Review of Jeng et al., Aerosol impacts on regional climate: chaotic or physical effect?

This study proposes an analysis of the role of weather and climate stochasticity impacting the response of the Indian monsoon precipitation to Arabian dust regional radiative forcing. It relies on the statistical analysis of a large ensemble of regional short-term simulations based on a global, dust interactive, atmospheric model to discuss the regional significance of ‘true’ physical dust induced response vs purely chaotic internal variability response. It also examines the number of ensemble members needed to achieve converging and robust results. It outlines that some sub-region like central India where precipitation response depends on meso-scale weather system organization are more prone to internal variability compared to other region where the impact of large scale flow dominates. The study concludes that most of existing studies looking at dust impact on Indian monsoon precipitation did not properly account for these effects, explaining divergence in results especially for central India.

The topic is definitely relevant to ACP, the paper and methodology are in general appropriate, clearly written. Although the topic of stochastic effects / internal variability affecting sensitivity and climate change studies has been explored, seeing it applied on the specific issue of dust /Indian monsoon interaction is definitely interesting in my opinion. Overall I find the paper suitable for publication, after taking into account the following points :

Major comment :

My main criticism concerns the Authors generalizing their conclusion a bit quickly regarding other existing climatic studies. Indeed the proposed simulation protocol uses a lot of members (50), but it also explore the dust induced response on a relatively short time scale of 20 days, characteristic of June of a given year. So each members includes a limited number of meso-scale events for example. Climatic studies are often based on multi-year simulation and examine the impacts of dust on seasonal and yearly averaged precipitation, they thus includes many events and part of the internal variability effect might be smoothed out when averaging. I am not saying that the internal variability does not affect these studies, I am sure it does, but perhaps the convergence toward a consistent physical signal is achieved faster (i.e. with less members) when dealing with climate length simulations. I advise the Authors to be cautious when making conclusion “at climate scale” and regarding “the Indian Monsoon”, or to clearly demonstrate how their results can be be generalized.

Regarding the radiative forcing and significant precipitation response obtained through a robust ensemble average, it would be also good to recall that the corresponding patterns and magnitude reflect a specific june 10-30 2016 situation, which does not cover the entire variability of dust - Indian monsoon interactions.

Finally I did not really understand the method for discussing the validity of statistical significance tests in few ensemble member studies, please see specific comments.

Minor/specific comments :

Title : Perhaps a title more focused on dust and the region of study would be more appropriate.

L45 -50 ; An other useful reference focusing on regional climate models Internal variability

O’Brien et al.. Clim Dyn 37, 1111–1118 (2011). https://doi.org/10.1007/s00382-010-0900-5

L 141 : the simulations are 20 days long, representing a specific month of a specific year. Can we say this protocol « captures the Indian monsoon » ? To me the experiment is closer to a meteorological experiment than a climatological experiment. The intra-seasonal and interannual variability of the ISM are here not captured.

L150 : Could you please mention at this stage if only dust radiative effects or both dust radiative and microphysical effects are taken into account in the experiments ? I understand it was stated in the conclusion.

Question: Despite the simulation time scale being relatively short and since IV develops from small perturbations, can the fact that SST are forced in your experiments affect the noise to signal ratio (and so the relative impact of internal variability) ? Fixed SST creates basically a constant supply of energy and moisture for the perturbations to develop without consistent dampening, perhaps this is likely to enhance stochasticity especially in convective regions. This would also be a contextual difference with climatic studies which consider an interactive ocean /SST.

2.2.2 Generating Perturbed Initial Conditions for Ensembles: It seems that two distinct perturbation protocols are presented but I did not really understand why at this stage. Are they compared later on ?

L 295 : This robust effect of dust on precipitation (100%enhancement) is here quite large and significant. This is a strong result that would need to be more discussed in light of other studies. As I mentioned earlier, caution should be taken regarding how this result can be representative of “dust impact on monsoon” regarding the time-scale addressed.

L360 : check also the previous O’Brien et al. ref which identifies a similar behavior in the convergence as a function of ensemble members.

L 365 : As stated earlier, these studies are based on longer model integrations where the temporal average might already dampen the IV effect seen at shorter time scale . In other words perhaps a 10 member ensemble considering multiyear, seasonal means (which includes many events) could be more robust than the author suggest based on 50 members ensemble of 20 day simulations (which each includes a limited numbers of events). When considering multiyear seasonal means, the convergence towards a physical effect in function of ensemble members might be perhaps faster. So less ensemble member required.

L385. Figure 9. I was wondering how different are the radiative forcings (especially TOA) for E1 and E2.

L395 and Figure 9: About statistical significance: I did not really grasp the method and conclusion here. If you select the E1 and E2 samples to be representative of a type of precipitation response, you automatically increase the statistical significance of the results just due to this preferential sampling, compared to a sample which would contain members with variable type of responses. From this I don’t see how to conclude that statistical tests applied to climatic simulations with small ensemble are not meaningful. Maybe I missing something (or a statistical background), that could be explained furthermore.

L420: Particularly I think when convection is an important component of the meso-scale systems. Orography induced meso-scale system for instance might be less chaotics in term of response to dust.

L435: and other studies.

This manuscript investigates the chaotic vs. physical effects of dust aerosols on Indian Summer Monsoon (ISM) precipitation using large ensemble simulations with the iAMAS model. The study focuses on a 20-day period in June 2016 and quantifies the spread and convergence of dust-induced impacts using 50-member ensembles. The authors use the Indian Summer Monsoon (ISM) system as a case study to show that even with the same physical forcing (e.g., dust aerosol), the simulated response varies widely due to initial condition perturbations. The novel aspect lies in highlighting the limitations of small ensemble sizes in drawing robust conclusions about aerosol effects.

The major comments are:

1. The final paragraph of the introduction should more clearly articulate the main objectives of the study and provide a concise roadmap of the manuscript’s structure. Currently, the paragraph combines motivation and definitions without explicitly stating specific research questions or outlining the paper’s structure.

2. The description of the iAMAS model is currently scattered within the introduction, primarily through citations to previous studies. However, a dedicated and concise model description paragraph is missing from the Methodology section, which is where readers expect to find details about the modeling framework used in the experiments. I recommend moving the relevant model description content from the introduction to Section 2.1, ensuring it covers key features (e.g., dynamics, resolution, aerosol treatment, radiation, and physics schemes) in a self-contained manner. This will improve the clarity and reproducibility of the study.

3. The manuscript lacks clarity on aerosol treatment, especially dust. Please specify:

• Whether both direct (radiative) and indirect (cloud) effects are included. If only direct effects (ARI) are used, this should be clearly stated and justified.
• Whether aerosol-cloud interactions (ACI) are active, and if not, why.
• Whether aerosols are internally or externally mixed, and what assumptions are made regarding their optical and hygroscopic properties.

I recommend that the authors include a dedicated subsection or an expanded paragraph in Section 2 covering these aerosol processes in sufficient detail.

4. The choice to simulate only 20 days during the early monsoon season (June 10–30, 2016) warrants further justification. This short time frame captures only the monsoon onset and not the full seasonal evolution, intraseasonal variability, or withdrawal phase. While the period may have been chosen to isolate certain synoptic features or reduce computational cost, the manuscript should explicitly state the scientific rationale for selecting this specific window. Additionally, it would strengthen the study to discuss how representative this period is of broader monsoon-dust interactions. If this is intended as a case study, that should be clearly stated to avoid overgeneralization of the results.

5. Figure 3c (AOD from the “Sensitive” simulation) appears to show nearly no aerosol loading over much of South Asia, including the Indo-Gangetic Plain — a region known for high anthropogenic aerosol concentrations even during the monsoon period. Since the “Sensitive” case only excludes Arabian dust emissions, anthropogenic aerosol emissions should still be present in the simulation or was it only dust emissions enabled?

6. Figure 10 suggests that ensemble sizes beyond 30 members yield only marginal improvements in the convergence of dust-induced precipitation responses. Given the computational cost associated with running large ensembles, could the authors clarify whether they consider 30 members as an optimal threshold for similar studies? Additionally, do they recommend any specific criteria or diagnostics to determine when further increases in ensemble size (e.g., to 40 or more) are justified?

7. Would the authors expect similar sensitivity and ensemble size requirements if the primary response variable were temperature rather than precipitation? Could temperature fields, given their typically lower chaotic variance, be used to isolate aerosol radiative effects with smaller ensemble sizes?

8. Since the study focuses predominantly on dust aerosols and specifically targets the Indian Summer Monsoon, I recommend the review title to more accurately reflect this focus.