the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Feb 2026
Dynamic drag partitioning in GEOS-Chem (v. 14.2.3) eliminatessource function and tuning, revealing equifinality of atmosphericdust observations
Abstract. Large surface roughness elements (typically vegetation) extract momentum from near-surface winds partitioning it between the vegetation canopy and the unsheltered soil surface, altering the spatial distribution and intensity of dust emission. Data for global drag partition varying over space and time are limited and classical dust models often adopt simplified assumptions or neglect the effect of roughness entirely, relying instead on empirical dust source function to reproduce observed atmospheric dust. Such approaches ignore the three-dimensional spatiotemporal sheltering provided by vegetation and the soil roughness which limits the physical fidelity and predictive capability of dust models. Here we implement the global sheltering parameterization based on surface albedo data in the dust emission scheme of a global atmospheric chemical transport model and evaluate its impact on simulated global dust emissions. We compare simulations from this sheltering parameterization with default settings from classical dust modelling and validate our results against atmospheric dust observations following the widely accepted approach. This is the first study to evaluate the performance of the albedo-based sheltering parameterization globally. The approach eliminates the need for an empirical source function to constrain sources and avoids the need to tune dust emission models. It produces atmospheric dust loadings consistent with atmospheric dust observations despite substantially altering the frequency and spatial pattern of dust emissions. Evidently, different patterns of dust emission frequency can produce similar atmospheric dust loadings, revealing that previous studies intent on reproducing general patterns of large-scale atmospheric dust burden have overlooked the equifinality of dust emission frequency. Accurate dust emission magnitude and frequency remain hampered by entrainment threshold fixed over space, static over time with endless sediment supply, consistent with all other models.
- Preprint
(2589 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-5872', Anonymous Referee #1, 29 Apr 2026
- CC1: 'Reply on RC1', Hongquan Song, 02 Jun 2026
- AC1: 'Reply on RC1', Adrian Chappell, 02 Jun 2026
-
RC2: 'Comment on egusphere-2025-5872', Anonymous Referee #2, 26 May 2026
This paper uses an albedo-based drag partition to simulate the global dust cycle using the DEAD dust emission scheme in the GEOS-Chem model. Three experiments are carried out, two of which use default settings (Exp 1 with source function, Exp 2 without source function), and the third one (Exp 3) excludes the source function but uses a detailed (albedo-based) drag partitioning. Year 2016 is simulated at horizontal resolution of 0.5° × 0.625° for emission and 2° × 2.5° for transport. Each simulation is scaled to produce the same total annual dust emission. The paper concludes that the albedo-based configuration (Exp 3) eliminates the need for dust source function because it produced similar results for atmospheric dust as in Exp 1 (equifinality). While the attempt to show equifinality is interesting, I find the overall methodology problematic and interpretations misleading for the reasons discussed below. Therefore, I recommend rejection with encouragement for resubmission after addressing the comments below.
MAJOR COMMENTS
1. Incorrect Interpretation of Dust Source Function
A dust source function (Ginoux et al., 2001) is primarily used to represent sediment availability, not drag partitioning. In such models (e.g., Zender et al., 2003), the drag is typically simplified (as in Exp 2) and an additional control is applied to scale dust emissions from vegetated areas. Therefore, the comparison between Exp 1 and Exp 3 is problematic and misleading, as it incorrectly treats the source function as a representation of drag partitioning.
2. Models Should Be Viewed Holistically, Not by Breaking Parts Apart
Exp 2 removes the source function and is compared with Exp 3, leading to the conclusion that Exp 2 performs worse than Exp 3. This is expected, as Exp 2 uses a time-static drag partitioning without spatial variability (L125, Eq. 4), while Exp 3 applies a more comprehensive drag partitioning. This comparison would be meaningful if Exp 2 reflected a standard modeling configuration. However, dust models are generally set up to use a combination of a source function and a simple drag formulation, rather than applying the simple drag in isolation. As a result, it is unclear what additional insight Exp 2 provides.
3. Inadequate Evaluation/Comparison
As also noted in the comments from Referee #1, an analysis of individual dust events is needed to better test the fidelity of the albedo-based approach and its comparison with Exp 1. Moreover, (L220–229) the use of observational datasets from different years than the model simulation would not be very appropriate due to interannual variability in meteorology and land cover that affect the dust cycle.
MINOR COMMENTS
- I believe the concerns about the albedo-based method raised in Okin, 2023 are not addressed. I think dividing by the isotropic reflectance parameter does not fully remove the spectral dependence due to mixing (inseparability) of spectral (compositional) and geometric effects. In this regard as well, simulating many individual dust cases would help test the utility of the method.
- Furthermore, the paper does not state how often the albedo parameters update. Using monthly or annual averages may limit the advantage of using the satellite data by smoothing important temporal variability in surface roughness.
- Sect 3.1 L280-285: The paper appears to show deficiencies in previous model configurations by isolating model components. Removing the controlling factor (S) will lead to dust emissions from forests, as expected.
- It is not clear how comparisons were made but when comparing with observations (e.g., AERONET), it is better to first collocate instantaneous values, rather than computing daily averages separately.
- L144-147: Several previous studies also have explicitly stated that, to be strictly correct, the u* at the soil surface should be used (e.g., Kok et al., 2014). However, I do not see a particular need to highlight this point here because the drag factor in the referenced ‘classical’ model (Exp 2) remains spatially homogeneous.
- Equifinality appears to hold only in seasonal or annual means, as atmospheric dust may differ more between Exp 1 and Exp 3 at shorter timescales. This characterization is important in stating equifinality.
- L414-417: As a source of model uncertainty, the paper does not mention the issue of treating air density as a fixed quantity in many dust modeling studies, including in the cited Chappell et al., 2023 papers. Because air density varies spatially and temporally in a nonlinear way, using a globally uniform single value removes an important thermodynamic control on dust emission. This can introduce an error of ~20% in dust fluxes. For this and other issues in dust modeling (e.g., sandblasting efficiency), Joshi, 2024 (e.g., its Sect 3.6) should be used.
REFERENCES
- Kok, J.F., et al., 2014. An improved dust emission model – Part 1: Model description and comparison against measurements, Atmos. Chem. Phys., 14, 13023–13041, https://doi.org/10.5194/acp-14-13023-2014, 2014.
- Okin, G.S., 2023. Shadow is related to roughness but MODIS BRDF should not be used to estimate lateral cover. Remote Sens. Environ. 292, 113581. http://dx.doi.org/10.1016/j.rse.2023.113581.
- Joshi, J.R., 2024. Dust model sensitivity to dust source mask, sandblasting efficiency, air density, and land use: Implications for model improvement. Atmospheric Pollution Research, 15, 102230. https://doi.org/10.1016/j.apr.2024.102230. (accepted manuscript available via Google Scholar)
Citation: https://doi.org/10.5194/egusphere-2025-5872-RC2 - AC2: 'Reply on RC2', Adrian Chappell, 02 Jun 2026
-
RC3: 'Comment on egusphere-2025-5872', Anonymous Referee #3, 10 Jun 2026
This study implements an albedo-based drag partitioning scheme in the global GEOS-Chem model and simulates dust emissions at 0.5° × 0.625° resolution. Three model runs are performed using the source function without a drag partitioning algorithm (Exp1), the default simplified drag partitioning algorithm without the source function (Exp2), and the proposed albedo-based drag partitioning algorithm without the source function (Exp3). Comparisons among the three simulations and evaluations against observations reveal similar results for Exp3 and Exp1, leading to the conclusion that the dust model exhibits equifinality.
Investigation of the albedo-based drag partitioning scheme at a global scale is valuable. However, I have a fundamental concern about the study design. The source function is replaced with drag partitioning, although the two represent different physical processes. Additionally, the conclusion regarding equifinality is weakened by the coarse temporal resolution of the observational evaluation. I therefore recommend rejection at the current stage and encourage resubmission after the study design is revised, potentially with additional model simulations.
Major comments
1. The study is designed with the intention of replacing the dust source function with the drag partition parameterization. Fundamentally, however, a dust source map represents the geographic availability or relative erodibility of soil particles, whereas drag partitioning represents how surface roughness divides wind stress between roughness elements and the exposed soil. These are distinct processes. For example, a flat concrete surface may have low aerodynamic roughness but cannot emit mineral dust. The two treatments are therefore not mutually exclusive, and both processes should be represented in dust models. I agree with the authors that static source functions have limitations, but they should be improved rather than eliminated.
On this basis, Exp1 and Exp3 are not physically equivalent configurations. Their similar performance provides useful information about model sensitivity to separate subprocesses, but it does not justify replacing the source function with drag partitioning. I believe simulations using both simplified and albedo-based roughness should also include the source function. Alternatively, the source function could be disabled to isolate the uncertainty associated with it. However, in such cases, the simulations should not be directly compared with observations without acknowledging the missing sediment-availability constraint.2. The equifinality conclusion is based on monthly-to-annual averaged simulations (Figs. 5-8). However, Fig. 4 indicates that the occurrence of dust emission events differs between Exp1 and Exp3. Observations of DOD, concentration, and deposition flux are less sensitive to transient, episodic dust emissions, especially when evaluated at coarse temporal resolution. Comparisons at these coarser temporal scales therefore present only limited aspects of equifinality.
Minor comments
1. Line 11: The first sentence of the Abstract may require grammatical editing.
2. Line 96: Please cite several studies that use a similar “widely accepted approach.”
3. Line 143: Because the paper focuses on implementing the albedo-based drag partitioning algorithm, it is relevant to include in the main text a description of the proposed roughness scheme, especially in the context of existing schemes. A discussion of its strengths (e.g., greater temporal variability than traditional fixed-value parameterizations and relatively easy implementation) and limitations (e.g., limited applicability to dense canopies) would be appreciated (Chappell and Webb, 2016). Key comments raised by Okin (2023) could also be discussed, including whether normalization of the MODIS BRDF using the fiso parameter adequately removes interference from background darkness and other factors.
4. Line 191: It would be helpful to provide a map of the static source function, perhaps in the Supporting Information.
5. Line 316: In Fig. 4, the annual peak mass emission from MECA (around July) increases substantially in both Exp2 and Exp3 relative to Exp1 and reaches approximately twice the annual peak emission from WNA or ENA. However, this increase is not reflected in the monthly DOD in Fig. 6. What are the possible explanations for this difference?
6. Line 353: Please describe the “Reference” star and the surrounding semicircles in the figure caption.
References
Chappell, A., and Webb, N. P.: Using albedo to reform wind erosion modelling, mapping and monitoring, Aeolian Research, 23, 63–78, https://doi.org/10.1016/j.aeolia.2016.09.006, 2016.
Okin, G. S.: Shadow is related to roughness but MODIS BRDF should not be used to estimate lateral cover, Remote Sensing of Environment, 292, 113581, https://doi.org/10.1016/j.rse.2023.113581, 2023.
Citation: https://doi.org/10.5194/egusphere-2025-5872-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,166 | 284 | 76 | 1,526 | 286 | 220 |
- HTML: 1,166
- PDF: 284
- XML: 76
- Total: 1,526
- BibTeX: 286
- EndNote: 220
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study applies an albedo-based drag partitioning approach to GEOS-Chem (v. 14.2.3) for simulating dust emission globally. The authors compare the simulated spatial and seasonal distribution of dust emission from the current albedo-based drag partitioning approach with those obtained from two existing dust emission schemes that used (1) an observation-based dust source function that masks out non-erodible area, and (2) a simplified vegetation roughness drag partitioning approach, and further validate against atmospheric dust observations and dust deposition observations. The comparison and validation lead to the conclusion that the albedo-based drag partitioning approach eliminates an empirical source function to constrain dust sources and avoids the need to tune dust emission models.
I generally appreciate the equifinality issue within dust modeling as raised by the authors, and the albedo-based drag partitioning approach appears as an encouraging alternative to the classic dust source function approach. However, I have several concerns about the present manuscript regarding (1) the applicability this new approach being overly stated, and (2) the deviates between the current and previous approaches being insufficiently explored. Given the uncertain outcome of my proposed additional work, I recommend rejection of the current manuscript and encourage resubmission.
Major concerns:
A main purpose of the current study seems to eliminate the use of dust source function because, according to the authors, (1) the source function is empirical and not physical, (2) it is static over time. While I agree with these two critiques about the present usage of dust source function, the present albedo-based drag partitioning approach has its own limitations by nature. First, I would not call the present approach as “purely physical”, because (1) the aerodynamical shadow of roughness elements is empirically linked with surface albedo, and (2) the formulation of the present approach, e.g. equation (8), is semi-empirical.
Second, because of the semi-empirical nature of equation (8), it is hard to directly use it for predicting dust emission under climate change because to do so will require a radiative transfer module in the land model to calculate the surface albedo that accounts for the shadow of vegetation. In contrast, a more physics-based drag partitioning approach, e.g. as used in Leung et al. (2024), is more directly applicable in a climate or Earth system model than the present albedo-based drag partitioning approach.
If the authors refine the applicability of the current albedo-based drag partitioning approach to chemical transport models that deal with air quality problems during the satellite era when albedo observation is available, I believe it is fair to say that the albedo-based drag partitioning is an encouraging approach for capturing the episodic and dynamical nature of dust emission. Therefore, I suggest to rephrase the entire manuscript to clearly state the applicability of the present approach. Yet, the relative performance of different dust emission modeling approaches should be further evaluated, which I will cover in the next section of my review.
（1） Which dust source function do you actually use in Exp 1? What is it based off? While the authors state it is the default GOCART setting and cite the Ginoux et al (2001) paper, they also mention that it is “based on the geographical distribution of atmospheric dust”. But indeed the dust source function used in the GOCART default setting was not from atmospheric dust, but from surface topography and has a physical foundation in terms of sediment cumulation (Ginoux et al. 2001). I believe the default DEAD model use the same or similar idea for generating dust emission model (Zender et al. 2003). Please check the dust source function used in your Exp 1 and clarify its empirical or physical foundation.
（2）In Exp2, z0 is obtained at what temporal scale? I see a subscript m in equation (4) but not on line 123. Are they the same thing? Leung et al. (2024) used similar, simplified vegetation drag partitioning formulation, but aerodynamical roughness length above canopy is updated with dynamical vegetation. They show good performance of the dust emission scheme when compare the day-to-day variation in DAOD with multiple sources of observation. So I wonder if the vegetation drag partitioning in your Exp2 reflects the dynamical nature of vegetation, or is it set static? If it is currently set static, could you follow the approach of Leung et al. (2024) but use satellite observation to derive a dynamical vegetation drag partitioning and compare it with your albedo-based drag partitioning approach? Do you still get the conclusion that Exp2 overly flatten the seasonal cycle of dust emission.
（3）If my understanding is correct, a key strength of the albedo-based approach is that the albedo field updates daily, which reflects the dynamical nature of soil erodibility. However, the current comparison on the simulated dust fields between the three experiments, as well as the simulated versus observed dust fields, mostly focus on the climatological scale. Could you show some extreme cases that are successfully captured by your albedo-based drag partitioning approach but missed by the other two schemes? Can you also show the statistics for all three schemes regarding their performance on extreme dust outbreak events?
（4）When you do comparison between model and observation, are all the observational datasets restricted to the year 2016? At least for dust deposition from Albani et al. (2014), it is unlikely that the observational data covers year 2016. I think the temporal match is necessary for model-observation comparison, especially in terms of the extreme events as I proposed in my previous comment.
Minor suggestions
Reference
Leung, D. M., Kok, J. F., Li, L., Mahowald, N. M., Lawrence, D. M., Tilmes, S., ... & Pérez García-Pando, C. (2024). A new process-based and scale-aware desert dust emission scheme for global climate models–Part II: Evaluation in the Community Earth System Model version 2 (CESM2). Atmospheric Chemistry and Physics, 24(4), 2287-2318.
Bullard, J. E., et al. (2016), High-latitude dust in the Earth system, Rev. Geophys., 54, 447–485, doi:10.1002/2016RG000518.