The manuscript describes the heterogeneous freezing of water microdroplets containing pure microcline (mc), montmorillonite (mm), or quartz (qu) particles, as well as binary mixtures thereof, as observed and analyzed in a microfluidic droplet assay device.

As natural mineral dust is typically a mixture of various minerals, one key assumption underlying many experiments on the ice nucleation ability of mineral dust is that individual dust particles do not interact in their ice nucleation ability; i.e., their joint effect can be described as the surface-weighted sum of the constituents.

This manuscript presents a counterexample to this assumption by showing that the addition of montmorillonite to a microcline suspension can even decrease the ice nucleation ability of the microcline particles. The authors discuss this in the light of chemical interactions between the two minerals. This is an important observation that will probably stimulate additional work, and therefore, the manuscript should be published in ACP.

The manuscript is written very well, and the findings are presented clearly and concisely. I have several minor questions and suggestions, which are listed below. 

In total, the manuscript contains four parts:

1. Description of the apparatus, especially the addition of a semi-automated freezing detection scheme using a neural network for image analysis (Figs 1, 2)
2. Presentation and discussion of the results obtained for the individual minerals mc, qu, mm. (Figs. 3, 4)
3. Presentation and discussion of the results obtained for mixtures of the minerals at different mixing rations (Figs. 5, 6).
4. Presentation of the results obtained for a sample of Arizona Test Dust (ATD)

General remarks:

Part 4 of the manuscript seems somewhat unrelated to the main part,  could the authors comment in the introduction why it was included?

Remarks on Abstract:

Line 3: Please delete the word “Unfortunately”. Previous lab experiments have focused on individual constituents for a good reason, and for most of the results shown in this manuscript, the above-stated assumption on the additivity of ice nucleation efficiency holds. C.f. also to my remark 3 on part 3.

Remarks on part 1:

1. Please give some references to the CNN setup and/or software packages used. In the central panel of Figure 1, there are abbreviations in the legend, that are not explained.
2. In Fig. 2, it appears that the number of droplet freezing events detected varies strongly between the 5 mixtures. Please explain. What was the total no. of droplets within the field of view for each experiment in Fig. 2? Please give this number in the description or in the Figure caption.

Remarks on part 2:

1. In all freezing curves (Figs 3, 5, 6) a frozen fraction curve for pure water is reproduced from Isenrich et al. 2022. Shouldn´t blank reference measurements with pure water be repeated regularly to exclude contaminations in the tubing? Please comment on this in the manuscript.
2. In Fig. 3, the n_s values collapse to a single curve for quartz, as they do in Fig. 7 for ATD. For mc and mm however, they are disjunct between the concentrations used. Please comment on this surprising finding in the manuscript.
3. It would be helpful to discuss the expected error margins in the data as given e.g. in Fig. 3. If possible, give error bars for one exemplary sample.

Remarks on part 3:

1. In all freezing curves, but particularly in Figs. 5 and 6, the individual symbol shapes are hardly discernible. I do not have an easy solution for this, but plotting symbols open instead of closed and/or at a smaller size might help the reader.
2. It would be helpful if expected “fraction frozen” curves could be constructed for the various mixtures from the results of part 2 using the individual n_s values and surface areas (the expected additive outcome) to compare to the measurement results. This would probably also avoid plotting too many individual data points in Figs. 5 and 6.
3. Discussion section: Even though the non-additive behavior in the ice nucleating efficiency of mc and mm is surprising and interesting, it should be noted that it is only apparent if mm is present in a 100x higher concentration than mc, a situation that is not very likely to be encountered in nature in an individual droplet.

No specific remarks on part 4 or the appendices.

Schardt et al. performed a study of the ice-nucleating activity of mixtures of three different mineral types, in addition to the activity of the individual minerals within a droplet microfluidic device. The main findings showed that K-feldspar microcline dominated the ice-nucleating activity in the binary mixtures when present, even at much lower concentrations than the other mineral in the mixture; While this is often generally accepted as being the case, this is the first time to my knowledge that it has actually been shown rather than being simply expected. An interesting finding was that a mixture of montmorillonite-microcline showed a slight decrease in ice-nucleating activity at cooler temperatures compared to microcline only and the authors provide a theoretical explanation for this phenomenon.

The experiments are well designed and performed and the results confirm some pre-existing notions while offering new interesting findings about montmorillonite-microcline that warrant further exploration. However, I am not convinced that there is enough here for publication as a research article in for ACP as is and may be more suitable for publication as an ACP Measurement Report or perhaps as a paper in AMT. What would elevate the manuscript for publication in ACP would be a further small subset of experiments to test the hypothesis about the effect of Na+ ions in solution from the montmorillonite influencing the microcline. This could, for example, be through testing of the leaching of Na+ from montmorillonite at different concentrations via methods such as ICP or atomic absorption spectroscopy and testing of the influence of these concentrations of montmorillonite on microcline and the influence of Na+ concentrations on microcline.  Given this, I have a few major comments and some minor comments.

Major comments:

1. When each experiment was performed three times, does this mean that a new suspension was prepared and tested each time, or three repeats were performed on the same suspension? I’d be intrigued as to whether a new suspension of “mc 0.001 wt% + mm 0.1 wt%” would act in the same manner. It is also interesting that the bulk of the slope for this binary mixture is very similar to one of the mc 0.001 wt% repeats; there is more of a “tail” as the authors mention, but overall the main part of the line does not look vastly different to this particular mc 0.001 wt% experiment.

1. Page 15: This section is an interesting discussion and could explain some of the effects seen in the mm + mc curve. This is arguably the key finding of the paper,and would benefit greatly from exploration of this. For example, could the concentration of Na+ ions leached from the montmorillonite be estimated or measured and then experiments performed on microcline using aqueous solutions containing similar concentrations of Na+ or a range of concentrations to test the hypothesis?

1. Page 16, line 11: The ATD discussion should be in its own section. However, it is unclear why ATD is being discussed here. What is the context for its inclusion amongst a discussion about binary mixtures of minerals? I appreciate that ATD is in itself a mixture of minerals, but there is no apparent link or discussion between ATD and the other results shown here.

Minor comments:

1. Page 2 line 7: Was XRD not performed on the quartz sample to ensure purity/composition?

1. Page 5, line 14: Some quartzes have been shown to exhibit changes in ice-nucleating activity even when suspended in water for a relatively short time (see Harrison et al. (2019; doi: 10.5194/acp-19-11343-2019)), while sonication is also a very energetic technique. Have the influences of these effects been assessed here to ensure there is no loss in quartz activity due to being in suspension for too long (particularly if also being heated by the sonication process)? Or the potential effect of sonication itself on activity – though I am unaware of any mention in the literature that sonication has an adverse effect.

1. Page 5, line 20: What was the SEM analysis for? Determination of particle size distribution? EDS of the particles for composition analysis?

1. Section 2.3: The CNN-Expert method seems to be the best way of analysing the data is a rapid but more accurate (than CNN only) manner, but not as good as “Expert only”. Is the slight lack of accuracy compared to “Expert only” accounted for in the uncertainties in the data later on?

1. Page 10, line 8: Include reference to Harrison et al. (2019) when referring to variation in activity.

1. Page 10, line 25: As the authors discuss, it would make sense that the ns(T) values may be lower than expected since the larger particles have been removed compared to in the literature. Have the authors also considered that these very dense particles might also be settling in the syringe and tubing, thereby further reducing the size range of the particles that can actually enter the device for analysis?

1. Page 10, line 25: Also, it is very interesting to see the results for microcline compared to the literature, with the ns(T) values falling short of the “expected” activity. Both Peckhaus et al. (2016; doi: 10.5194/acp-16-11477-2016) and Tarn et al. (2018; doi: 10.1007/s10404-018-2069-x) saw similar behaviour when comparing microcline results from similar sized droplets to the literature and this phenomena likely warrants some attention in future. Further, Harrison et al. (2016; doi: 10.5194/acp-16-10927-2016) demonstrated that different microclines can have varying activity. Many microcline studies tend to specifically use one of a handful of samples to ensure results can be compared across techniques/experiments.

1. Page 10, line 27: As noted in an earlier comment, some quartzes can lose activity after being suspended in water for even a relatively short period of time and this is worth bearing in mind here.

1. Figure 3: Please add y-error bars to the ns(T) values, these should at least be calculated from the uncertainty in droplet volume, uncertainties in the measurement of concentration when preparing the suspensions and the BET values.

1. Page 12, line 8 (and Figure 4): It is unclear what the SEM analysis is for, I had assumed particle size distribution as mentioned in the text but this does not seem to have been measured.

1. Page 13, line 1: This part on binary materials should have its own section with heading.

1. Was there no background (pure water) assay performed for these experiments? The baseline data is all from Isenrich et al. (2022), which is quite unusual given that some of the results in this study encroach into the homogeneous freezing area.

1. Page 15, line 7: “SWy-2 montmorillonite is a clay mineral that releases Na+ ions when suspended in water and these ions may exchange with the K+ ions at the surface of the microcline.” – please provide a reference for this.

1. Page 16, line 14: The ATD has also not been processed in the same manner as the other minerals apparently, since a specific surface area has now been assumed based on another study for a particle size range that is completely different to that used for the other minerals (which had been filtered to 0.45 micron). Was the ATD also filtered? This becomes more confusing as it is revealed that the particle size has a large effect on the specific surface area, but it is unclear what particle sizes were used in these experiments and in that case why a normalisation to 22 m2/g is used.

1. Page 18, line 8: This should also be a new section.

1. Figure A1: Given that the feldspar sample contained 97 % microcline, should the literature parameterisations used for comparisons in the earlier figures not be scaled to this? I assume the literature parameterisation (e.g. Harrison et al. (2019)) is represented as 100 % microcline? Perhaps it would not make a substantial difference, but given the experimental values reported here are lower than the literature data this could be one of the reasons (in addition to those discussed elsewhere).

1. Page 19, line 34: “the freezing behavior of Arizona Test Dust (ultrafine fraction) also followed its mineralogical composition” – I do not think that this was particularly discussed in this way in the Results section, it was very unclear what the point or outcome of these ATD experiments was, particularly since none of the other mineral data were plotted.