How accurate are operational dust models in predicting Particulate Matter (PM) levels in the Eastern Mediterranean Region? Insights from PM Surface Concentrations

Eleftheriou, Andreas; Mouzourides, Petros; Kouis, Panayiotis; Kalivitis, Nikos; Katra, Itzhak; Vasiliadou, Emily; Savvides, Chrysanthos; Yiallouros, Panayiotis; Neophytou, Marina K.-A.

doi:https://doi.org/10.5194/egusphere-2025-2739

Preprints

https://doi.org/10.5194/egusphere-2025-2739

Preprints

27 Jun 2025

| 27 Jun 2025

How accurate are operational dust models in predicting Particulate Matter (PM) levels in the Eastern Mediterranean Region? Insights from PM Surface Concentrations

Andreas Eleftheriou, Petros Mouzourides, Panayiotis Kouis, Nikos Kalivitis, Itzhak Katra, Emily Vasiliadou, Chrysanthos Savvides, Panayiotis Yiallouros, and Marina K.-A. Neophytou

Abstract. This study provides the first comprehensive assessment of eleven operational dust forecast models and a multi-model ensemble in predicting ground-level Particulate Matter (PM) concentrations in the Eastern Mediterranean Region (EMR), with a focus on Cyprus, Greece, and Israel. Ground-based observations from regional background stations support model performance assessment across different PM fractions (PM₁₀, PM_2.5, and coarse particles), using established statistical metrics (correlation coefficient, R, Mean Bias, MB, and Root Mean Square Error, RMSE). The results reveal substantial variability in accuracy, with R values ranging from −0.24 to 0.91 depending on site and event subset. NASA-GEOS consistently achieves the highest correlation (R = 0.71 at Cyprus), indicating accurate representation of dust transport. In contrast, SILAM and EMA-REG4 perform poorly, with low correlations (R = 0.10 and −0.24, respectively) and significant estimation errors (MB = −90.34 µg/m³ for EMA-REG4). The NOA-WRF model effectively captures extreme dust events, with R = 0.91 during the 95th percentile of PM concentrations in Greece. Most models perform better for coarse PM, with the BOOT methodology indicating reduced scatter and bias during dust storm days. However, no model performs optimally across all sites and conditions, highlighting the need for location-specific tuning and evaluation. The study underscores the importance of refining model configurations and improving parameterizations to enhance forecast accuracy. Future efforts should incorporate localized data and further develop region-specific models to improve the operational use of these systems in early warning protocols for mitigating public health impacts.

Received: 09 Jun 2025 – Discussion started: 27 Jun 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Andreas Eleftheriou, Petros Mouzourides, Panayiotis Kouis, Nikos Kalivitis, Itzhak Katra, Emily Vasiliadou, Chrysanthos Savvides, Panayiotis Yiallouros, and Marina K.-A. Neophytou

Status: final response (author comments only)

RC1:
'Comment on egusphere-2025-2739', Anonymous Referee #1, 26 Aug 2025
Review of egusphere-2025-2739 entitled “How accurate are operational dust models in predicting Particulate Matter (PM) levels in the Eastern Mediterranean Region? Insights from PM Surface Concentrations” by Andreas Eleftheriou et al.

General

The manuscript entitled “How accurate are operational dust models in predicting Particulate Matter (PM) levels in the Eastern Mediterranean Region? Insights from PM Surface Concentrations” by Andreas Eleftheriou et al. provides an assessment of the performance of eleven operational dust forecasting models and a multi-model ensemble through comparisons against surface PM measurements at three sites in the Eastern Mediterranean Region (EMR), Ayia Marina (AM) in Cyprus, Be’er Sheva (BS) in Israel and Finokalia (FKL) in Crete. The evaluation is done using specific established statistical metrics, namely correlation coefficient, R, Mean Bias, MB, and Root Mean Square Error, RMSE. The obtained results reveal a substantial variability in the models’ accuracy that no single model consistently achieves accurate predictions across all three regions and in all conditions (entire study period and days with high dust loadings).
The manuscript adds to the scientific community’s knowledge about the performance of operational dust models. Nowadays, a significant effort is made on developing and implementing such models to monitor dust levels in the atmosphere, and also warning the public about hazardous dust episodes, at regional or global scales. Given that these models differ between them in their spatial resolution, meteorological drivers, emission schemes or data assimilation procedures, it is important to intercompare them and to draw conclusions on which model(s) outperform. In this meaning, the study is interesting, although the conclusion drawn is not clear as to which model does so, in overall. While this is a bit disappointing, the study proves and convinces the reader that this happens given the multi-parametric problem, in the sense that many and combined factors play role and determine the overall model performance. While some questions remain unanswered (as explained below), it is more or less understood that a model can perform well at some site, while not in another, or better/worse under different dust loading conditions.
Based on the above, and the fact that the analysis is correct and complete at a significant level, while the text is well organized and written, I recommend publication of the manuscript subject to some corrections and recommendations for revision suggested below.

Main Comments
Further reference to the existing literature dealing with Mediterranean dust storms/episodes should be made. In particular, additional references should be made to papers dealing with increasing/decreasing desert dust storms in the Mediterranean Basin.

The style of discussion of the results obtained is a bit tiring for the reader. For example, discussing the results station by station in section 4.1 may get the reader tired and lost in the details given. What matters is the comparative analysis, which is discussed in the last paragraph (section 4.1.1). This kind of discussion can be expanded referring to Figures 3 and 4. Also, Table 3 can be restructured to show rankings of the 11 models performance regarding the 3 different stations and the statistical metrics.

Improve and make more descriptive the captions of some Figures (e.g. Figures 9, 10, 11) to help the reader to more easily and rapidly understand what is shown.

The style of discussing results in section 5 should be improved by: (i) reporting the values of statistical metrics wherever recalled/reported in the text, (ii) specifying if the statements made with reference to the models’ performance concern the entire study period or other conditions, e.g. the 95^th percentile pf observed PM concentrations.

Provide arguments to infer responses to questions as to why a model, e.g. the NASA-GEOS, outperforms at AM and BS but not in FKL, or why a model, e.g. NOA-WRF outperforms during intense dust events, but does not the same in overall.

Minor comments
Line 64: Yet, the satellite detection algorithms have progressed with time, including geostationary satellites as well (e.g. Kolios and Hatzianastassiou, 2019) exempt from the limitations of the polar orbiting satellite-based algorithms.

Line 93: the reference “Varga et al., 2014)" misses in the list of references.

Section 2: Dust storms in the MB and their spatiotemporal characteristics are better captured by satellites (e.g. Gkikas et al., 2013, 2016).

Line 110: The range of measured/exported particles’ size should be given (it is necessary information to be used in the comparison with the corresponding sizes of the 11 models) since differences with models can partly explain the PM overestimations/underestimations by the models.

Line 131: replace “develop …” by “developed …”.

Lines 133: It would be useful to indicate these dust storm days on Fig. 2, probably using different symbols for each station. Thus, readers can have an idea about the intensity of these dust storms at every site. Also, are there common days in the 3 stations out of the reported 106, 88 and 101 dust storm days?

Table 1: What kind of radiation interactions are those reported in this Table? Aerosol-radiation interactions or others as well, e.g. aerosol-cloud? Please specify.

Table 2: at what height is the first level above surface for the NASA_GEOS and NOA_WRF models, why are the corresponding values missing in the Table?

Line 167: use a parenthesis after the sum symbol, i.e. put the difference “M_i-O_i” in a parenthesis.

Line 179: relative bias and relative RMSE would be equally interesting metrics to show (as bias and RMSE).

Figure 3: Change “MEDIAN” to “MMM” for consistency with the rest of paper. Do the same in Figures 5 and 7.

Line 310: Figures 3 and 4 are not discussed and mentioned in the text, reference to them should be made.

Line 330: does “… R=0.62 …” should read “… R=0.55 …”?

Discussion in section 4.1.2: It is also interesting to discuss if models perform better or worse for high-dust events compared to all cases, also providing possible explanations for the improvement/deterioration. Some models, e.g. NOA_WRF, perform better for high-dust events in some locations (in FKL in this case) while doing worse in other locations (BS and AM), why does this happen?

End of section 4.1: A general assessment should made referring to whether the findings for the identified dust days by Achilleos et al. (2020) methodology are similar to those drawn from the high-dust days analysis of the previous section or not. It is essential to see if the methodology applied to identify dust events affects the results referring to the comparative model’s performance analysis (keeping in mind that basically they should not do so).

Discussion of Figures 9, 10 and 11: Make a comment on existing differences between Figures 10 and 11 since the nature of the results shown on these figures is about similar (as they both refer to days with high dust loadings).

Lines 502-505: This is a (probably the most) typical statement reflecting the complexity of the problem addressed, concerning the distinction of which is/are the model/models that perform better than others, overall. In spite of the differences existing between the models at various levels, they are all performing more or less similarly, perplexing the situation/problem.

Lines 508-510: Remove the empty line and link the text (from “and” to “Table 2”).

Section 5: The results show that different factors, e.g. vertical/horizontal resolution, height of first model level etc. are used as criteria for evaluating the performance of the models. Yet, it seems that some models are superior to others with respect to a specific factor, while other model(s) are superior with respect to other factor(s). Thus, it seems that a kind of counterbalance exists, leading to a roughly similar performance of the models. Would it be possible to draw, based on the overall performance of the models, a conclusion about which factor/factors is/are the most important for the model’s performance?

Lines 591-595: What is reported here is somewhat worrying. Why/how the performance of a model should change depending on the selected approach/method for the evaluation?

References
Gkikas, A., Hatzianastassiou, N., Mihalopoulos, N., Katsoulis, V., Kazadzis, S., Pey, J., Querol, X., and Torres, O.: The regime of intense desert dust episodes in the Mediterranean based on contemporary satellite observations and ground measurements, Atmos. Chem. Phys., 13, 12135–12154, https://doi.org/10.5194/acp-13-12135-2013, 2013
Gkikas, A., Basart, S., Hatzianastassiou, N., Marinou, E., Amiridis, V., Kazadzis, S., Pey, J., Querol, X., Jorba, O., Gassó, S., and Baldasano, J. M.: Mediterranean intense desert dust outbreaks and their vertical structure based on remote sensing data, Atmos. Chem. Phys., 16, 8609–8642, https://doi.org/10.5194/acp-16-8609-2016, 2016
Kolios, S.; Hatzianastassiou, N. Quantitative Aerosol Optical Depth Detection during Dust Outbreaks from Meteosat Imagery Using an Artificial Neural Network Model. Remote Sens. 2019, 11, 1022
Citation: https://doi.org/10.5194/egusphere-2025-2739-RC1
- AC2:
  'Reply on RC1', Petros Mouzourides, 11 Sep 2025
  We sincerely thank Reviewer 1 for the constructive and supportive review of our manuscript. We appreciate the recognition of the study’s contribution and the thoughtful comments that will help us improve the clarity, structure, and interpretative depth of the manuscript. We appreciate the recognition of the study’s contribution and the thoughtful comments that will help us improve the clarity, structure, and interpretative depth of the paper. We fully agree with the conclusion that no single model can be considered as consistently well-performing across the Eastern Mediterranean. This is also a key message of our study, as the evaluation shows that model skill depends strongly on site-specific characteristics. In general, NASA-GEOS tends to achieve the best scores, but its performance still varies between locations. Proximity to dust sources and local topographic features clearly influence model skill, which explains why a model may perform well at one site but less so at another, or under different dust loading conditions. We will emphasize this point more clearly in the revised conclusions to avoid any ambiguity. Below we provide our responses to the main points raised.
  Main comments
  On literature references: We agree and will include additional references dealing with Mediterranean dust storms and long-term trends, such as Gkikas et al. (2013, 2016) and Kolios & Hatzianastassiou (2019). This will strengthen the background section and better situate our study within the regional literature.
  
  On the style of discussion (Section 4.1): We acknowledge that the location-by-location description is lengthy and can be repetitive. In the revision, we will streamline the text, focus more on comparative analysis across sites and models, and expand the synthesis paragraph with explicit reference to Figures 3 and 4. We will also explore restructuring Table 3 to highlight rankings of model performance.
  
  On figure captions: We will revise the figure captions to more clearly explain what each figure shows, enabling the reader to better understand them without referring back to the main text.
  
  On Section 5: We will improve the discussion by systematically including values of statistical metrics when mentioned in the text, and by clarifying whether statements refer to the entire period or to subsets (e.g., high-dust events). We will also expand on why specific models (e.g., NASA-GEOS, NOA-WRF) perform differently across sites and under different dust loading conditions. As suggested, we will further discuss how certain configuration settings (e.g., resolution, first model layer height) may contribute to these differences and whether specific factors appear more influential overall.
  
  Minor comments
  We will address all the minor points raised, including:
  Adding more relevant references where needed (Varga et al., 2014; Gkikas et al., 2013, 2016; Kolios & Hatzianastassiou, 2019).
  
  Clarifying particle size ranges in Section 2.
  
  Indicating dust storm days in Figure 2 and commenting on overlaps between stations.
  
  Clarifying the type of radiation interactions reported in Table 1.
  
  Correcting notation, typographical errors, and formatting issues (e.g., empty lines, consistency in figure labels etc).
  
  Discussing model performance for high-dust events versus all days, and clarifying differences between evaluation approaches (e.g., Achilleos et al. methodology vs. 95th percentile).
  
  Once again, we thank the reviewer for the comprehensive and constructive feedback. We are confident that the suggested revisions will substantially improve the readability, contextualization, and scientific contribution of the manuscript.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2739-AC2
RC2:
'Comment on egusphere-2025-2739', Anonymous Referee #2, 05 Sep 2025

This manuscript presents a multi-model evaluation of 11 operational dust forecasting systems and a multi-model median ensemble against ground-based PM observations in the Eastern Mediterranean Region. The topic is timely and relevant, as dust events have important health and air quality implications and systematic model evaluations in this region are scarce. The study is generally well-structured and the dataset is valuable, but the manuscript would benefit from stronger interpretative discussion of the results.

Some comments for further improvement are provided below.

Major

The introduction would benefit from additional references to previous dust model evaluation studies (e.g., SDS-WAS activities, Saharan/Middle Eastern validation work). This would strengthen the justification for the claim that very few studies have used near-surface monitoring data and none focused on the EMR.

Section 4.1 is overly detailed and repetitive, presenting results station by station. This can be quite tiring for the reader and obscuring broader insights. The discussion should be streamlined to emphasize comparative analysis across models and regions and accompanied by a thorough discussion on the reasons behind this behavior (see next comment).

The authors already provide a valuable discussion of how model configurations (horizontal/vertical resolution, first model layer height etc.) may influence forecast skill. However, this discussion remains somewhat generic and is not always tightly connected to the evaluation results. They should consider emphasizing more on this component.

Minor

Line 93: the reference “Varga et al., 2014" should be added in references.

Line 120: there is an empty line and the “Table 2” is in bold.

Table 2 misses the "Height first layer" for some models. If the information is not available online, the authors should consider contacting the teams in charge of the systems.

Line 510: there is an empty line and the “Table 2” is in bold.

Line 587: Please clarify terminology. The study uses a multi-model median (MMM), not a multi-model mean. Referring to it as an “average” is potentially misleading.

Citation: https://doi.org/10.5194/egusphere-2025-2739-RC2
- AC1:
  'Reply on RC2', Petros Mouzourides, 11 Sep 2025
  We sincerely thank Reviewer 2 for the constructive and insightful comments that will help us improve the clarity and impact of the manuscript. Below we provide our responses to the main points raised.
  
  Major comments:
  On the Introduction: We agree with the reviewer and will strengthen the literature review by adding references to previous dust model evaluation studies, including SDS-WAS activities and validation work in Saharan and Middle Eastern regions. This will better contextualize our study and support the statement regarding the limited number of evaluations using near-surface monitoring data in the EMR. For completeness, we note that our manuscript already cites Basart et al. (2012), which evaluated BSC-DREAM8b over Northern Africa, the Mediterranean, and the Middle East. However, this is a model evaluation rather than an assessment of SDS-WAS activities as a program. We also cite García-Castrillo & Terradellas (2017, WMO SDS-WAS report), which focused on the Canary Islands, but not the Middle East. We will therefore add more recent references, where available, to strengthen the context.
  
  On Section 4.1: We acknowledge that the current location-by-location (station-by-station) presentation is lengthy. In the revised version, we will streamline this section to reduce descriptive repetition and focus more on comparative insights across models and regions.
  
  On model configuration discussion: We appreciate this important point. We will enhance Section 5 by connecting specific findings from our evaluation to the role of configuration settings (e.g., resolution, first model layer height). Where possible, we will illustrate how these settings may explain differences in performance among the models, thereby making the discussion more tightly linked to the results. We note, however, that as we are evaluating existing operational systems, we do not always have full access to the details of model design and implementation. To address this, we have carefully reviewed the available references describing the development of each model and have contacted the respective model teams for additional information where possible. In our discussion, we will therefore remain within the level of information documented in the published literature and provided by the model developers.
  
  Minor comments
  We will add the missing reference to Varga et al. (2014) in the Introduction.
  
  The formatting issues with empty lines and bold text at Lines 120 and 510 will be corrected.
  
  We will update Table 2 to include the “Height of first model layer” for all models. We had contacted the teams responsible for developing and maintaining the models, and we will carefully recheck our records to ensure that no information has been overlooked.
  
  We will clarify the terminology regarding the multi-model approach, specifying that we use the multi-model median (MMM) rather than the mean.
  
  Once again, we thank the reviewer for the valuable feedback. These revisions will strengthen both the contextualization and interpretative depth of the manuscript.
  
  Citation: https://doi.org/10.5194/egusphere-2025-2739-AC1

Andreas Eleftheriou, Petros Mouzourides, Panayiotis Kouis, Nikos Kalivitis, Itzhak Katra, Emily Vasiliadou, Chrysanthos Savvides, Panayiotis Yiallouros, and Marina K.-A. Neophytou

Viewed

Total article views: 467 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
401	54	12	467	13	21

HTML: 401
PDF: 54
XML: 12
Total: 467
BibTeX: 13
EndNote: 21

Views and downloads (calculated since 27 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	46	4	1	51
Jul 2025	54	26	7	87
Aug 2025	121	21	4	146
Sep 2025	180	3	0	183

Cumulative views and downloads (calculated since 27 Jun 2025)

Month	HTML	PDF	XML	Total
Jun 2025	46	4	1	51
Jul 2025	54	26	7	87
Aug 2025	121	21	4	146
Sep 2025	180	3	0	183

Viewed (geographical distribution)

Total article views: 464 (including HTML, PDF, and XML) Thereof 464 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 17 Sep 2025

Short summary

Desert dust storms are a significant environmental concern in the Eastern Mediterranean. This study compared eleven forecasting models to see how well they predict dust levels in the atmosphere. By checking their results against in-situ and satellite measurements, we found that some models work better than others, but none are perfect. These findings can help improve forecasting systems, making them more reliable and useful for protecting public health and preparing for extreme dust events.


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