Mixed-Phase Microphysical Evolution in Large Eddy Simulations of Tropical Cumulus Congestus: Developing and Evaluating a Laboratory-based Ice Multiplication Parameterization of Freezing Drops

Stanford, McKenna; Keinert, Alice; Fridlind, Ann; Kiselev, Alexei; Knopf, Daniel; Ackerman, Andrew; Leisner, Thomas; Lawson, Paul

doi:10.5194/egusphere-2026-4713

Preprints

https://doi.org/10.5194/egusphere-2026-4713

Preprints

21 Aug 2026

| 21 Aug 2026

Mixed-Phase Microphysical Evolution in Large Eddy Simulations of Tropical Cumulus Congestus: Developing and Evaluating a Laboratory-based Ice Multiplication Parameterization of Freezing Drops

McKenna Stanford, Alice Keinert, Ann Fridlind, Alexei Kiselev, Daniel Knopf, Andrew Ackerman, Thomas Leisner, and Paul Lawson

Abstract. Ice microphysical processes modulate cloud structure, evolution, and Earth's radiative balance, yet secondary ice production (SIP)—whereby fragmentation enhances ice number concentrations (N_ice) beyond what ice-nucleating particle (INP) populations alone can explain—remains poorly constrained. We develop a drop-shattering parameterization based on laboratory-observed pressure release event frequencies during drop freezing and evaluate it, alongside a water-activity-based immersion-freezing model for primary ice formation, in large eddy simulations (LES) of a tropical cumulus congestus case from NASA CAMP²Ex—the second of a two-part study extending liquid-phase results from Part I into the mixed-phase region. Bin and double-moment bulk simulations are evaluated against in situ aircraft observations from 0 to -15 °C. The baseline parameterization negligibly enhances N_ice; a 10× multiplier on per-event splinter numbers—reflecting substantial production uncertainty—increases N_ice by 1–2 orders of magnitude. The bulk scheme reaches localized maxima near 10³ L^-1, while the bin scheme reaches 10–20 L^-1, reflecting fundamentally different collision-kernel structures between the schemes. A primary-secondary ice feedback emerges exclusively in the bin scheme, driven by INP enrichment of precipitation-sized drops through collision-coalescence and INP accumulation; this feedback is absent in the bulk scheme due to its lack of aerosol core mass tracking. The 10× parameterization partially reconciles a 1–2 order-of-magnitude deficit in simulated concentrations for sizes > 200 μm relative to observed particle size distributions, with turbulence-induced collision enhancement essential for conditioning SIP efficiency. Together, these bin and bulk implementations provide a foundation for improving SIP representation in large-scale models.

Received: 14 Aug 2026 – Discussion started: 21 Aug 2026

Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Chemistry and Physics.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

McKenna Stanford, Alice Keinert, Ann Fridlind, Alexei Kiselev, Daniel Knopf, Andrew Ackerman, Thomas Leisner, and Paul Lawson

Status: final response (author comments only)

RC1: 'Comment on egusphere-2026-4713', Anonymous Referee #1, 21 Sep 2026

The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-4713/egusphere-2026-4713-RC1-supplement.pdf

Citation: https://doi.org/10.5194/egusphere-2026-4713-RC1
RC2:
'Comment on egusphere-2026-4713', Anonymous Referee #2, 22 Sep 2026
REVIEW for manuscript #egusphere-2026-4713: “Mixed-Phase Microphysical Evolution in Large Eddy Simulations of Tropical Cumulus Congestus: Developing and Evaluating a Laboratory-based Ice Multiplication Parameterization of Freezing Drops” by Stanford et al.

Overview
This manuscript develops a new parameterization for secondary ice production (SIP) by fragmentation of freezing drops (FFD), grounded in KIT laboratory measurements of pressure release event (PRE) frequency, and evaluates it in DHARMA large eddy simulations of a CAMP2Ex tropical cumulus congestus case using both bin and double-moment bulk microphysics, with primary ice represented by the water-activity-based immersion freezing model (ABIFM). The work is well motivated and carefully documented. The parameterization is built on a physically distinct and clearly documented quantity. The central quantitative findings, however, rest on a 10× multiplier applied to the laboratory-derived splinter number, without which the new parameterization has essentially no effect on simulated ice number concentrations. The different responses between the bin and bulk schemes are expected. My specific comments are listed as follows.

Major comments
The 10× multiplier is the load-bearing element of the study and is not bounded. Every substantive result, such as the 1–2 order-of-magnitude Nice enhancement, the PIP–SIP feedback, the partial closure of the >200 µm deficit, requires it, and the laboratory-constrained baseline is explicitly reported as negligible. As written, the reader cannot tell whether 10× is a defensible upper bound, a central estimate, or the smallest factor that produced a visible signal. I recommend at minimum two additional bin-scheme sensitivity runs (e.g., 3× and 30×) so the reader can see where the response becomes nonlinear and whether the feedback in Section 4.3 survives a weaker multiplier.

The splinter-number uncertainty is treated as a scalar when the underlying data suggest it is distributional. K75 report a mean of 9.4 splinters per drop, but the authors note that 7 of 75 drops produced 50–150 splinters each and dominated the total (lines ~121–125). A deterministic per-drop yield fitted to that mean will systematically misrepresent a strongly skewed population, and a 10× multiplier applied uniformly to all drops is arguably a crude proxy for exactly that skewness. Please discuss this explicitly, and state whether a stochastic or tail-weighted treatment would be more faithful to the laboratory evidence. This also bears on the Section 4.3 feedback, which depends on a small number of large, INP-enriched drops.

2 assumes separability, Nt(D,T) = nPRE(T)·f(D), with nPRE constrained at a single drop size (300 µm). TY77 spans both size and temperature; those data should be used to test whether the temperature dependence is in fact independent of size, and the result of that test reported.

The asymptote argument for T < −30 °C is a property of the quadratic, not of the physics. Eq. 1 has its maximum at T ≈ −31 °C and decreases for colder temperatures, so “an asymptote for temperatures colder than −30 °C” (line ~107) and the suggestion that “linear extrapolation beyond −30 °C may be valid” (lines ~147–148) are not supported by the fit form. Since the authors intend the parameterization to be used by others, please state clearly that the turnover is an artifact of the polynomial, and either truncate at the vertex or specify the recommended behavior beyond it.

The bulk scheme’s large SIP response may be a structural artifact, yet it is the configuration reported as best matching observations. In Section 4.2 (lines ~347–354) the authors speculate that splinters added to the cloud ice category inherit the collision efficiency of the pre-existing, larger crystals, producing a cascading amplification absent in the bin scheme. Section 4.4 (lines ~448–450) then states that the bulk scheme “closes the gap with observations for sizes >200 µm more effectively than the bin scheme.” If the amplification is a moment-integration artifact, that agreement is obtained for the wrong reason and should not be presented as a point in the bulk implementation’s favor. This needs to be resolved rather than speculated on, for example, by carrying splinters in a separate small-ice category, by imposing a mean-size constraint on the cloud ice distribution receiving splinters, or by a process-rate decomposition showing how much of the collisional SIP tendency originates from newly added splinters. At minimum the claim in Section 4.4 must be qualified.

The size ranges for 2D-S and HVPS used are ambiguous (lines 286-287). Please clarify it. The manuscript states (lines ~287–289) that particle-by-particle phase discrimination was not performed, so observed Nice is unavailable, and the comparison proceeds on total PSDs. The headline conclusion concerns a 1–2 order-of-magnitude deficit in concentrations for sizes >200 µm, but at −7.4 and −11.1 °C a substantial fraction of those particles could plausibly be supercooled drops, in which case the deficit is partly a liquid-phase bias rather than an SIP deficit (which the authors partly acknowledge for the 1 °C level). Please bound the ice fraction of the observed >200 µm population using the available information, such as 3V-CPI habit classification, 2D-S roundness or area-ratio thresholds, or Nevzorov LWC/TWC partitioning on the Learjet passes, even approximately. This is the single change that would most strengthen the evaluation.

The primary-ice bottleneck conclusion is asserted but not tested. Section 5.1 argues that low ABIFM freezing rates, not the per-drop splinter yield, are the fundamental limitation, and Appendix C uses the Hallett-Mossop equivalence to reinforce it. But the paper also notes that immersion-freezing parameterizations carry one to two orders of magnitude of uncertainty, and the c and m coefficients used here are derived from ACE-ENA marine boundary layer measurements applied to a western Pacific tropical case, with forg ≈ 0.3 inferred indirectly from bulk κ. A pair of sensitivity runs scaling Jhet (or NINP) by ±1 order of magnitude would show whether the bottleneck conclusion is robust or is itself a consequence of the chosen primary-ice parameterization. Given that this framing carries much of the discussion and conclusions, I regard this experiment as important.

The observational baseline is thin for the strength of the claims. The evaluation rests on six cloud passes with transect lengths of 0.63–6.84 km and 4–25 contributing PSDs (Table 2), and only two temperature levels have more than one pass. Observed variability is therefore essentially unconstrained at 1 °C and −11.1 °C, yet the comparison is used to support order-of-magnitude statements. In addition, all simulated updraft passes are retained although observed passes reached w up to 20 m s⁻¹ while simulated maxima are weaker. Please either stratify the comparison by w or show the simulated w distribution alongside the observed values so the reader can judge how comparable the samples are, and temper the quantitative language where a single pass defines the observation.

The argument against an active Hallett-Mossop process in this case is weaker than the conclusions imply. Section 5.2 rests on a contrast between two cloud passes (CP 2 and CP 3) at similar temperatures but different cloud-top depths. That is suggestive, but it is a sample of two, and Appendix C shows that Hallett-Mossop alone shifts the Nice distribution by 1–2 orders of magnitude, more than the baseline FFD mechanism. I would recommend softening “observational evidence from this case does not support an active Hallett-Mossop process” to a statement about the absence of positive evidence, and bringing the Appendix C result forward into the main text, since a reader may reasonably conclude from it that rime splintering is the more efficient mechanism in this configuration.

Minor comments
Line 6: Please expand CAMP2Ex on first use in the Abstract.

Line 63: “this Part II study extends evaluation of to the ice phase”, missing word.

Lines 95-96: add reference here

Line 96: “one large fragments vs. ten tiny splinters” — should be “one large fragment”.

Line 114: “Two experimental studies meet this criteria” — “this criterion”.

Please state the units of Nt explicitly in Eq. 2 (splinters per freezing drop) and give the diameter range over which the linear size fit is supported by data versus interpolated.

Line 140: “A 3D surface of Nt is also provided in in Fig. 4” — duplicated “in”.

Section 5.1, line ~484. The statement that the KIT formula yields ~1–2 splinters per 300 µm drop between 0 and −15 °C should be reconciled with the comparison to Lachapelle et al. (2025) (5–8 per drop) and Pfeifer et al. (2025) (1–6 per drop) in Section 2, which appears to hold only for millimeter-sized drops. Making that size dependence explicit at both places would avoid an apparent inconsistency.

Simulation naming is inconsistent between Table 1 and the text. Table 1 lists BIN_NOTURB, BIN_NOTURB_SIP, and BIN_NOTURB_SIP_10X; the text uses BIN_SIP_NOTURB and BIN_SIP_10X_NOTURB (lines ~214–215), and Section 4.5 and Fig. 14 refer to BIN_CNTL_NOTURB, which does not appear in Table 1 at all. Please unify throughout and ensure Table 1 lists every simulation shown in the figures.

Section 4.2 versus Conclusions. The text reports maximum Nice in BIN_SIP_10X of ~10 L⁻¹ (Fig. 7c), while the abstract and conclusions state 10–20 L⁻¹. Please make these consistent and state which diagnostic (domain maximum versus mass-weighted domain mean) each number refers to.

Figures 2, 3, and 4 overlap substantially in content. Figure 4 is largely a restatement of Figure 2 and of Figure 3b. Consider merging Figures 2 and 4 and moving one panel to an appendix.

Section 4.3, lines ~397–401. It should be feasible to do 3D process-rate diagnostics in this study.

Figure 7 caption: “(d)-(e)” should read “(d)-(f)”.

Figure 8 caption: the panel letters are mislabelled — “(a,b) bin-scheme … and (b,d) bulk-scheme … for (a,b) cloudy grid points only and (c,d) cloudy updraft grid points” is internally inconsistent.

Figure 11 caption: “ql > 10⁻³ g kg⁻³” should be “g kg⁻¹”.
Citation: https://doi.org/10.5194/egusphere-2026-4713-RC2

McKenna Stanford, Alice Keinert, Ann Fridlind, Alexei Kiselev, Daniel Knopf, Andrew Ackerman, Thomas Leisner, and Paul Lawson

Viewed

Total article views: 390 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
193	109	88	390	90	75

HTML: 193
PDF: 109
XML: 88
Total: 390
BibTeX: 90
EndNote: 75

Views and downloads (calculated since 21 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	60	28	17	105
Sep 2026	123	78	69	270
Oct 2026	10	3	2	15

Cumulative views and downloads (calculated since 21 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	60	28	17	105
Sep 2026	123	78	69	270
Oct 2026	10	3	2	15

Viewed (geographical distribution)

Total article views: 379 (including HTML, PDF, and XML) Thereof 379 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 03 Oct 2026

Short summary

Clouds containing a mixture of liquid and ice are poorly simulated because ice crystals multiply via poorly understood mechanisms. This study uses laboratory experiments to represent ice multiplication from freezing drops and evaluates that representation in high-resolution simulations against aircraft observations. Ice fragment production increases ice crystal numbers but is modulated by the evolution of raindrops prior to freezing, with direct implications for weather and climate models.


Total:	0
HTML:	0
PDF:	0
XML:	0