the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Jul 2026
Droplet sedimentation regulates liquid water in marine advection fog: Large-eddy simulations with interactive sectional microphysics
Abstract. Marine advection fog strongly affects visibility and maritime operations, yet numerical models frequently overestimate its liquid water content. Although fog-top radiative cooling is known to promote condensation, how size-dependent gravitational droplet sedimentation redistributes and removes the resulting liquid water remains poorly quantified. Here, we use the University of California, Los Angeles Large-Eddy Simulation model coupled with the sectional aerosol–cloud microphysics module from a Lagrangian perspective to quantify this process during a typical advection-fog event over the northwestern Pacific. Longwave radiative cooling near the fog top promotes condensational growth and the formation of large droplets, generating intermittent sedimentation signals that propagate downward through the fog layer. Lead–lag correlations show that sedimentation initiated near the fog top reaches the fog base after approximately 1 h. The liquid water path budget identifies fog-top radiative cooling as the dominant source (about 30 g m-2 h-1) and gravitational sedimentation as the largest microphysical sink (approximately 20 g m-2 h-1), offsetting nearly two-thirds of the radiatively driven production. Droplets with radii larger than 10 μm dominate the sedimentation flux. Sensitivity experiments confirm the robustness of this mechanism: removing sedimentation causes excessive liquid-water accumulation and substantially deepens the fog layer, while suppressed collision–coalescence or enhanced aerosol loading weakens sedimentation by limiting large-droplet formation. These results identify droplet sedimentation as a vertically coupled control on liquid-water redistribution and fog-layer evolution, implying that marine-fog parameterizations need to represent sedimentation and large-droplet microphysics explicitly.
- Preprint
(6446 KB) - Metadata XML
-
Supplement
(387 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-2980', Anonymous Referee #1, 22 Aug 2026
-
RC2: 'Comment on egusphere-2026-2980', Anonymous Referee #2, 01 Sep 2026
This manuscript investigates the role of droplet sedimentation in regulating liquid water in marine advection fog using large-eddy simulations coupled with the SALSA microphysics scheme. The representation of fog remains a challenge in numerical weather prediction, and a detailed examination of the processes controlling fog liquid water is therefore worthwhile.
However, I have substantial concerns regarding the positioning of the manuscript relative to previous work and the evidence supporting its central mechanistic interpretation. In particular, several key aspects presented as motivation or novel physical insight, including the importance of sedimentation for fog liquid water, the sensitivity of sedimentation to the representation of the droplet size distribution, and the coupling between cloud-top sedimentation, entrainment, and the radiative feedback, have been established in previous studies. The last of these is long established in the stratocumulus literature (Ackerman et al., 2004; Bretherton et al., 2007), which the manuscript does not cite. Relevant fog-specific work, including Schwenkel and Maronga (2019, 2020), Richter et al. (2021), Boutle et al. (2022), and Rodriguez-Geno and Richter (2024), is likewise not discussed or insufficiently distinguished from the present contribution. The statement at L94–97 that Barve et al. (2025) provided the first reproduction of the observed bimodal droplet spectrum in marine fog using an LES coupled to a Lagrangian cloud model is therefore, not correct in light of the literature listed above. Richter et al. (2021) simulated marine fog with an LES–LCM specifically on the bimodal droplet-size distribution.
The element that could be potentially novel is the test those mechanisms mentioned above to marine advection fog over an SST front, which is a different regime from the nocturnal radiation fog and coastal fog cases treated previously, together with the multi-day evolution and the fog-to-stratus transition. This is a worthwhile contribution, but it is a case study that confirms and extends an established mechanism rather than one that identifies a new mechanism, and the manuscript should be framed accordingly.
I therefore recommend that the authors substantially reconsider the relevant literature and revise the manuscript accordingly. Previous work should be properly acknowledged wherever the present interpretations reproduce or extend already established findings. Most importantly, the authors need to state clearly what is genuinely new relative to these studies and reframe the manuscript around that contribution. In my view, this requires sufficiently substantial revision and repositioning that I recommend rejection of the present version with the possibility of resubmission after major restructuring.
I recommend rejection of the present version and encourage resubmission after substantial restructuring and additional analysis.
Major comments
Major1
As mentioned above, the author needs to properly acknowledge the previous works. and reframe the novelty of the work.
The treatment of collision–coalescence could be potentially interesting. Some previous studies (e.g., Schwenkel and Maronga (2020) and Boutle et al. (2018)) indeed neglected the collision–coalescence. Rodriguez-Geno and Richter (2024), however, examined this process directly for marine fog using a Lagrangian cloud model, and reported that collisions open a pathway for aerosol activation through collisional activation.
The physical interpretation in this paper, however, is incomplete. The authors state that collision–coalescence weakens the bimodal structure near the fog base while also noting that the responsible mechanism remains unclear. More importantly, the interpretation appears internally inconsistent. The authors first state that the bimodal structure becomes less pronounced in NoCollCoal, implying that collision-coalescence promotes the large-droplet mode, but subsequently state that collision-coalescence weakens the bimodal structure. This apparent contradiction should be resolved before the associated mechanism is discussed. Given the emphasis placed on the bimodal droplet-size distribution and on large-droplet formation, this requires clarification.
Major2
The horizontal grid spacing appears very coarse relative to the shallow fog layer. The mean fog-top height discussed in the manuscript is approximately 280~m, whereas the grid spacing is $\Delta x=\Delta y=150$~m and $\Delta z=5$~m near the surface. The corresponding three-dimensional filter width is approximately 48~m, which is a substantial fraction of the fog-layer depth, meaning much of the energy containing eddies may not resolved. The domain also contains only 20 × 20 grid columns in the horizontal, which limits both the range of resolved eddy scales and the horizontal sampling available for the domain-averaged statistics. This is directly relevant to the intermittent, streak-like sedimentation events described in Section 4.3.
Most previous LES studies of fog have employed considerably finer grid spacing, often of order a few meters to several tens of meters, and Schwenkel and Maronga (2020) also discussed the sensitivity of fog simulations to very fine spatial resolution (isotropic 2m grid box). Yang et al. (2024) used a horizontal grid spacing of 20 m and a vertical spacing of 1 m below 50 m over a 1500 m × 1500 m domain, and stated explicitly that this resolution was chosen in order to capture the delicate radiative and turbulent processes.
The choice of a 150~m horizontal grid spacing therefore requires much more justification. In particular, the authors should discuss whether the relevant turbulent structures within such a shallow fog layer are adequately resolved and, provide a resolution-sensitivity test demonstrating that the main conclusions are robust.
Major3
The manuscript repeatedly discusses cloud-top radiative cooling, entrainment, and the interaction between sedimentation and the vertical evolution of the fog layer, but the turbulent vertical-velocity scale is not analyzed. The paper that the author referred to: Yang et al. (2025), for example, discussed the ratio $v_{\mathrm{sedi}}/w_{\mathrm{turb}}$ (i.e., Rouse number) when assessing the relative importance of sedimentation and turbulence. This comparison is important because the relevance of gravitational settling depends on its magnitude relative to turbulent vertical transport, rather than on the terminal velocity alone.
This consideration is also relevant to the effective sedimentation radius inferred in the present study. The estimate based on $\Delta z/\Delta t$ implicitly interprets the propagation speed of the sedimentation signal as a particle fall velocity. Such an interpretation should be evaluated together with the resolved vertical velocity and the SGS turbulent velocity scale. I therefore suggest that the authors quantify the relevant turbulent velocity scales and discuss the inferred sedimentation velocity in this context.
Major4
Figure~13 is truncated at a radius of 20~$\mu$m, whereas the manuscript places considerable emphasis on droplets with an inferred radius of approximately 25~$\mu$m. The droplet-size range most directly relevant to this interpretation is therefore not shown. I suggest extending the radius range of Fig.~13 and, make the large-droplet tail easier to evaluate.
Because the interpretation relies strongly on the large-droplet tail, particularly the inferred radius of approximately 25~$\mu$m, the authors should provide additional information on the effective spectral resolution in this size range. Since the cloud-droplet bins are defined by dry diameter while wet sizes evolve prognostically, please clarify how the wet radii shown in Fig.~13 are diagnosed and how many bins effectively represent droplets in the approximately 15--30~$\mu$m radius range. This information would help assess how well the large-droplet tail relevant to sedimentation is resolved.
In addition, the CTRL and NoSedi experiments are absent from Fig.~13, making it difficult to assess how the droplet-size distribution changes between the observationally evaluated control simulation and the sedimentation sensitivity experiment. Including these cases would substantially improve the interpretation.
The very long time range shown in Fig.~13 also makes the evolution of individual sedimentation events difficult to follow. For example, around 00:00 UTC on 1 July, parts of the droplet-size distribution appear to terminate rather abruptly. From the current figure it is difficult to determine whether this behavior reflects rapid growth toward larger sizes followed by sedimentation, or another process. A shorter-time-scale view of selected events, together with a clearer representation of the large-droplet tail, would make the interpretation much more transparent.
Major 5
The authors also suggest that the substantial difference between the previous bulk-microphysics results reported in Yang et al. (2024) and the present bin-microphysics results is mainly related to sedimentation and the representation of the large-droplet tail. This may be the case, but the comparison currently involves two substantially different microphysical frameworks.
Moreover, simulations in Yang et al. (2024) used a much finer horizontal resolution, while the present simulations use a different domain size and resolution. These differences make a direct attribution of the contrasting results to microphysics alone difficult. Given that these studies form a closely related sequence of papers, I suggest that the authors quantify, the sedimentation flux obtained with the bulk and bin microphysics schemes under otherwise comparable simulation conditions.
Furthermore, because the manuscript places substantial emphasis on large droplets with an inferred radius of approximately 25~$\mu$m, a size-resolved decomposition of the sedimentation flux would be useful together with the corresponding droplet number concentrations. This would directly demonstrate which part of the large-droplet tail dominates the sedimentation flux.
References
Ackerman et al. (2004), Nature, 432, 1014–1017.
Bretherton, Blossey, and Uchida (2007), Geophys. Res. Lett., 34, L03813.
Boutle et al. (2018), Atmos. Chem. Phys., 18, 7827–7840.
Schwenkel and Maronga (2019), Atmos. Chem. Phys., 19, 7165–7181.
Schwenkel and Maronga (2020), Atmosphere, 11, 466.
Yang et al. (2021), Mon. Weather Rev., 149, 3183–3203.
Richter, MacMillan, and Wainwright (2021), Boundary-Layer Meteorol., 181, 523–542.
Boutle et al. (2022), Atmos. Chem. Phys., 22, 319–333.
Rodriguez-Geno and Richter (2024), Q. J. R. Meteorol. Soc., 150, 4580–4593.
Yang et al. (2024), Atmos. Chem. Phys., 24, 6809–6824.
Barve et al. (2025), Q. J. R. Meteorol. Soc., 151, e5022.
Yang et al. (2025), Proc. Natl. Acad. Sci. USA, 122, e2505421122.
Citation: https://doi.org/10.5194/egusphere-2026-2980-RC2 -
RC3: 'Comment on egusphere-2026-2980', Anonymous Referee #3, 11 Sep 2026
The manuscript “Droplet sedimentation regulates liquid water in marine advection fog: Large-eddy simulations with interactive sectional microphysics” presents a series of fog LES conducted with the UCLALES-SALSA model. The primary emphasis is on droplet sedimentation, and the authors use a baseline simulation along with some limited sensitivity experiments to explore LWC budgets within an advection fog case. This study is a follow-on to previous work by some of the authors, using a more realistic treatment of droplet microphysics.
There are certain aspects of this work that are relevant and valuable to the ACP community. Overall, however, I’m left wondering exactly what we have learned, scientifically, for two key reasons:
1. One of the main conclusions is summarized in lines 509-513, in that droplet sedimentation is a large sink of liquid water and can regulate other properties of the fog layer. Is this new? It seems fairly intuitive that large droplets settling out would be a large sink, and I am not aware of any studies or models which argue to the contrary. In a similar vein, the authors talk about the deficiencies of standard microphysical models, but almost make it seem like they don’t account for sedimentation at all. Perhaps I am missing something, but I think the authors would need to do a better job providing the current state of understanding of the role of sedimentation, both in observations and models, to better highlight what new insights this study provides.
2. At the same time, I also end up wondering how many of the results in this manuscript are subject to non-negligible numerical issues. This includes things like grid resolution but also modeling approximations. In some places, this can be alleviated with more model details, but in others I think the authors would have to demonstrate that their conclusions aren’t being largely (or entirely) driven by numerical artifacts. At the very least, this might require a grid convergence process.Below I have itemized several points throughout the text that I think would need to be addressed, some of which provide more details on my broader concerns above.
Line 74: I think the authors should be explicit in the units of what they call “flux”. The number flux would be the concentration times the Stokes settling rate, which would scale like d^2. I believe that the d^5 comes from converting number to water volume?Line 94: Barve et al. (2025) was not the first to see this. Years before, Schwenkel & Maronga (2020) and Richter et al. (2021) both observed bimodal droplet distributions using LES with a Lagrangian droplet model. Furthermore, in these models, since they use Lagrangian microphysics, the effects of settling were naturally incorporated and properly assigned as a function of droplet size. I suppose one could argue that sedimentation was not “explicitly examined”, but it was certainly incorporated into the model in a physically meaningful way.
Line 103-108: There is other work that should be cited here too, like Pope & Igel (2025)
Line 131: I think the authors need to make a stronger case for what exactly is missing in previous models. They simply say that processes are “highly simplified”, but what exactly does this mean in the context of sedimentation? Given the motivation for this study, I think the authors need to take more time explaining how this process is typically represented, what the shortcomings are, and how their results shed light. The current text throughout almost makes it seem like sedimentation isn’t even accounted for in other models, which is not true.
Equation 171: How reliable is this? My understanding is that these Nd and LWP retrievals, especially for fog, are notoriously unreliable. Perhaps I am mistaken
Line 197: Could the authors be a bit more specific as to how aerosols and droplets are treated? Is there a single continuous size spectrum that contains both unactivated aerosols and activated droplets? Or are there two different distributions – one for droplets and one for aerosols? Related: How is activation handled within this framework? The authors simply say that it doesn’t “rely on empirical CCN parameterizations”, but this doesn’t say what it actually does.
Line 208-213: The grid resolution seems potentially problematic. For many numerical schemes, aspect ratios beyond 10:1 can yield enormous numerical errors, and the authors are using a ratio of 30:1 at the surface. For fog specifically, studies have also shown that the fog properties are highly sensitive to grid resolution (Maronga & Bosveld 2017, Wainwright & Richter 2021). Furthermore, for such coarse horizontal resolution, I would expect that many of the processes that the authors consider important (turbulence, droplet settling) would be poorly resolved, and that near the surface they would be largely parameterized. I think the authors would need to demonstrate that their conclusions are not sensitive to grid resolution or aspect ratio.
Line 239-241: I don’t understand this: how is the SST from observations used, but at the same time a “cosine function” was used to fit it? Does the SST behave perfectly periodically in time/space along the trajectories shown in Figure 2? Something about this does not make sense. The blue line in Figure 4 likewise does not look sinusoidal.
Line 246-256: This is related to my comment above about the reliability of remote sensing for aerosol properties, but how confident are the authors in these numbers for the time/location they are trying to represent? Why do they not show a comparison of the initial condition alongside the observed estimates in Appendix A?
Line 260: How exactly is the sea salt emission implemented into SALSA? The corresponding bins at the surface are simply populated at this rate? Something else?
Equation 3: Where does subgrid turbulence come into play here? The horizontal resolution is very coarse, and yet nothing is said about the subgrid contribution to the various LWP tendencies. Near the surface, my guess is that these actually dominate. This seems very concerning given that it is precisely this balance that the authors hope to understand better.
Fig S1: Given how many times Figure S1 is referenced in the text, why is it included only in the supplemental information?
Figure 4 and elsewhere: These are horizontally averaged profiles?
Line 382: This statement seems central to the objective of this study, but is stated without much evidence. There seem to be many differences between this study and Yang et al. (2021, 2024), as well as many possible nonlinear explanations for reductions of near-surface LWC. To simply state that it “is likely related to the frequent sedimentation of fog droplets from the fog top” seems like pure speculation. I think the authors either need to reword this, or provide proof that this is the explanation for the difference in near-surface LWC.
Line 405-406: Again, this seems like speculation and not a direct interpretation of the authors’ results. Since they are actually injecting sea salt aerosols, do they actually see that this is necessary to maintain in-fog aerosol populations?
Line 520: Related to above: this is precisely where near-surface representation of mixing and dispersion is critical. The authors conclude that it’s the absence of sedimentation that leads to large near-surface LWC (which I agree is at least partially true), but with no mention of the LES wall model or how near-surface LWC is being deposited in the absence of the sedimentation term, this again seems like speculation. It might be the case that this LWC exceeding 0.4 g/m^3 is entirely a numerical artifact. (note that I’m not arguing that it necessarily is, only that it’s a possibility which the authors seem to be ignoring)
Line 536+: How is collision coalescence parameterized in the context of SALSA? As with other points made above, it is hard to put things into perspective without any details on the numerical implementation.Conclusions: This section is a little odd, in that it repeats a lot of the detailed, quantitative findings. I think it would be much more impactful if it was more concise and summarized the key findings, along with a broader perspective on where this fits into the community’s understanding of fog processes and modeling.
References:D. H. Richter, T. MacMillan, and C. Wainwright, “A Lagrangian cloud model for the study of marine fog,” Boundary-Layer Meteorology, vol. 181, pp. 523–542, 2021, doi: 10.1007/s10546-020-00595-w.
J. Schwenkel and B. Maronga, “Towards a better representation of fog microphysics in large-eddy simulations based on an embedded Lagrangian cloud model,” Atmosphere, vol. 11, p. 466, 2020, doi: 10.3390/ATMOS11050466.
N. H. Pope and A. L. Igel, “Counteracting influences of gravitational settling modulate aerosol impacts on cloud-base-lowering fog characteristics,” Atmospheric Chemistry and Physics, vol. 25, pp. 5433–5444, 2025, doi: 10.5194/acp-25-5433-2025.
B. Maronga and F. C. Bosveld, “Key parameters for the life cycle of nocturnal radiation fog: a comprehensive large-eddy simulation study,” Quarterly Journal of the Royal Meteorological Society, vol. 143, pp. 2463–2480, 2017, doi: 10.1002/qj.3100.
C. Wainwright and D. H. Richter, “Investigating the sensitivity of marine fog to physical and microphysical processes using large-eddy simulation,” Boundary-Layer Meteorology, vol. 181, pp. 473–498, 2021, doi: 10.1007/s10546-020-00599-6.
Citation: https://doi.org/10.5194/egusphere-2026-2980-RC3
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 195 | 88 | 26 | 309 | 27 | 19 | 25 |
- HTML: 195
- PDF: 88
- XML: 26
- Total: 309
- Supplement: 27
- BibTeX: 19
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of the manuscript egusphere-2026-2980
Droplet sedimentation regulates liquid water in marine advection fog: Large-eddy simulations with interactive sectional microphysics
Summary
This study uses the UCLA large eddy simulation model to learn more about the liquid water budget of a maritime fog. The model is coupled to a specific high resolution, multi-bin microphysics model. The study focusses in particular on the role of droplet sedimentation in governing the LWP, the radiative cooling at the fog top and the fog depth. They conclude the representation of droplet sedimentation is key for the fog evolution and visibility. Fog is challenging to forecast in NWP models, so more fundamental understanding from the LES point of view. The paper is also well written, concise and with high quality figures. Here and there some additional details in the modelling setup are needed, and in my view the conclusion section need to be rewritten.
Recommendation: Major revisions
Major comments:
-1. In my view the Conclusion section needs to be rewritten. The section is very long and does at first instance only repeat the results section. The Conclusion section should answer a well-posed research question that is postulated in the Introduction section. Moreover I prefer to have the Discussion section in front of the conclusion section. The Discussion section should focus on the framing what has been added by this study compared to earlier work, and what is still to be done.
-2. I would love to see an assessment of the UCLA model distribution of subgrid vs resolved scales in the modelled fog over time and in the vertical. At the fog top usually the stratification is strong and small scales are not resolved. This may inhibit the use of LES as technique. Please add some assessment.
-3. Related to the previous point: I will mention it below as well, but some kind of sensitivity study to the model vertical resolution will be appreciated, and will mature the manuscript. A DZ of 5 m as used here is relatively coarse these days, especially for a weather phenomenon dominated by small scales. The mentioned Waerstad study already used 2 m grid spacing 8 years ago (an earlier generation supercomputer), so I would love to see the robustness of the model results in this study against the vertical resolution.
- 4. Analysis. The paper focusses very much on the LWC profiles in time, and tries to relate that to the sedimentation flux, which is logical given the goal of the paper. However I think it would be good show some time series of u* and/or w* and/or TKE at a relevant level to show how the turbulence develops throughout the simulation. Turbulence is a key transport mechanism for liquid water as well, so it would be good how it interacts with radiative cooling at the top, and with the sedimentation flux.
Minor comments:
Ln 48: models also are characterized by excessive mixing, so sometimes fog and low clouds disappear too quickly or do not form. Please rewrite
Introduction: role of initialization and lead time on the NWP/modelling skill is not discussed in the Introduction. If a model is initialized over the ocean with sparse observations, one cannot blame the model physics, neither the poor resolution. In addition, although over land I have seen several studies for the Cabauw tower sites where the 24-48 h forecast was better than the 0-24 h forecast. So spin up and lead time play a role. Please discuss.
Ln 162: 2-m visibility?
Ln 175 (and later on in more detail). A quality assessment of ERA5 for the selected case is not presented? ERA5 usually does not do a good job on fog, for the modelling reasons provided in the introduction. Same for MERRA-2. Please provide an assessment what is the quality of your initial conditions.
Moreover, I do not understand why different reanalysis products have been used for the physics and for the air quality. Using CAMS data from ECMWF for the air quality would imply same physics and dynamics would have been used, which is more consistent.
Ln 186: “UCLALES is a three-dimensional boundary-layer model with turbulence closure”. I do not get this, if it is and LES model, then I expect there is NO turbulence closure, since the turbulence should be largely resolved. So the paper need to make more clear whether it works with and LES model or with a RANS model.
Ln 205: … evaluated in simulations of liquid-phase clouds. Please add some words what was the outcome of the evaluation.
Ln 208-213: please add some words of justification for the chosen settings. Is 5 m sufficient vertical resolution not too coarse? Especially at the fog top where we expect a strong inversion, and LES models are known for having problems at strong stratification, especially in coarse resolution (eddies are too small at the top). For example, Wærsted et al (2019) used 2 m vertical resolution for modelling a dense radiation fog with substantial gravitational settling of droplets.
Ln 236: was the geowind constant with height? Please add.
Ln 242: was the divergence constant with height? At the surface there cannot be any divergence … (w=0 at surface). So how was the vertical w and divergence profile?
Ln 266: the 10-m wind speed was here set the same as the geostrophic wind speed. This violates the idea that the 10-m wind speed is reduced compared to the geowind by boundary layer friction. Please comment.
Ln 281: the reference to Waersted et al. (2018) is missing in the reference list. I think it should be Waersted et al. (2019).
Ln 336: Please add some justification for the use of the Kunkel formula. After 1984 multiple formula’s have been proposed as alternative as well. The Kunkel formula was also not derived for the sea fog, so some more justification of its use is needed. At least some kind of sensitivity assessment is needed to show to what extent your conclusions depend on the specific use of Eq 4.
Ln 337: it is important to note in which unit the LWC needs to be entered in the formula and what is the unit of VIS in that case.
Figure 4: please add an uncertainty measure (measurement uncertainty or confidence interval) to the dots in panel b.
Ln 384: “intermittent downward-extending streak-like features”. Please provide a explanation for this behaviour. Personally I think this is a numerical artefact. In case the fog grows by 1 grid cell, the fog layer as a whole gets extended by 5 m, which means all thermodynamic variables need to be redistributed again. At least this is how I have seen it in fog simulations in1D models