| 09 Sep 2026
Role of INP activation schemes and concentrations in controlling ice formation pathways in Warm Conveyor Belts
Abstract. Warm conveyor belts (WCBs) in extratropical cyclones transport moist boundary-layer air into the upper troposphere and form extensive mixed-phase cloud bands and cirrus cloud shields. The microphysical pathways governing ice formation within these systems remain poorly quantified, particularly with respect to heterogeneous ice nucleation and secondary ice production. We investigate the impact of uncertainty in the ice formation representation on clouds in a WCB using the ICON model with an ice modes two-moment scheme that explicitly tracks ice formed by immersion freezing (IMM), deposition nucleation (DEP), homogeneous droplet freezing (FRZ), homogeneous solution freezing (HOM), and secondary ice (SEC). We test the sensitivity of ice formation pathways to three dust INP activation schemes, perturbations of INP concentration, and three secondary ice mechanisms (rime splintering, collisional breakup, droplet fragmentation).
Ice number concentrations in the cirrus regime respond strongly to both INP abundance and activation scheme. Enhanced INP concentrations strengthen heterogeneous modes, particularly DEP, leading to DEP-dominated cirrus and suppression of homogeneous nucleation. Reduced INP concentrations promote compensating increases in HOM and FRZ. Secondary ice from rime splintering and collisional breakup dominates number concentrations in low level mixed-phase clouds but is negligible in cirrus. Importantly, altering the INP activation scheme produces changes comparable in magnitude to variations in INP concentration with scaling coefficients between 0.1 and 15, highlighting substantial uncertainty in heterogeneous ice nucleation assumptions. Overall, the partitioning among ice modes is highly sensitive to microphysical assumptions, but the balance between liquid-origin and in-situ cirrus remains primarily controlled by WCB dynamics and sedimentation.
Review of "Role of INP activation schemes and concentrations in controlling ice formation pathways in Warm Conveyor Belts"
by Lüttmer et al.
Manuscript summary
The authors use ICON in limited-area mode at 6.5 km, together with their ice-mode two-moment scheme, to study how ice forms in a warm conveyor belt (WCB) during a NAWDEX cyclone in October 2016. Because the scheme tracks cloud ice by formation pathway (IMM, DEP, FRZ, HOM, SEC), they can follow how three dust INP activation schemes (HA, UL, PH), scaled INP concentrations (HA+, HA−) and three secondary ice mechanisms (RS, BR, DF) affect the ice along LAGRANTO WCB trajectories.
The key conclusions are
I enjoyed reading this work. The ice-modes framework is a neat way of following where ice comes from, and the question fits ACP well. That said, some parts of the model setup that matter for interpreting the results are not described, and the manuscript has quite a few errors in figures, cross-references and text. I recommend minor revisions.
Main comments
Aerosol formulation and the link to INPs
I could not find a proper description of how dust, and hence INPs, is represented. Are the Hande et al. (2015) fields climatological profiles, do they vary horizontally, and which season is used? Since they were derived for Europe, are they representative of the North Atlantic inflow of this WCB? PH uses altitude as its vertical coordinate (l. 208–209), so its dust distribution also differs from the other two schemes.
I was also surprised by the relaxation of n_iact on a 10-min time scale (l. 225–227). This replenishes INPs soon after a cloud evaporates and, as far as I can tell, does not account for scavenging. It could favour deposition nucleation in the outflow.
The size of the INP perturbation is also described inconsistently. Line 213 and Fig. 1a suggest roughly one order of magnitude in each direction. Lines 396 and 512 refer to two orders, l. 393 and 447 to "several", and the abstract to scaling coefficients of 0.1–15.
Also, the claim in l. 50–52 that comprehensive aerosol models are too expensive for sensitivity studies is too strong. ICON-A-HAM (Salzmann et al., 2022) and EMAC (e.g., Bacer et al., 2018) are regularly used for ice nucleation sensitivity experiments, albeit at coarser resolution. They would also be good examples to cite at l. 57–59.
WBF process and saturation adjustment
Saturation adjustment is never mentioned, and the WBF process is mentioned only in passing (l. 457), yet both shape the mixed-phase results. If saturation adjustment is used for cloud liquid, as in the standard ICON configuration, mixed-phase cloud sits at water saturation, where DEP and HOM are switched off (l. 327–329). The finding that DEP and HOM ice in mixed-phase cloud comes from sedimentation is therefore partly built into the model. How fast that sedimenting ice grows at the expense of liquid, and hence the fractions in Fig. 3b, also depends on how deposition is formulated when liquid is present.
The WBF process only operates under certain conditions (Korolev, 2007). Its treatment can matter as much as, or more than, the nucleation scheme for phase partitioning (e.g., Tan and Storelvmo, 2016; Vella et al., preprint). Saturation adjustment is also known to affect simulated aerosol–cloud responses (Lebo et al., 2012). Consider discussing this work against your results.
Convection, vertical velocity and time step
At 6.5 km, convection is in the grey zone, but the list of parameterisations (l. 90–93) does not say whether a convection scheme is used. If it is, how is detrained ice assigned to the ice modes? If not, how is the embedded convection known to occur in WCBs (Oertel et al., 2020) represented?
I am also concerned about the vertical velocity used for homogeneous nucleation. Kärcher and Lohmann (2002) depend strongly on updraft speed, and at this resolution the resolved w misses much of the small-scale variability that drives cirrus formation. This could bias the HOM–DEP competition, which is central to the paper, towards DEP.
Ice origin is only tracked in the cloud-ice category
Since aggregation feeds snow (l. 147) and riming feeds graupel, the origin information is lost once ice leaves the cloud-ice category. That is a large share of the frozen water. Snow accounts for 33–43 % of the frozen mass in cirrus (Fig. 3d), and snow and graupel together for almost 90 % in mixed-phase cloud (Fig. 3a). The unattributed part also differs between experiments, as HA+ moves more mass into snow (l. 366–369), which could bias the comparison. Please state this limitation clearly, and discuss or, ideally, quantify its effect on the partitioning.
Robustness and interpretation
Everything rests on one case and one trajectory start time, and some of the reported differences are small. Examples are the median n_tot values of 0.008, 0.009 and 0.01 cm⁻³ (l. 266–269) and fraction changes of 0.03–0.07. Without some estimate of variability, it is hard to know which of these are meaningful. The similarity of the WCB dynamics across experiments (l. 258) is key to the argument but is "not shown". It is also unclear whether trajectories are recomputed for each run.
I would also be careful with the statement that changing the scheme "while keeping INP concentrations comparable" (l. 515) matters as much as changing INP abundance. UL and PH were tuned to HA only for immersion freezing (l. 210–212), and PH deposition is about an order of magnitude weaker (l. 221). PH therefore partly acts as an INP reduction for DEP.
Lastly, the argument that fewer crystals in UL and PH still deplete the vapour (l. 300–302) would be much stronger with RH_i statistics. These are arguably the most relevant outflow quantities for the UTLS (Schwenk et al., 2025).
Minor comments
Technical corrections
The manuscript contains several typos and editorial errors, and I would ask the authors to proofread it carefully. Some examples are listed below.
References suggested