Review of "How well are aerosol–cloud interactions represented in climate models? – Part 2: Isolating the aerosol impact on clouds following the 2014–15 Holuhraun eruption" by Jordan et al

This manuscript compares general circulation model simulations of the effects on clouds and radiation from the 2014-15 Holuhraun volcanic eruption in Iceland to satellite observations of clouds and aerosols. 

In general the paper is well written, but needs some important clarifications as noted in specific comments below. I have several major concerns:

1. I’m confused why the authors use monthly means and do not try to actually use daily data, especially from models to get a more process focus. This lessens the utility of the paper, but I think they have the data to look at. 

2. The author focus on the 'predominantly volcanically polluted (PVP) region using SO2 values. But many of the effects are outside of the PVP regions. Why is that? This might be a case where the PVP region could be better defined if it were done daily rather than monthly given the evolution of emissions and meteorological variability. 

I would strongly advise looking at daily data with at least one model to see if it matters for the correlation and attrition, and the volcanically affected or not.

3. More model description is warranted (see specific comments below).

Specific comments:

Page 4, L88: Absent all volcanic emissions? Or just Holuhraun?

Page 5, L92: Explain how the GCMs do this (or not). Are aerosols and cloud drops prognostic in all the models? Or are cloud drops diagnostic (set as a function of aerosols and activation). I assume CNRM is diagnostic number, but autoconversion (Kessler) does NOT depend on drop number, while it does for Menon and KK? A bit more explanation is warranted.

Page 5, L93: Aerosols affect entrainment in the models? How? It thought most GCMs did not include this.

Page 8, L178: so that implies that the area is not really pristine, but can be affected by anthropogenic emissions as well as Holuhraun.

Page 8, L182: but you are still averaging over months. Since you have the meteorology, why not look at daily data? More points, larger gradients. This smooths out the analysis and reduces the chances your differentials are affected by averaging.

Page 8, L186: what is the ‘null hypothesis’? Are you testing significance? Stippled points are significance? Please clarify.

Page 9, L190: but the PVP region also does NOT have a significant change.

Page 9, L191: again, this argues that region is polluted.

Page 9, L192: I would argue they do NOT capture observations well since the largest observed changes are NOT in PVP regions, while the models have largest changes in the PVP regions.

Page 10, Figure 3: are these box plots individual location averages? What gives the spread. It has not been well defined.

Page 11, L208: what about the ‘polluted’ region S. Of the PVP region? Shouldn’t you comment on that: seems like MODIS might have a larger effect than the models.

Page 11, L214: exasperated is not the right word. I think you mean ‘increased’ or exacerbated. That’s still a bit awkward.

Page 11, L228: Isn’t the frequency of precipitation (and susceptible cloud) also important? It’s not really the mean, it’s the number of days that it is effective at precip suppression.

Page 12, L:230: Figure 4: is this a lot or a little precip in the PVP regions. It’s hard to tell from the figure or from what you have said here.

Page 12, L237: Are they really ‘excellently’ capturing the spatial distribution of LWP. What are the potential issues in measuring LWP?

Page 13, L239: I don’t think the ensemble matching the magnitude of the observed really means anything: if you only included one model from each family you might get a different answer. Some models have very different patterns. I think you could do more to look at the spatial distribution of effects: why are there large effects OUTSIDE of the PVP regions in models and obs? Do they correlate with aerosols? Nd? Precip even? Does this hold at a daily scale? Not just s smeared out monthly average.

Page 14, L246: But figure 6 shows some large spread in met effects (large blue box and whiskers) and of different sign. What is going on?

Page 15, L254: that seems pretty self evident and consistent with lots of other work.

Page 15, L257: you should figure out how KK varies across this subset of models: is it tuned differently?

Page 15, L258: is the lack of CF response shown anywhere?

Page 17, L266: Is a reader to read in figure 8 that almost all of the points have significant differences? even ones where the difference is nearly zero (e.g. regions where it switches from positive to negative). That does not seem correct for significance testing…

Page 17, L268: it’s not noise if it is significant? It’s significant meteorological variability right?

Page 17, L285: if clouds get thicker, then low clouds will trap more LW and reduce OLR. I think it’s a consequence of higher TAU (more LWP). The LW offsets the SW somewhat.

Page 17, L291: also due to more daylight….

Page 19, L304: this global efficiency seems very dependent on location and timing of emissions no? Is it really relevant? You only put in emissions for Sept and Oct? Did you calculate effects through February? If not, why not? Seems simple to do.

Review of “How well are aerosol–cloud interactions represented in climate models? – Part 2: Isolating the aerosol impact on clouds following the 2014–15 Holuhraun eruption” by Jordan et al.

In this study, the authors utilize the 2014-15 Holuhraun eruption as a natural laboratory through which they explore aerosol-cloud interactions (ACI) in an ensemble of state-of-the-art general circulation models (GCMs). Following on from Part 1 of this paper, which addresses the modeling of the eruption’s emissions from the volcanic plume, the authors focus on ACI in marine stratocumulus clouds. To this end, 8 models are assembled with a variety of different cloud microphysics parameterization schemes, with a primary focus on modeled precipitation suppression. The authors contrast the output from models where the eruption is present to models where it isn’t and then compare these results to MODIS observations. They find that, while the models seem to adequately capture the Twomey effect (the first indirect effect), there remains considerable disagreement in modeled adjustments (the second indirect effect).

The paper’s writing is clean and generally easy-to-follow, and it’s well-written overall. However, as it stands, the paper feels lacking in certain details, and some revisions and additions are recommended before publishing. These are described point-by-point below.

General comments

1. In the paper, the authors identify precipitation suppression as the expected mechanism for enhanced adjustments. For the majority of models, the autoconversion parameterization used is Khairoutdinov and Kogan (2000) (hereafter KK2000), which relies on several parameters (Nd exponent etc.) to determine the rate of autoconversion. It is suggested (P20, L333) that differences in the configuration of KK2000 between models may account for these differences. I think this is likely true, but more analysis here to interpret the causality of these differences would make the result more robust. How do these KK2000 parameters compare between models? Do the autoconversion parameters match up with the outcomes shown in Figure 7? Exploring these parameterization differences would be a straightforward pathway towards discerning the causality behind the differences in the model and strengthening the results, allowing the authors to more specifically describe the discrepancies that led to this result.
2. Following on to the previous point, additional explanation of the ACI pathways present in the model would be useful for the interpretation of results. As discussed, precipitation suppression is the main pathway the authors are concerned with in this paper (although, as mentioned, ACI effects on entrainment are present in models, though weakly). How, mechanistically, through the parameterizations implemented, do the authors expect GCM ACI to manifest from the enhanced SO2? Are there significant structural differences between microphysics schemes that may play a part in output disagreement, or are you more concerned with parametric differences?
3. For this work, only two months are analyzed: “September and October 2014 when the eruption is strongest” (P3, L76). However, this is not the whole eruption, and considerable additional SO2 was emitted after the end of October. Why not continue this analysis through the remaining months, especially given that you’ve already created the framework? The exclusion of the remaining eruption is not well-motivated, and as-is undermines the completeness of your conclusions. Before publication this should be addressed, either in the form of a stronger motivation for neglecting the remainder of the eruption or by including those months in the analysis.

Specific comments

P3, L52: Here and elsewhere, you describe the region as “near-pristine”. While I don’t disagree that the region can get pristine conditions, I imagine that this must be highly dependent on the prevailing winds/meteorology (as has been observed in the Azores to the south, see Gallo et al., 2023). To that point, you later describe how anthropogenic emissions from the UK affect a large region below the PVP domain. I don’t think this is necessarily a huge problem, given that you’re accommodating for anthropogenic emissions by mirroring them in your Hol and NoHol, but if you wish to characterize the region as such, a relevant climatology citation may be useful here. Alternatively, additional specific commentary regarding the prevailing background aerosol regime in this region to the Holuhraun-dominated regime may be useful for “setting the stage”.

P8, L186-7: The statistical methods used throughout are a bit unclear. What is the “local” null hypothesis that is being rejected? Can you describe, briefly, the FDR method? More elaboration is required here, especially given how frequently these methods are used throughout the rest of the paper.

P11, L214: I am unsure if “exasperated” is the right word here- did you mean “exacerbated”?

P17, L290-1: While it is not unreasonable that the time during which the eruption was most intense had the highest RF, could this not also have been affected by the differing solar radiation between the two periods of time? At these extreme latitudes, the amount of sunlight changes significantly over the course of September and October – I wonder if that may be influencing the relative RF magnitudes during these months, and if that sunlight difference can be easily normalized between the two periods.