the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 Jul 2026
Deciphering the effect of mixing state on ice nucleation using single particle measurements and probabilistic simulations of mixed dust-biological ice nucleating particles
Abstract. Ice-nucleating particles (INPs) in soils are complex mixtures of mineral dust and biological material, yet the influence of particle mixing state on ice nucleation activity remains poorly constrained. Here, we generated synthetic soils using montmorillonite and Snomax to evaluate the suitability of current analytical methods for characterizing and predicting the ice nucleation activity of complex particle mixtures. Particle composition was characterized using single particle mass spectrometry (SPMS) and computer controlled scanning electron microscopy with electron dispersive x-ray (CCSEM/EDX), as well as Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) for surface chemical analysis. Ice nucleation experiments were conducted using a Continuous Flow Diffusion Chamber (CFDC). CCSEM/EDX classified particles predominantly as externally mixed while SPMS demonstrated greater sensitivity to Snomax and identified more internally-mixed particles. Residual particles from both techniques showed enhanced biological markers compared to the bulk population. Frozen fractions did not scale with mass mixing ratios, and simulations indicated that mixing state had minimal effect on ice nucleation activity when total Snomax mass was conserved. We also found some limited evidence that incorporating particle mixing state from measurements improves simulated frozen fractions. These results demonstrate that accurate prediction of INP activity in dust-biological mixtures requires constraints on the abundance and particle-level distribution of active biological material, rather than bulk or population-average compositional descriptions alone.
- Preprint
(6058 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (extended)
- RC1: 'Comment on egusphere-2026-3882', Anonymous Referee #1, 22 Sep 2026 reply
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 147 | 87 | 20 | 254 | 26 | 27 |
- HTML: 147
- PDF: 87
- XML: 20
- Total: 254
- BibTeX: 26
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of “Deciphering the effect of mixing state on ice nucleation using single particle measurements and probabilistic simulations of mixed dust-biological ice nucleating particles” by Cornwell et al., 2026
Synthetic soil consisting of montmorillonite and Snomax at four mass ratios are characterised with three single-particle techniques and the particle’s ice nucleation activity and ice residual composition are determined with a CFDC/PCVI setup. The results are compared against a probabilistic freezing simulation under several assumptions about particle mixing state. The central findings are that frozen fraction does not scale with the nominal mass mixing ratio, that mixing state topology matters far less to the simulated outcome than total Snomax mass, and that none of the three composition techniques achieves closure with the CFDC measurements.
General comment:
The research addresses an important open question and uses a comprehensive set of measurements to address it, but there are multiple issues in the presented analysis that make this study incoherent.
First, the residual composition data that the paper's closure argument depends on are inconsistent with the paper's own account of how these particles freeze. At -20°C, where the pure dust curve (Fig. 9a) shows no activity, dust classified particles still make up 15 to 45% of the measured ice residuals, up to 2400 times more than independent freezing of components, the paper's own modelling assumption, would predict. Pure Snomax residual composition at -20°C, the one dataset that would directly verify what is freezing in the temperature range is missing.
Second, the miniSPLAT classification narrative that Section 3.1 builds its central "far more sensitive than CCSEM/EDX" argument turns out to be less solid at every point checked. The classification percentages quoted in the text for the 100:1 and 1000:1 soils don't match bulk values in Tab. 3, by up to 30%. The specific evidence offered for "enhanced Snomax signal in residuals," has a small effect size (Cohen's d 0.11-0.22) with 43-45% distributional overlap.
Third, the probabilistic simulation behind the paper's other headline conclusion, that particle-level mixing state matters less than bulk composition, rests on a montmorillonite nucleation rate that, when checked against both the paper's own measurement data and independent size-selected montmorillonite literature (Ickes et al., 2017; Pinti et al., 2012), appears to substantially understate the activity, on a biological probability parameter with no physical upper bound that is exceeded for part of the assumed population, and on a mixing geometry. The conclusion currently follows as much from these choices as from the underlying physics.
Specific comments:
Line 33: Provide supporting references for the higher ice nucleation activity of organic-rich dust compared to mineral dusts.
Line 38: The part of the question about the conceptual framework can be answered. Tang et al., 2026 published the CNT based conceptual framework of ice nucleation in mixed particle populations. Their findings should be incorporated.
Line 43-44: Explain why single particle experiments are more suitable for measuring internally mixed particles.
Line 51: Clarify what the mentioned studies have to do with ambient measurements of INP composition. The described experiments were carried out with INP of known composition. Also, Augustin-Bauditz didn’t use a freezing assay but LACIS, which can not be used for ambient measurements.
Line 57: Specify what real world applicability. The measured curves are considerably more binary (flat plateau, then step) and the montmorillonite component considerably weaker, than the gradual, continuously narrowing transitions reported for real agricultural soil dust (O'Sullivan et al., 2014; Steinke et al., 2016; Tobo et al., 2014). Replace the assertion with a substantiated discussion of these differences and their causes.
Line 70: Snomax is introduced as a general proxy for biological ice nucleation activity. Coauthor Steinke et al. (2016) found that heat treatment did not reduce a real soil's ice activity, suggesting non-proteinaceous entities can also drive soil ice activity, which Snomax, being entirely heat-labile, cannot represent.
Line 75: State the actual masses or dilution scheme used, especially for 1000:1 and 100’000:1. The used notation X:1 SS could be introduced here. It is missing an explanation in the current version.
Line 82, 253-254: Clarify the difference in “processed similarly” that can explain the removal of 98+, 100+ signal.
Line 88: Why was the Teflon bag used for Snomax instead of the chamber?
Line 93: Explain what “resolution of 10” means in this context. It seems not to be the sheath-to-sample flow ratio.
Line 94: Explain why it is considered easier to sample 700nm particles with the AAC than DMA.
Line 95: Multiple charged particles contribute mainly if small particle sizes < 200nm are selected. For 700nm they are negligible.
Line 101: Specify if offline samples are for SEM and how they were prepared.
Line 105-110, Table 2, 3 (-20°C residual rows): Dust classified particles make up 15% (100:1 SS), 45% (1000:1 SS), and 75% (100’000:1 SS, N=4 only) of the measured -20°C residuals. Given the pure dust frozen fraction is ≈10⁻⁴ at this temperature, there is a 2-3 order of magnitude mismatch that the same check at -34°C does not show. Explain how to reconcile this discrepancy. No composition data are reported for pure Snomax residuals at -20°C, despite pure Snomax freezing equally well at -20°C and -34°C, and residual data being available for Snomax at -34°C. This is the measurement that would directly confirm what is freezing in the temperature range f* is calculated from (line 339-340).
Line 108-109: According to Fig.7 in Wex 2015, Snomax containing samples should freeze at -10°C. Also clarify at what T and RH the nucleation section was for these experiments.
Line 118, 120, Fig. D1: SEM/EDX standards report roughly 35 to 46% carbon on pure montmorillonite dust particles, which have no structural carbon. At the stated 20 kV accelerating voltage, the Kanaya-Okayama range gives an interaction volume of approximately 4 to 5 μm for both carbon and an aluminosilicate approximating montmorillonite, five to ten times larger than the ~0.5-0.9 μm particles analysed. This makes substrate sampling from the carbon grid unavoidable at this voltage, and negligible accompanying nitrogen argues against generic hydrocarbon contamination as the source. State whether the quantification corrects for substrate contribution and reflect this in the Section 4 "lacks the sensitivity" framing (line 418). Choosing a lower voltage (typical 5kV) instead 20kV would have narrowed the volume and helped to reduce the substrate signal. Address how this affects the particle classification.
Line 121, Table 1: Two of the five SEM/EDX classes are not mutually exclusive, "Dust and biological" against "Dust" for 0.1≤P<0.5%, MD≥4%, CNO<96%, and "Processed biological" against "Processed dust and biological" for 1≤MD<4%. Clarify the resolution rule used by the classification code.
Section 2.3.1: consider including SEM images showing pure and internally mixed particles.
Line 144: clarify why the signal disappears.
Line 144-146, Figure 4: The classification thresholds (log10(84+,86+) "greater than the 0.0891" for Snomax, "less than 0.0272" for dust) are mathematically incorrect as written, log10 of a relative peak area can never be positive. The numbers only work as raw, linear RPA cutoffs. Drop "log10" from the threshold description or replace the numbers with their log-transformed equivalents. Separately, those cutoffs appear to sit at the edges of the pure Dust and pure Snomax distributions shown in the same figure rather than being independently derived, and the pure Dust distribution spanning roughly log10 ≈ -3 to -1.6 that its upper tail approaches the dust/mixed boundary. Report the misclassification rate obtained by applying the stated thresholds to the pure Dust and pure Snomax reference populations themselves, since this is directly checkable with data already in hand and bears on every "internally mixed" fraction reported in Section 3.1.
Figure 4: The distributions appear to be diagonal segments joining bin heights rather than a conventional histogram, which visually broadens the tails of each distribution. Consider replotting as an actual histogram, or provide the underlying bin edges and counts, so the separation between populations can be assessed accurately.
Table 3: Several entries report mean ± SD from samples too small to be a meaningful uncertainty (e.g., n=2 for the −20°C dust residual, n=4 for the 100,000:1 SS residual).
Line 162: clarify if diffusion driers are part of the CFDC or upstream sample conditioning.
Line 164: specify the T, RH conditions used for experiments and residence time in each instrument section.
Line 165 ff.: Clarify if ice nucleation occurs in the nucleation section or exclusively in the evaporation section where ice cannot grow.
Line 181: provide the used loss factor.
Line 206-213: The mapping from the five SEM/EDX classes to the three summary categories (Bio, Bio+Dust, Dust) used in Figure 2 a) is never stated explicitly. Explain how the separation works.
Line 209: Provide explanations of why Snomax didn’t coat the dust and why Snomax was expected to coat the dust. Provide SEM images for illustration.
Figure 2: According to your Fig. 9 the ice nucleation activity of dust at -20°C is negligible or not detectable. Therefore, in Fig 2. e) and f) there should be no residuals containing only dust and for the pure dust experiment, no residuals at all. Explain where the dust residuals come from at -20°C.
Line 227-238: The statement that 10:1, 100:1, and 1000:1 soils "have mass spectral features that look similar to the pure Snomax" is not supported by Fig. 3, where the 1000:1 standard spectrum (j) shows its dominant peak shifted to roughly m/z 40-45, unlike Snomax, 10:1, and 100:1 (a, d, g), which share a dominant peak at roughly m/z 70-75. The stated classification percentages ("nearly 90%, 60%, and 45%... classified as Snomax," "1%, 3%, and 4%" dust) do not match Tab 3's own bulk values (83.6%, 47.2%, 15.9% Snomax; 4.2%, 16.1%, 47.6% dust). Recompute these percentages and reconsider whether the "1000:1 is still Snomax-like" survives the correction.
Line 241-245: providing the log10 of values would allow to compare with Fig.4. The reported enhancement of 84+/86+ in residuals versus standards is real and significant. However, the effect sizes are small (Cohen's d = 0.11-0.22), corresponding to roughly 43-45% distributional overlap between standard and residual populations, a random residual particle has close to even odds of falling below the standard mean. Report effect size or overlap alongside the significance claim. This should be analysed more carefully as it is one of the two pieces of evidence offered for "components responsible for initiating ice nucleation are primarily derived from Snomax" (line 246-248).
Line 245: The threshold is comfortably below, at 0.089. Clarify the interpretation provided here.
Line 249: The dust marker (98+/100+) disappears on wetting regardless of freezing, so residual particles classified "Dust" in Tab. 3 are identified by absence of the Snomax marker rather than a positive signal. Confirm this was accounted for deliberately and consider it alongside the -20°C residual mismatch (line 105-110) and the threshold circularity (line 144-146) as contributing mechanisms.
Figure 3: Indicate 84+, 86+ with vertical lines.
Line 265: Dubble check if this estimate holds if C% is corrected (see comment below).
Line 271-273: The bimodal size distribution, with a secondary mode at 200 nm "for all samples," does not appear equally present across samples. Check the 1D size histograms and confirm whether "all samples" is accurate. Provide an explanation for the second mode.
Line 275: Provide the underlying K-S statistics, p-values, or threshold, e.g., in a table alongside Fig. D1 to back up "Statistically similar" size distributions.
Line 276: Clarify why the size distribution of size selected particles can be dissimilar. In Fig.5 100’000:1 has a pronounced small particle fraction peak. 100:1 and dust seem much more similar and have a low Cohen’s d. Check the interpretation of Cohen’s d<0.2 negligible difference.
Line 278-279: On the contrary, Fig. D2 and Fig.6 show that residuals and standard size distribution are most different for Snomax, almost 2 std. deviations apart.
Line 283-284: Clarify how the hypothesis is in line with the fact that AAC and miniSPLAT processed the same particles. Provide SEM images showing particles before and after the soil creation process.
Line 288, 518: Provide the density of Snomax and montmorillonite K10. It could be expected that Snomax has about half the density of montmorillonite. Provide an explanation of the similar ρeff considering the shape factor from the SEM analysis.
Line 294: In contradiction, Fig. B1 shows a broad size distribution.
Fig A2, Fig 6: In addition to the wetting experiment size distribution, Fig. A2 a) shows the same data as Fig.6 a) and A2 b) the same data as 6 f). But there are noticeable differences in the size distributions that should be identical. Check the data used. The “growth” of Snomax is incredible considering the mass or density difference of a factor greater than 2. Also, Snomax grows to the size selected by AAC is surprisingly. The “standard” measurement should be repeated.
Line 295, 299: Clarify the relation of dva to the AAC selected diameter. According to Eq. B1 density and shape factor also affect the dca. Estimate by how much the shape factor would need to change through restructuring.
Line 298: As there is literature showing no restructuring, SEM not showing it (Fig. 5), consider that miniSplat data for standard Snomax is corrupted.
Fig. 6: Compare the measured ρeff to the calculated ρeff using the density of Snomax and M10 together with the shape factor from SEM in Fig.5.
Line 315: Clarify why this equation could give the coverage. There is no physical reason to assume this linear composite scales linearly with the fraction of particle surface covered by Snomax material.
Line 321: Snomax consists of bacterial cell fragments. Clarify the assumption that Snomax should cover the dust and not just stick to it as separate fragments. Provide supporting evidence that Snomax contains soluble, ice nucleation active material that can cover a dust particle. SEM images would be helpful.
Line 322-324: That surface coverage "remains largely unchanged" from 100:1 to 1000:1 is a stretch given the uncertainties, 4.1±0.5% and 3.1±0.2% do not overlap.
Line 329-330: Clarify if 255 K is a limit due to the wall temperature of the upper section activating particles into droplets.
Line 342: The ToF-SIMS surface coverage enrichment at the 10:1 soil tracks the same non-linear trend as the biological frozen fraction. State this as an explicit mechanistic hypothesis (preferential Snomax adsorption via montmorillonite's cation exchange capacity, line 69) rather than a correlation.
Line 344: Specify “similar trend” and what can be concluded from this similarity.
Figure 9: Dubble check the data from previous studies. There seem to be fewer data points in Hartmann et al., 2013 Fig. 3 or Wex et al., 2015 Fig. 7 than shown here.
Line 345: explain why it is plausible that Snomax is unevenly distributed across particle sizes.
Line 363: Scenarios of external and partly external mixtures are not explained in the Appendix C.
Line 365-366: On the contrary, Fig.10 shows overprediction of measurements in several panels. Show the simulated pure dust curve overlaid on the measured points, with the exact fitting procedure stated (see line 491).
Line 366: Soften the conclusion, the resulting mixing ratio of the used soil is not known.
Line 368: The statement that mixing state alone does not significantly affect the simulated ice fraction is a consequence of lambda being assumed proportional to particle volume, not an independent finding. State this as model-dependent; correct the same unqualified restatement in Section 4 (line 422-423).
Line 372: Quantify “better job”, it mainly indicates that the detected amount of Snomax is closer to reality for these samples for the respective instrument.
Line 380: explain why including imprecise information on mixing state is better than guessing.
Line 381-382: Specify the circumstances and how it varies with mixing state and analytical technique.
Figure 10, Line 198: Clarify why Monte Carlo simulations are needed when both the dust term and the Snomax term have analytical solutions. At 100’000:1 ratio the wiggles in Fig.10 are sampling noise, not physics. The analytical solution would avoid misinterpretation.
Tang et al., 2026 show that for monodisperse particles the internal mixture is always the more efficient (higher frozen fraction) mixing state than external mixture. The same could be expected, but Fig 10 a), b) show the opposite. Please clarify the influence of the particle size distribution and the core shell assumption on the simulation result.
Line 391: Using measured residual composition (Figure 11) to check the freezing model's correctness is inconclusive, and given the -20°C mismatch (line 105-110), a residual-measurement artefact is more probable.
Line 397: This is not a prediction. The comparison is circular and does not demonstrate that the treatment of freezing is good. The simulation reproduces the input fractions Fig. 11 a) b) are therefore the same as Fig. 2 a) b) which show the input.
Line 405-406: Based on the ice nucleation data, a more plausible explanation is that the amount of Snomax in the synthetic soil was lower than the amount identified by the SEM/EDX classification. The SEM/EDX doesn’t lack sensitivity, the miniSPLAT classification is vastly overestimating it.
Line 406: Clarify what supports this suggestion. More plausible is that the sample synthesis didn’t produce the intended mass ratio.
Line 410: Because the bulk mixing ratio is ill-defined from soil synthesis and analytical techniques, this conclusion is not supported by this work.
Line 417-421: Avoid implying miniSPLAT is detecting the mixing state more accurately, it is not. On the contrary, the classification is further from the real composition revealed by the ice nucleation experiment. Instead of leaning in an instrument sensitivity discussion focus on the classification introduced in this paper. The classification does not work. It is not an issue of instrument sensitivity.
Line 424: Specify what evidence in what circumstances and present the evidence.
Line 425: “utility varied by mixing ratio” is a euphemistic description for “it didn’t work”. The mixing of the sample is what the analytical technique is supposed to provide.
Line 427: A two-component system cannot be considered complex.
Line 431: The insensitivity is unexpected. Provide an explanation using the work of Tang et al., 2026 as basis.
Line 435: specify the limitations of analytical techniques and simulation and define how they change for different model systems. An informative experiment would have been using externally mixed, size selected particles in different ratios as calibration of analysis technique and ice nucleation experiment.
Line 437: That miniSPLAT and ToF-SIMS provided better information is not supported by the data shown in Fig.10.
Line 440: Stronger supporting evidence is needed for the utility of SPMS considering the SPMS could overestimate the bio particle abundance more than SEM/EDX.
Line 441: Clarify what coated systems are referred to and provide supporting evidence that they are less sensitive to mixing state.
Line 442: The guidelines are not justified based on what is presented in this study and should be removed.
Line 443-445: There are too many issues with this study to support the conclusion. Especially because the synthetic dust could not be characterized.
Line 450: Clarify how restructuring changes chemical composition.
Figure B1: specify what particles were measured and the AAC settings. Explain why the distribution is wider than dva and why the APS was used instead. The APS is not included as part of the setup shown in Fig.1.
Line 486: The assumed geometric standard deviation of 1.2 for the AAC selected size distribution is used without explanation of its origin. Add justification and a sensitivity range as well as how the frozen fraction changes based on it.
Line 493: The montmorillonite Jhet is roughly 10 -100 times lower than in Ickes et al. (2017) or Pinti et al. (2012). Clarify the units of the fit function and how it was obtained.
Line 495: Hartmann et al., 2013 used LACIS described as cloud chamber on line 46 in the introduction not a CFDC.
Figure C1: clarify what is meant by “corrected” λ.
Line 501: Explain how fitting the volume dependence was done with only one measured size of Snomax particles and provide more detail what Fig.C1 displays and how it relates to the CFDC measurement. Lambda is linear in particle volume with no stated upper bound and the Snomax size distribution exceeds the diameter where lambda exceeds 1. Cap lambda at 1.
Line 507: What is meant by “at the first step”?
Line 508: The core-shell geometry assumes a coherent Snomax shell, but Snomax is disintegrated, freeze-dried cell material, what likely adsorbs onto montmorillonite is released protein, not a shell with the same volume-proportional INA content calibrated on intact particles.
Line 535-538: Fig D1, D2 show Cohen’s d not K-S.
Technical corrections:
Line 19: provide a more appropriate citation from a peer reviewed article, e.g., Field 2017
Line 22: clarify what the Cornwell 2023 citation is in support of.
Line 71: "Pseudomona Syringae" should be Pseudomonas syringae
Line 87: “100 L Teflon bag” instead of “100L teflon”
Line 98: refer to Fig.1
Line 138: Add the closing bracket.
Line 140, 230, 236, 242, and throughout the manuscript: space before comma instead after 84+, 86+
Line 146: 84+ plus sign not in exponent.
Tab.3: add units to ρeff
Line 149: TOF-SIMS
Line 161: ice nucleation chamber
3.1. and Line 204, 222: capitalize Single
Line 220: Remove the stray “and”
Line 231: (Creamean et al., 2013)
Line 252: (see Appendix A).
Line 253: missing space after comma.
Line 260: “techniques” instead of “particles”.
Line 275: 100:1 instead of 10:1.
Line 286: “density” instead of “concentration”
Line 281, 287, and throughout the manuscript: use either Dva or dva for diameter.
Fig. 5: add a colorbar
Fig. 6: x-axis units should be nm. Add units to ρeff
Line 306: Remove the stray “with”
Line 328: The manuscript switches from °C to K. Use one or the other throughout the manuscript.
Line 328: "propreties" should be properties.
Line 361: Remove the stray “in”
Line 363: Appendix C, not 2.5
Line 376: missing space after comma
Line 391: Remove the stray “and and”
Line 393: should it read: simulated residual -34°C particle type … using experimental characterisation for standard samples
Line 397: under-predicted
Figure 11 c) Dust has wrong color code.
Figure A2: Figure doesn’t match the caption. Units are missing from the x-axis. Panel c mentioned in the caption is missing.
Figure B1: add units to x-axis
Line 529: define ai
Line 537: Remove the stray “the”
Fig D1, D2: captions are swapped. Consider cutting the mirrored part of Fig. D1.
References:
Augustin-Bauditz, S., Wex, H., Denjean, C., Hartmann, S., Schneider, J., Schmidt, S., Ebert, M., and Stratmann, F.: Laboratory-generated mixtures of mineral dust particles with biological substances: characterization of the particle mixing state and immersion freezing behavior, Atmos. Chem. Phys., 16, 5531–5543, https://doi.org/10.5194/acp-16-5531-2016, 2016.
Field, P. R., Lawson, R. P., Brown, P. R. A., Lloyd, G., Westbrook, C., Moisseev, D., Miltenberger, A., Nenes, A., Blyth, A., Choularton, T., Connolly, P., Bühl, J., Crosier, J., Cui, Z., Dearden, C., DeMott, P., Flossmann, A. I., Heymsfield, A. J., Huang, Y. H., Kalesse, H., Kanji, Z. A., Korolev, A., Kirchgaessner, A., Lasher-Trapp, S., Leisner, T., McFarquhar, G., Phillips, V., Stith, J., and Sullivan, S.: Secondary Ice Production – current state of the science and recommendations for the future, Meteor. Mon., 58, 7.1–7.20, https://doi.org/10.1175/AMSMONOGRAPHS-D-16-0014.1, 2017.
Hartmann, S., Augustin, S., Clauss, T., Wex, H., Šantl-Temkiv, T., Voigtländer, J., Niedermeier, D., and Stratmann, F.: Immersion freezing of ice nucleation active protein complexes, Atmos. Chem. Phys., 13, 5751–5766, https://doi.org/10.5194/acp-13-5751-2013, 2013.
Ickes, L., Welti, A., and Lohmann, U.: Classical nucleation theory of immersion freezing: sensitivity of contact angle schemes to thermodynamic and kinetic parameters, Atmos. Chem. Phys., 17, 1713–1739, https://doi.org/10.5194/acp-17-1713-2017, 2017.
O'Sullivan, D., Murray, B. J., Malkin, T. L., Whale, T. F., Umo, N. S., Atkinson, J. D., Price, H. C., Baustian, K. J., Browse, J., and Webb, M. E.: Ice nucleation by fertile soil dusts: relative importance of mineral and biogenic components, Atmos. Chem. Phys., 14, 1853–1867, https://doi.org/10.5194/acp-14-1853-2014, 2014.
Pinti, V., Marcolli, C., Zobrist, B., Hoyle, C. R., and Peter, T.: Ice nucleation efficiency of clay minerals in the immersion mode, Atmos. Chem. Phys., 12, 5859–5878, https://doi.org/10.5194/acp-12-5859-2012, 2012.
Steinke, I., et al. (2016), Ice nucleation activity of agricultural soil dust aerosols from Mongolia, Argentina, and Germany, J. Geophys. Res. Atmos., 121, 13,559–13,576, doi:10.1002/2016JD025160.
Tang, W., Arabas, S., Curtis, J. H., Knopf, D. A., West, M., and Riemer, N.: The impact of aerosol mixing state on immersion freezing: insights from classical nucleation theory and particle-resolved simulations, Atmos. Chem. Phys., 26, 9221–9255, https://doi.org/10.5194/acp-26-9221-2026, 2026.
Tobo, Y., DeMott, P. J., Hill, T. C. J., Prenni, A. J., Swoboda-Colberg, N. G., Franc, G. D., and Kreidenweis, S. M.: Organic matter matters for ice nuclei of agricultural soil origin, Atmos. Chem. Phys., 14, 8521–8531, https://doi.org/10.5194/acp-14-8521-2014, 2014.
Wex, H., Augustin-Bauditz, S., Boose, Y., Budke, C., Curtius, J., Diehl, K., Dreyer, A., Frank, F., Hartmann, S., Hiranuma, N., Jantsch, E., Kanji, Z. A., Kiselev, A., Koop, T., Möhler, O., Niedermeier, D., Nillius, B., Rösch, M., Rose, D., Schmidt, C., Steinke, I., and Stratmann, F.: Intercomparing different devices for the investigation of ice nucleating particles using Snomax® as test substance, Atmos. Chem. Phys., 15, 1463–1485, https://doi.org/10.5194/acp-15-1463-2015, 2015.