Review of Manuscript

Title: Strong control of the stratocumulus-to-cumulus transition time by aerosol: analysis of the joint roles of several cloud-controlling factors using Gaussian process emulation

Author(s): Sansom et al.

MS No.: egusphere-2025-3104

MS type: Research article

This study investigates the role of aerosol concentration and other cloud-controlling factors in the stratocumulus-to-cumulus transition using a perturbed parameter ensemble (PPE) of large-eddy simulations and Gaussian process emulation. By sampling a wide range of environmental conditions, the authors systematically evaluate both individual and joint effects of key parameters, such as boundary-layer aerosol, inversion strength, and autoconversion rate, on transition time and rain water path. This approach allows for a more comprehensive understanding of the processes driving the transitions and the conditions under which mechanisms like drizzle depletion become dominant.

In my opinion, the LES and emulation framework are both comprehensive and valuable. The manuscript is well-structured, and the analyses are presented clearly. That said, several clarifications and expanded discussions would further improve the manuscript, and I have outlined these in the comments below. I recommend publication after these points are addressed.

General Comments:

• The exclusion of SST as a perturbed parameter may limit the exploration of the deepening–warming transition pathway relative to the drizzle-depletion mechanism. Although the consequence of excluding SST is mentioned briefly in the Results section, its potential importance deserves further discussion in the Conclusions to highlight how its omission may affect the balance of transition pathways represented.
• The calculation of output variable, transition time, is subject to noise, and while this is acknowledged in the manuscript, additional clarifications would be helpful. Specifically, the definition of transition time based on cloud fraction (fc) thresholds of 0.9 and 0.55 appears somewhat subjective. It would really strengthen the study to perform a sensitivity analysis using alternative fc thresholds and report the resulting differences in transition time.
• The model configuration, including the reference trajectory, forcings, and boundary conditions, is idealized, and this limits the realism of the simulations. Although the manuscript notes this briefly in the “perturbation method” section, it should be stated more explicitly in Section 2.1 and reiterated in the Conclusions.
• All results presented are based on idealized modeling, which makes it difficult for direct comparison with observations. A more detailed discussion would be helpful on how the results can be interpreted considering this limitation. For example, to what extent can we trust the results of this study? Could some be artifacts of the model structure? Have previous studies assessed the fidelity of this LES model and its microphysics scheme for stratocumulus-to-cumulus transitions?
• I am not sure if generalizing the results would be straightforward, especially since they are based on a specific region and season. It would improve the manuscript to clearly state the geographical location (Subtropical Northeastern Pacific) and season (currently not mentioned in the manuscript) both in the Abstract and the Conclusions. While the exclusion of semi-direct aerosol effects and highly polluted conditions is noted briefly, this should be clarified in the Conclusions.

Specific Comments:

• L21–23: It would make sense to change this sentence to something like: “Low clouds in the subtropics have a cooling effect on the planet. However, global climate models (GCMs) project a future decrease in their cloud fraction, which would reduce that cooling effect, amplify warming, and contribute to a positive cloud feedback.”
• L69–70: Please check the grammar.
• L79: Is there a better term to use here in place of “calculated”? I am not sure to understand it here.
• L84-92: Please provide more detailed descriptions of Gaussian process emulation and PPE, how they are related, and how they benefit the study of clouds and their transitions, so the concepts are more understandable to readers unfamiliar with these methods.
• L92: Change “in Sansom et al. (2024) we used” to “Sansom et al. (2024) used.”
• L106: Mention the period of the study (date, year, etc.) in Sandu et al. (2010), which your study is based on.
• L108: So, is the reference case an average among many trajectories?
• L112–118: Mention the MONC and CASIM versions, if possible.
• L121: You mentioned wind profiles are retained. How about temperature and humidity (assuming they are used as forcing) in the boundary layer and free troposphere? Also, is subsidence a forcing in your LES, or is it constant for all runs?
• L123: What is the spin-up period in your simulations? Also, specify the initial and final SST values in your study.
• L127-138: It would make more sense if this paragraph were moved up right after introducing CASIM. Also, I assume CASIM is only active in the boundary layer, correct? Then, how does it interact with aerosols in the free troposphere? Did you use a constant or time-varying value for the free-tropospheric aerosols? Does CASIM have a surface source of aerosol in your study? These can be clarified in the manuscript.
• L130: Can you add the particle size distribution (normal, gamma, etc.) for the Aitken, accumulation, and coarse modes?
• Section 2.2 and Table 1: Some explanation is needed regarding how these six variables and their ranges were selected. Were they based on previous studies? Have you done sensitivity tests to exclude other variables? As I mentioned earlier, SST is a very important variable especially for deepening-warming transition. Surface wind speed and subsidence at the inversion level are also critical parameters that excluded. Also, you should clarify whether the values of these variables are chosen at the initial time or averaged along the trajectories.
• L177: “Latin Hypercube” is a critical component of your study, so it would be helpful to dedicate a paragraph or a few sentences briefly describing it and justifying its use. What is its benefit compared to assigning equally spaced values within the range of each variable?
• L182: Here, you correctly mention that the model setup is idealized. This should also be stated when describing the LES setup and reiterated in the Conclusion.
• L189–191: It is important to explain this “new understanding.” What are the ranges of each variable that result in stratocumulus clouds and a transition within your setup?
• L191: Add the final number of total experiments and the number of transitioning experiments.
• L197: Mention the locations of these campaigns (Subtropical Eastern Pacific?).
• L198: This is an important point and should be mentioned in the Abstract and/or Conclusions: this study considers clean to moderately polluted aerosol conditions.
• Figure 1 caption: Mention the total number of experiments shown here.
• L205–206: Are these thresholds rather arbitrary? Have you done sensitivity tests to define transition time based on different values?
• Figure 3: Add the total number of emulators and also the number of MONC simulations. Also, clarify that rain water path is averaged over each simulation (if I understood correctly from the text).
• L254–255: To be more accurate, change this to: “shows three snapshots of liquid water path (L) and water mass mixing ratio (MMR) from the beginning, middle, and end of the simulation, along with time series of fc, L, and rain water path (R).”
• L267: Do any of those studies provide observations for comparison?
• L270: Remind the reader that Yamaguchi et al. (2017) uses an aerosol-aware cloud microphysics scheme similar to yours. It would be helpful to comment on which of the other studies mentioned above also use an aerosol-aware scheme. Also, are Yamaguchi et al. (2017) sensitivity tests based on bulk microphysics?
• Figure 4: Since you discuss boundary layer depth in the text, can you show it on the vertical cross-sections? You defined fc as MMR greater than 0.01. Can you confirm this is the lowest MMR value shown in the cross-sections? Since this case is well studied, I assume observational (at the very least, satellite) exploration has been done by others. It would make sense to compare your simulations with those observations and discuss the fidelity of your model in simulating this case.
• Figure 4 caption: Add “path” after “Snapshots of liquid water.”
• L278–279: When you mention “subset mean in Fig. 5c is a similar shape to the base simulation,” please add that the transition in the base simulation occurs later compared to the mean. Also, change “is” to “has” in that sentence.
• L286–291: You should add that the number of simulations with warm SST is not enough to draw a definitive conclusion. Also, why not select SST as a seventh parameter? SST is such an important factor in the transition that leaving it out needs justification. This relates to the earlier comment on the criteria for including or excluding parameters.
• Figure 5: Add the number of simulations shown in each panel. In panel d, add the mean SST at T1 or T0 for each group to show how different they are. Also, why was the value of “296 K” chosen?
• Figure 6: The inset is missing. Also, what specific time do these points correspond to in each panel showing two parameters?
• L305–306: It would be helpful to add more explanation on the calculations and preparation of Figure 7. You previously stated that 1000 emulators were created. How do you have 1 million points here? It seems that Figure 7 involves more than just averaging the points from Figure 6, so more elaboration is needed.
• L306–307: How is this quantified? Based on the inset in Figure 7, Na has the highest impact, followed by b_aut to a much lesser extent. Delta theta variance is very low, similar to BLz variance.
• Figure 7 caption: Some methodology details seem to be missing, making the first sentence difficult to understand for readers unfamiliar with the method. Also, in the last sentence, do you mean “variance in transition time”?
• Figure 8: Panels k, l, and o are redundant; they are the same as in the previous figure and should be deleted.
• Figure 9: Some tick labels are overlapping. The highest value on the y-axis (20) in panel c is hidden beneath the lowest value on the y-axis (0.0) in panel b. The “80” and “0” values in panels c and e appear as “800.”
• L378–380: The length scale of open-cell stratocumulus clouds is usually greater than 10 km, so I do not think your LES setup could simulate them.
• L384–385: Based on Figure 10, BL Na and delta theta are also important factors. Specifically, BL Na (for all cases and for those with higher R) and delta theta (for cases with lower R) have higher correlations with transition time. This should be mentioned in the text.
• Figure 10: I recommend moving panel f from Figure 9 to Figure 10. That would make all panels in each figure consistent and avoid showing panel f before panel d in Figure 9. It would also help justify having Figure 10 in your paper. Currently, its explanation in the text is less than two lines.
• L386–387: Are the correlation values mentioned here and in Figure 9f statistically significant? Given the small number of simulations and low correlation values in some categories, some of these may not be significant. It would be good to calculate and report statistical significance here.
• L431–437: The issue is not limited to autoconversion. There are various assumptions and tuning parameters throughout a microphysics parameterization. That’s why intercomparison studies often show a wide range across different models and schemes.
• L450–455: Can you elaborate on how this parameterization limitation might bias the drizzle-depletion mechanism or other processes? In particular, I think the shallow boundary layer in your case delays the transition time in the baseline and other simulations.
• L457: If your microphysics include the coarse mode, it should be mentioned in the methodology section where you describe Aitken and accumulation modes.
• L460–461: There is no figure or explanation about the impact of different aerosol modes on the results. If you performed sensitivity tests, please include them in the supplementary material, along with an explanation in the Results section.
• L465–472: This is an important challenge in defining transition time. Another contributing factor is the diurnal cycle: it is possible that some cases labeled as “transition” simply reflect this cycle. This caveat can be added here.
• L465–472: Related to the previous point, you should mention here that the definition of transition time is based on T1 and T2, which themselves are based on fc > 0.9 and fc < 0.55, respectively. These values seem somewhat subjective or arbitrary, so varying them might change some results (unless you’ve done sensitivity tests).
• L474–478: As I wrote earlier and you correctly acknowledged here, the length scale of open-cell stratocumulus clouds is usually larger than the domain size used in your study and in the reference you cite. I reviewed that reference, and although they mentioned open cells in the abstract and other sections, they do not show open-cell morphology. In fact, they note in the middle of the paper that their domain is too small to resolve open cells. So, to avoid confusion and be more accurate, it is best to remove any conclusions about open cells.

Review of "Strong control of the stratocumulus to cumulus transition time by aerosol: analysis of the joint roles of several cloud-controlling factors using Gaussian process emulation" by Sansom, Johnson, Regayre, Lee, and Carslaw, EGUSphere manuscript 2025-3104.

An ensemble of large eddy simulations of the stratocumulus to cumulus transition are performed, building off the Sandu and Stevens (2011) case study with variations in initial conditions (boundary layer depth, moisture and aerosol concentration, the jumps in moisture and potential temperature across the inversion) and microphysics (the dependence of autoconversion on the cloud droplet number).  The joint variation of these parameters is chosen using a latin hypercube sampling with the goal of maximizing the minimum distance between ensemble members in this high-dimensional space.  A brief inspection of the behavior of this ensemble is made before the focus shifts to training the Gaussian process emulator, which "interpolates" properties of the ensemble in the high-dimensional space to show their parametric dependence more clearly.  The analysis of the emulator focused on how the timing of transition and the mean rain water path depend on the various parameters, with the strongest dependence on initial aerosol, then inversion stability and auto conversion for the transition time.

Recommendation: Major revisions

The paper is well written and tells a nice story about the transition.  As a person who has made simulations like those in the ensemble, I wish the authors had shared more about the results of those simulations before shifting to the emulator results.  If there's another manuscript being prepared about those simulations, the authors could highlight that forthcoming manuscript in the paper but might still consider including a bit more in this paper.  The paper would also benefit from more interrogation into the drizzle-depletion vs. deepening decoupling transitions.  I'll make some suggestions along these lines in the major comments below.  Several of my suggestions would involve a fair amount of effort on the part of the authors, so I would understand if they chose not to pursue all of them.

=========================

Major comments (11/240 means p. 11, line 240):

1. A tremendous amount of computational effort was put into the ensemble of LES simulations, but the paper passes quickly over the actual simulations and spends more time talking about the smoothed/filtered/interpolated view of the simulation results presented by the emulator.  It would make sense to have a couple of figures (possibly in a supplement) that summarize the simulation results.  If the authors had some set of plots that they used to understand the broad behavior of the simulations, those would work well for this purpose.  If the authors don't have something like that, my suggestion would be a collection of time series: some subset of SST, inversion height, lifting condensation level, decoupling, accumulated precipitation, boundary layer aerosol concentration, cloud fraction, liquid water path, rain water path, shortwave cloud radiative effect.  Depending on the behavior of the runs, a presentation following left two panels in figure 3 of Chen et al (2024, https://doi.org/10.5194/acp-24-12661-2024) that contrasted the behavior of the drizzle-depletion, deepening-decoupling and no transition group of the simulations, might make the plots easier to decipher.  One or two phase space plots such as Figure 1 of Glassmeier et al, 2019 (https://doi.org/10.5194/acp-19-10191-2019) could also give the readers some idea of how the simulations are similar to and different from each other according to the different metrics.

2. I would also suggest a brief exploration of the contrast between simulations that exhibit drizzle-depletion and deepening-decoupling transitions (maybe resulting in one extra figure if the results are interesting to the authors).  A couple of possible questions: Does the boundary layer aerosol actually decrease in the drizzle-depletion simulations?  Does boundary layer aerosol decrease more than in the drizzle-depletion transitions than in deepening-warming ones, or is the initially low aerosol in the drizzle-depletion more important?

3. Framing the analysis around the influence of aerosol on the timing of the transition is well chosen.  However, the transition is of interest primarily because the associated changes in cloud cover likely cause significant changes to the radiative balance at top of atmosphere (TOA).  Might the authors take a look at how much of the difference in shortwave cloud radiative effect (CRE) or TOA net shortwave flux across the simulations is explained by differences in transition time?  (A consideration of longwave fluxes could also be included but would probably have weaker signals across the transition.)  Here, I recognize that --- if the transition occurred during nighttime hours --- the shortwave radiative fluxes might not vary much if the runs transitioned at different times when the sun was down.

4. Regarding sec 2.3: Would the transition threshold be less noisy/sensitive if the transition metric incorporated some time averaging of cloud fraction (i.e., three hour average f_c < a threshold value) or, alternatively, required that f_c remain below a threshold value for a few consecutive hours?  Might this narrow the emulator uncertainty as represented by the length of the vertical lines in Figure 3a?

==========================

Specific/minor comments

2/30-32: I would be inclined to reference Bretherton and Wyant (1997) at the end of this sentence.

2/32-33: Re-wording suggestion: "Once decoupled, the moisture is supplied to the stratocumulus by cumulus plumes emerging from sub-cloud layer, rather than eddies driven by cloud-top radiative cooling."  Is there really an interval where the stratocumulus moisture supply is "cut off"?  This might be true in warm advection cases when cold SSTs would lead to negative buoyancy fluxes at the surface and near-surface stable layers, but, in the present transition simulations where the SST increases with time, I would have thought that cumulus should be supplying moisture to the stratocumulus layer from the onset of decoupling.  With a small domain, there might be some intermittency in the cumulus occurrence, but I would be surprised if the horizontally-averaged total water flux at the top of the subcloud layer systematically falls to zero at the onset of decoupling in the MONC simulations studied here.  Does that actually occur?  One other point, the "warmer" sub-cloud layer referenced on line 33 is coming from the increase in SST over time.  The process of decoupling leads a cooling and moistening of the sub cloud layer relative to the cloud layer.

2/46-48: I would suggest adding a phrase/sentence to set the reader's expectations, something like, "While increases in aerosol or cloud droplet number concentrations might be expected to delay the transition due to precipitation suppression, Chun et al. (2025) ..."  The work of Ackerman et al (2004, https://doi.org/10.1038/nature03174) has some relevance to this, though I'm not sure whether a citation is needed here.

Also, my understanding is that Chun et al used a prognostic bulk aerosol scheme with two moments of the accumulation mode aerosol size distribution being predicted.  When I read "detailed microphysics scheme", I usually translate that as "bin microphysics scheme" but maybe that's just me.

5/126: Since precipitation is so important to the transition in at least some of the cases, I would encourage the authors to mention that precipitation onset tends to occur sooner in larger domains.  Yamaguchi et al (2017, Fig. 1) show this in transition simulations based on Sandu and Stevens (2011) and also include as nice discussion about this in the middle paragraphs on p. 2345.  Efrani et al (2022, Figure 6) show the effect of domain size on precipitation onset and transition timing in a case study with prognostic aerosol.

If the computational effort isn't prohibitive, could a couple of sample simulations (perhaps one drizzle-depletion and one deepening-decoupling simulation each) be simulated in a larger ~25 (or better yet ~50) km square domain.  It would be supportive of the present results if both simulations showed the same type of transition in the larger domain even if the time of transition changed.  Even if not, it could be mentioned as a qualification.  Making all of the simulations in a larger domain would be much more expensive, so there's no expectation here that they should all be redone in a larger domain.

6/166-7: Please specify the free tropospheric aerosol concentration for both Aitken and accumulation mode aerosol.  Because of entrainment, the boundary layer aerosol concentration would tend towards the free tropospheric value over time in the absence of any other sources and sinks.  If the free tropospheric aerosol concentration was smaller than the initial value in the boundary layer, this would make precipitation more likely towards the end of the simulation.

6/171: Why not replace -1.79 with b_{aut} in this equation?  Then, modify the following sentence towards the end: "... (both in kg kg^-1), N_d is the cloud droplet number concentration (cm^-3), and b_aut is the exponent of cloud droplet number concentration, which has the value of -1.79 in Khairoutdinov and Kogan (2000).  We perturb the value of b_aut to change the autoconversion rate, and this is one of the parameters varied in our PPE."

11/256-264: The diurnal cycle plays a role in these transition simulations, so it would be useful to note the local time of day of these snapshots in the text and/or caption.

12/Fig. 4: Suggestions:

- Add symbols in cloud fraction (and perhaps other panels at left) showing times of the panels (a), (b) and (c).

- Use a log colorscale for liquid water path.  The contrast between 20 and 40 g/m2 is arguably more important than the one between 200 and 400 g/m2.

- Add a timeseries panel at left showing the boundary-layer-mean aerosol concentration.

16/330-332: My reading of Figure 7i is that the time to transition grows shorter with a faster autoconversion timescale for both shallow and deep boundary layers.  It's not clear to me that precipitation is causing the transition to be delayed in shallow boundary layers.

As noted above (2/32-33), the language about the stratocumulus being "cut off" from its moisture source seems too strong in my mind.  

16/348-349: Regarding "Moist boundary layers allow thicker clouds to form, which would then take longer to dissipate through entrainment", wouldn't thicker clouds also be more likely to precipitate and reduce boundary layer aerosol through collision-coalescence scavenging?  It's not obvious to me that thicker clouds would delay transitions in all scenarios.

18/Fig. 8: Might fewer but larger panels at left tell the story as well or better?  That might allow the smaller panels at right to be larger, especially if the transition time colorbar was made vertical, so that those panels could be as large as the others.  The changes in the bottom row of panels at left seem much more gradual than the top row to my eye, so that dropping a couple of those might make the panel-to-panel changes more striking.

Also, label the panels at left so they can be individually reference in the text.  More broadly, labeling all of the sub panels with each figure would be helpful.

18/360: Regarding "For BLNa <100 cm−3, the transition time is very low and almost invariant to the other two parameters.", the transition time seems to vary by ~40 hours with changes in delta theta along the top edge of panel 8b.  If this sentence was referring to something else, maybe specify the figure panel explicitly for clarity.

19/Fig. 9: Might the maximum rain water path or the accumulated surface precipitation or some other metric divide the ensemble into subgroups with less overlap in the cloud fraction evolution?  Did the authors try other metrics and find that they behaved similarly?

19/378: If I understand things correctly, this behavior seems similar to that seen in the fixed-Nd simulations of Sandu et al (2008, https://doi.org/10.1175/2008JAS2451.1).

22/421-423: Could the "ultra-clean layers" observed during CSET be one stage (or the remnants) of such a transition?

23/439-440: It might have been nice to hear more about the tendency of the model to produce less drizzle earlier in the paper.

23/457: In addition to Wyant et al (2022), McCoy et al (2024, https://doi.org/10.1175/2008JAS2451.1) also looked at Aitken buffering in simulations of a case over the Northeast Atlantic Ocean.

=========================

Typographical/wording suggestions (Optional):

0. Does "boundary layer aerosol concentration" include both accumulation and Aitken mode aerosol?  If so, how are they partitioned?  If not, how do the two change over time?  Maybe include time series of both in the suggested plots in major comment 1 above.

2/22: "reduce" --> "weaken". I think the "weaken"/"amplify" contrast might work better.

4/97: "... also varies the dependence of cloud-to-rain autoconversion on the cloud droplet number concentration"

11/254: Insert "(see also, " before citations in parenthesis.

14/307: Left parenthesis before delta theta.

16/all: Maybe include the figure number each time the panels are referenced?  I had to look back a couple of pages to find the figure number.

16/348: "longer transitions" --> "later transitions"

18/367: Maybe add "(R)" after rain water path in the section title, as a reminder to the reader.

22/405: "In this study, we have used an LES cloud microphysics model with aerosol processing to create a perturbed parameter ensemble and explore the effects ..."

23/447: "... that the lack of rain feedbacks _on aerosol and cloud droplet concentrations_ in previous studies may partly explain ..."