dust models face when trying to incorporate minerals in their formulation: the distribution of minerals across aerosol particle sizes. The core of the study focuses on the methodological improvements included in the COSMO-MUSCAT model to improve the emitted minerals’ Particle Size Distribution (PSD) and the evaluation of a set of simulations against observations. The authors also report new observational data of mineral and elemental composition from two different and recent campaigns, which are of value for increasing our understanding of the dust mineralogy and its regional variations. The improvements included in the model as well as the observational data merit publication and are useful for the scientific community to advance in this field.
However, in my view, some aspects of the observational campaigns and the model evaluation methodology have to be clarified. Also, the manuscript's current presentation is quite dense and the structure makes it difficult for the reader to clearly identify the primary findings. I would recommend the authors to split the work in two different articles: one devoted to the observational campaigns and the other related to the modelling work and its evaluation. Alternatively, a synthesis of these two parts is recommended, moving to supplementary materials the non-essential information.
Please, see my specific comments below.
Specific comments:
The authors report information on two different experimental campaigns: DUSTRISK2022 and JATAC2022. The protocols, instruments and characteristics of the measurement methods are explained in some detail, however the actual measurements, their representativity and uncertainty ranges are not reported or clearly discussed in the main article. These are relevant details to assess their strengths and weaknesses as a target for model evaluation. Some of the details reported in sections 4.2 and 4.3 (e.g., Table 3) could be moved to supplementary materials, while more information on the results of the measurement campaigns and their uncertainties could be added in the main paper. For instance, the JATAC2022 campaign reports close to 0 % of illite and smectite in fine particle sizes in Cape Verde. Is this expected? How is this reconciled with dusts close to sources with illite contents up to 40%? On the other hand, the estimate of iron oxides content from JATAC2022 relies on information from other experimental studies (DiBiagio et al., 2019), how does this impact their reliability?
The description of the model evaluation procedure overrelies on a workflow schematic (Figure 3). I would recommend to expand the description and justify the different steps taken in the main text.
As mentioned in the general comments, some aspects of the paper organization could be improved for clarity. For instance, in the current structure:
My general recommendation would be to:
Technical comments: