General Comments:

Overall I find the manuscript to be acceptable, although a few minor suggested revisions are mentioned below. It is clearly written and well organized with supporting evidence and logic and easy-to-follow outcomes.  The main criticism (further discussed below) pertains to the persistent problem in some models of achieving the right outcome for the right reason.

Specific points:

1. The discussion in Sections 2.2 and 2.3 could be made part of the appendix directly instead of including in the main body.  The cloud scheme is already discussed in detail in the appendix so isn't it simple to keep all those details in one place?  The fact that multiple closure methods were attempted could be omitted and only the one picked could be described.  The rejected method seems impertinent to readers.  During the research, the authors discovered a closure idea that was inferior but that happens frequently in model parameterization development.  Which dead-end pathways to describe to readers is subjective, but it doesn't seem to add any insight directly to a physical problem being solved.  As one manuscript reviewer's opinion only, I would not require this to be addressed in a revision, so the editor can decide if there is mutual agreement among reviewers.

2. While I agree that the scale of model data versus observations is extremely different, I believe it is insightful to see a distribution of the fundamental raw model data error. A good example is found in Fig. 5 of Thompson et al (2024).  The frequency histograms of RHice in this manuscript's Fig. 6 provides a good indication of the changes in ARP-new vs. IFS and Obs, but a distribution plot of direct model error for every single IAGOS unfiltered observation is desired as well.

3.  Why are various models still not using a better physical representation of ice depositional growth from a physical means rather than using variants of saturation adjustment?  Efforts to create and use tuning knobs to handle ice supersaturation rather than updating inherent physical growth equations seems endless.  Eq. 1 is just another tuning knob component of three elements described in this paper:  (1) a calibration coefficient; (2) a simplistic temperature function; and (3) a closure method that doesn’t properly represent the physics as shown clearly in Fig. 5.  The “cliff” in the histogram is related to Eq. 1 and the sentence in Line 112: “Once the supersaturation threshold is locally exceeded, local adjustment is instantly obtained back to saturation.”  In other words, as humidity grows progressively larger, it will cross the threshold and then suddenly the RHice is instantly dropped (let’s say for example 145%) back to 100% while adding the excess vapor directly into solid phase.

There is no need to invoke a need for 2-moment cloud ice treatment to result in proper RHice forecasts.  This appears to be a common misconception.  A mass mixing ratio single moment scheme suffices with additional assumptions of ice spectral distribution.  A basic inverse exponential distribution with a Y-intercept parameter that can increase as ice mass increases while holding a slope constant is one such assumption.  This follows the most basic observations that more ice number comes with more ice mass.  From whatever assumptions are made for number distribution, the total ice number (or number within bins of specific size ranges) can be diagnostically calculated, which effectively turns a 1-moment scheme into a 2-moment treatment.  There is no solid evidence to say that 1-moment schemes are incapable of predicting the correct outcome compared to 2-moment schemes.

The essential problem of the microphysics is the lack of accounting for slow physical vapor depositional growth of ice.  Creating a new threshold for when to convert instantly the excess vapor over ice saturation into cloud ice isn’t solving the problem yet (as Fig. 5 clearly shows).  In fact, the method to create initial ice where none previously existed could be fine with the new technique, but once ice does exist in a grid volume, do not permit more ice to nucleate and use a “electrical capacitance” analogy to grow the existing ice by vapor deposition.  That way some of the excess (over saturation) water vapor can remain in gas phase and continue to permit RHice>100%.

Technical corrections:

I did not exhaustively list many technical corrections because the manuscript was relatively good overall and I am late submitting the review so I am optimistic that other reviewers made more suggestions.  Here are just a couple items.

L59: “verification methods deserve to be completed to accurately…” is awkward.  It is simpler to state that verifying RHice in general is needed as well as threshold-based (ISSR) conditions?

L60: “known to be a rather rare phenomenon in the atmosphere.”  It is not rare.  It occurs ~11% of the time in the entire atmosphere if you believe radiosonde data per Thompson et al (2024) or the manuscript's quote of 10% of the time from the IAGOS dataset.  That does not seem especially rare.  The phrase is basically repeated in L246.

Review of the Manuscript: "Modeling and verifying ice supersaturated regions in the ARPEGE model for persistent contrail forecast"

The manuscript presents a modification to the ARPEGE NWP cloud scheme to allow for the representation of ice supersaturated regions (ISSRs), with evaluation against aircraft in situ humidity data and comparisons to the Integrated Forecast System (IFS). The topic is timely and highly relevant to aviation climate impact mitigation strategies. The paper is generally well-motivated and provides a solid contribution toward improved ISSR representation. However, there are several issues that need to be further clarified or elaborated before publication.

Major Comments

1. While the study introduces a practical modification to ARPEGE’s cloud scheme, the extent of its novelty compared to earlier approaches (e.g., in IFS or other global circulation models) is not entirely clear. The paper would benefit from a more explicit discussion of how the proposed approach differs from existing contrail cirrus cloud parameterizations used in current GCMs (e.g., ECHAM, CESM).
2. The evaluation is primarily based on IAGOS aircraft measurements. While these are high-quality and relevant, the representativeness of IAGOS data (limited flight routes, sampling biases) should be more explicitly discussed. Could the conclusions change in regions less well sampled by IAGOS? The authors mention radiosondes as a possible next step, therefore an expanded discussion of current limitations in observational coverage would strengthen the evaluation framework. The authors should provide more detail on the dataset and discuss potential sampling biases.
3. The authors note that both ARPEGE and IFS underrepresent the highest values of RHice (>110–115%). This is an important limitation, since persistent contrails are most strongly associated with highly supersaturated regions. Could the authors elaborate on the physical or numerical reasons why models fail at these extremes? Addressing this could help guide future model improvements.
4. The modification to the cloud scheme is presented as computationally inexpensive and algorithmically simple, but the paper does not sufficiently explore whether introducing supersaturation impacts other parts of the model system. For example, how does it affect cloud microphysics, radiative transfer, or dynamics? Even if these effects are minor, they should be explicitly discussed.

Special comments:

1. L102–103: What is the value of Ccalib in this study? How do you get this value? Is Ccalib expected to change with variations in time and geographic location?
2. L147–150: To what extent is the cloud fraction sensitive to the choice of probability distribution? What alternative approaches exist for representing probability distributions beyond the normalized centered probability distribution? What rationale did the authors provide for selecting the normalized centered probability distribution over other methods?
3. L184–185: This study excludes the United States and Asia. How could that influence the results?
4. L186: Please provide an explanation of FL250 and FL450.
5. L215–218: Will the smoothing and undersampling alter the properties represented in the original IAGOS dataset?
6. Section 6.1: Providing results that show how ARP-new alters the cloud properties compared to ARP-op would be very helpful.
7. L562: Why did the authors choose to emphasize results at a neighborhood tolerance of 150 km?