the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Mar 2026
iDust-ut: A Global Wind Erosion Threshold Dataset for Enhanced Dust Forecasting
Abstract. Accurate global dust forecasting is essential for public health, transportation safety, and industry operations. Current models underestimate extreme dust events with pronounced regional biases, primarily due to poor parameterization of the wind erosion threshold (ut), a fundamental parameter representing the minimum wind speed for dust emission initiation. This study develops iDust-ut, an advanced global threshold dataset through a multi-source data fusion approach that integrates ground observations, satellite remote sensing, and multiple reanalysis datasets. Validation against independent field observations demonstrates high accuracy, with a correlation coefficient of 0.93 and a mean absolute error of 0.8 m s-1, far outperforming existing products. Additionally, this study introduces a model-adaptive threshold adjustment scheme that compensates for systematic wind speed biases across different numerical models. Based on annual 2023 evaluations across Northwestern China (for PM10) and global dust belt regions (for dust optical depth), implementation of the iDust-ut dataset with adaptive adjustment substantially enhances forecast performance of the iDust model compared to the approach of using a global constant threshold. Specifically, the Threat Score for extreme PM10 forecasting in Northwestern China increased by 108 % (from 17.39 % to 36.25 %), while the Threat Score for extreme dust optical depth simulation in global dust belt regions improved by 47 %. The enhanced model performance substantially outperforms the European Centre for Medium-Range Weather Forecasts (ECMWF) operational aerosol forecasts in many key metrics. The iDust-ut dataset offers an immediately deployable, computationally efficient solution for enhancing dust forecasting accuracy across various modeling systems. The iDust-ut dataset can be freely accessed via https://zenodo.org/doi/10.5281/zenodo.15580883 (Chong and Chen, 2026).
- Preprint
(1983 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2025-628', Anonymous Referee #1, 01 Jul 2026
-
RC2: 'Comment on essd-2025-628', Anonymous Referee #2, 19 Aug 2026
The article presents a new dataset characterising a key factor in dust emission parameterisations, the wind erosion threshold. This factor has a large impact on the results produced by these parameterisations and is potentially understudied, making this article a valuable contribution. My comments are separated into substantive points, which I believe should be addressed before publication, and technical or presentational ones, which are offered as suggestions. Comments 1 to 3 seem to me the most important. Several of my points may reflect gaps in my own reading rather than shortcomings of the manuscript, and I would be glad to be corrected on any of them.
Substantive comments
1) Independence of the validation dataset (Sect. 4, Table A2, L334, L373-375). Validation is performed against five field campaigns. The TaZhong campaign listed in Table A2 (83.66° E, 38.98° N) probably took place at or very near the Tazhong ISD site used as the worked example in Sect. 3.1 (83.667° E, 39.0° N), which would make the campaign measurements and the ISD records correlated. The same may apply to other campaigns, as the Fennec observations at Bordj Badji Mokhtar are also likely to be near an ISD site. Since a Gaspari-Cohn local correction is applied to bring the gridded field closer to nearby stations, validation at these sites may be contaminated by the field having been merged with nearby ISD data. Please state whether any validation locations fall within R_loc (L334) and, if so, consider excluding those stations from the localisation scheme. In addition, a leave-one-out scheme (repeating the computation N times, each time excluding one campaign from the dataset) would help demonstrate the robustness of the reported accuracy, which currently rests on five points.
2) Inconsistency between text, figures and abstract (L593-602, Fig. 8, L24-25, L631). The values quoted in L593-602 do not match those shown in Figure 8. For example, iDust_CTRL is annotated with a Threat Score of 6.50 for extreme DOD (>1.0) in Fig. 8j, but the text gives 7.28. Several other values in the same paragraph differ similarly, and the corresponding paragraph for Fig. 8i appears to be affected as well. Please check the values throughout and ensure that the headline improvements stated in the abstract (L24-25) and in L631 are consistent with the figures and the body of the text.
3) Equation (1). As written, Eq. (1) always gives DOD > AOD for any α > 0, which should not be possible since DOD is a subset of AOD. One of the terms probably has a negative sign. I would ask the authors to confirm that this is only a typesetting error and that it does not propagate into the code. I could not verify this myself, as the relevant script is not present in the indicated repository.
4) Code availability and reproducibility. I would like to acknowledge that the authors have already deposited the dataset and a sizeable part of the code openly, which made this review considerably easier. The dataset itself is provided in clear, well-annotated NetCDF files. The remainder of this comment is intended to strengthen an already good practice rather than to criticise it. The code should be accompanied by a means of recreating the software environment the authors used, for example a requirements.txt, a pyproject.toml with a lock file, or an equivalent. The code and the data currently share a single Zenodo record released under CC BY 4.0, which is a data licence; the code should additionally be made available under an explicit software licence (MIT, Apache-2.0, GPL, or similar). I also could not find the scripts that derive DOD from the MODIS and VIIRS products, which are needed to verify comment 3.
5) Derivation of the tuning coefficient C (Table 3, L500-501). Please describe how the dust emission coefficient C is derived for each experiment. It should be made clear that the improved performance is not attributable to the retuning itself, for example by repeating the CTRL experiment with C = 0.9 and C = 1.0.
6) Confidence of the spatial descriptions (L351-356, L381-385). The text at L351-356 presents a very positive general picture ("closely track each other", differences "typically below 1 m s⁻¹"), but the figures appear to show substantial divergence over North and South America and perhaps parts of South Asia. This is acknowledged at L368-369, so it would be helpful to state explicitly that the close agreement holds over the dust belt rather than globally. Similarly, L381-385 notes that regions of less frequent dust activity such as South America are assigned higher thresholds, but this appears to hold for the MERRA-2 result more clearly than for the ERA5 one.
7) Aerosol composition at the PM10 validation stations (L124-126). The assumption that dust dominates PM10 at the Northwestern China stations could be better supported. PM2.5/PM10 ratios from the same CNEMC network, or a nearby AERONET site where one exists, would be relatively quick proxies, though the authors may be aware of a supporting study that I was unable to find.
8) PM10 conversion coefficients (Appendix C). Please briefly explain how the linear coefficients used to convert dust mixing ratios to PM10 were derived, specifically the factors 0.4 applied to DD3 and 0.74 applied to DD4.
9) Framing of u_t as the fundamental source of error (L47-49, L14-16). While I agree that the emission parameterisation is a significant source of uncertainty in NWP dust forecasts, it is not the only one; transport, removal mechanisms, and assumed particle shapes and sizes also contribute. A softer framing would serve the introduction better, for example "One of the fundamental challenges [...]", unless the authors can support the stronger claim with specific references. The two works cited immediately before this sentence (Pitkänen et al., 2023; Chen et al., 2025) document the forecast errors but do not attribute them to the wind erosion threshold. The same phrasing appears in the abstract.
Technical and presentational comments
10) Colour maps and panel labelling in Figures 3 and 4. The u_t maps in Figure 3 show a continuous quantity with no natural midpoint but use a diverging colour map. A sequential, perceptually uniform map (such as the matplotlib default, viridis) would present the data more clearly and would also improve accessibility for readers with colour vision deficiencies. The same applies to Figure 4b. In contrast, Figure 8 correctly uses diverging colours, since bias has a natural midpoint at zero, and should be left as it is. There is also an issue with the panel labelling in Figure 3, which runs a, b, c, d, c, d, d, e and does not match the caption.
11) Figure 5 scatter plots. The scatter plots in Figure 5 are difficult to read because of the density of points. Panels (a) and (b) look almost identical, yet the first is annotated with Corr = 0.81 and the second with Corr = 0.44. A density or heatmap-style presentation would show the result more clearly. Units should also be stated for the annotated statistics, as the bias and MAE values do not appear consistent with axes labelled in 10⁴ m³ s⁻³.
12) Wordiness and typographical errors. The writing is sometimes wordier than it needs to be. For example, L111 could go from "This study implements a comprehensive quality-control framework through multiple criteria to ensure the reliability of threshold determination [...]" to "This study implements a quality-control framework to ensure the reliability of threshold determination [...]" without any change in meaning, making the article easier to read. Further examples are L621-626 and the first paragraph of the conclusions (L680-686). A short list of typographical errors: L256 "candinate"; L421 is missing "across" after "proxies"; L527 "exhibit" should be "exhibits"; L679 "adapation"; L697 probably needs "innovative" rather than "innovation".
I thank the authors for an interesting and useful contribution, and I would be happy to review a revised version.
Citation: https://doi.org/10.5194/essd-2025-628-RC2
Data sets
iDust-ut: A Global Wind Erosion Threshold Dataset for Enhanced Dust Forecasting Mei Chong and Xi Chen https://zenodo.org/doi/10.5281/zenodo.15580883
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 322 | 129 | 16 | 467 | 23 | 40 |
- HTML: 322
- PDF: 129
- XML: 16
- Total: 467
- BibTeX: 23
- EndNote: 40
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Please find my review attached. This is an excellent paper.