Please refer to the attached pdf file.

The paper presents a report on numerical experiments simulating a bow echo that impacted Poland in 2007. It shows that introducing small-scale temperature perturbations into the ERA5 analysis enables the COSMO model to simulate the bow echo correctly. The study is interesting, as the authors successfully simulate the bow echo and discuss the roles of cold pools and vertical wind shear. However, the paper is excessively long. Many sections contain unnecessary details that can be omitted, while other parts require more thorough discussions. I recommend a major revision.

Major comments

• Section 2, which describes the bow echo, could be shortened.
• Section 4, which "discusses a reconstruction of the initial conditions for prognostic experiments", could be merged with Section 3. This is because the changes in soil and surface conditions are kept in the subsequent experiments with CI. Furthermore, the results from nudging of soil and surface observations, as well as the modifications of surface heat fluxes, are not summarized in either the conclusion or the abstract. Therefore, the associated figures can be omitted while a concise description of the soil and surface changes, along with a brief summary, could be retained.
• Section 5, which "discusses the results of the experiment without the CI scheme", could be removed. The main relevance of this experiment lies in its failure to reproduce the bow echo.  While this provides justification for doing experiments with the CI scheme, the key message could be conveyed in a single sentence. Given that the paper focuses on the CI scheme, it is unclear why so much detail is dedicated to an experiment that does not use it.
• Section 7.2 on "Impact of shear modification on environmental conditions and deep convection without CI", raises a similar concern. Why is emphasis placed on experiments that do not involve the CI scheme?
• Section 8 on "Rear inflow jet in the EM0 forecast", includes two figures and half a page of text. This section should either be expanded with a more thorough discussion or removed altogether.

Minor comments

• Line 26, bow echoes are indeed a specific class of deep moist convection organization. As written Line 60, they are "developing under a significant external forcing". With this respect, how bow echoes can be self organized?
• Line 17, MCAPE and MCIN are usually named CAPE and CIN.
• Lines 30 and 52, "Europe [...] and Poland" This wording suggests that Poland is not part of Europe.
• Line 67, Uncertainties in the representation of turbulence also affect cloud organization (Machado and Chaboureau 2015, Tompkins and Semie 2017)
• Lines 377-380. In addition to the experiment with the use of the shallow convection parameterization, experiments to the turbulence parameterization could also have been conducted.
• Lines 396, 398, 406, 409 and 410. The CI scheme uses a few number of parameters. How their values are specific to the case study? Could the CI scheme be used for other case studies, in operational way?
• Line 422, "MCIN values locally diminish below 10 J kg-1". A value of 10 J kg-1 corresponds to a vertical velocity of 4.5 m s-1 that an air parcel should have to overcome the CIN barrier. It would be more meaningfull to examine a lower threshold of CIN, e.e.g 1 J kg-1 corresponding to a vertical velocity of 1.4 m s-1

References

• Machado, L. A. T., and J.-P. Chaboureau, 2015: Effect of turbulence parameterization on assessment of cloud organization. Mon. Weather Rev., 143, 3246–3262, http://dx.doi.org/10.1175/MWR-D-14-00393.1
• Tompkins, A. M., and A. G. Semie, 2017: Organization of tropical convection in low vertical wind shears: Role of updraft entrainment. J. Adv. Model. Earth Syst., 9, 1046–1068, https://doi.org/10.1002/2016MS000802

This paper addresses a case study of a rapidly developing bow echo over northeastern Poland on 21 August 2007. The issue addressed is the tendency of convection-permitting models (CPMs) to that represent convection with relatively poor resolution to suppress the initiation of convection.

The case study is a good example of a major, high-impact, bow echo over Poland. This was chosen because of its high impact, but, because it occurred in 2007, it leads to a numerical experiment which is relatively unrealistic in the context of contemporary convection-permitting NWP systems. In particular, the hind-cast is spun up from ERA 5 reanalysis. This is a good choice for 2007, but the analysis is relatively low-resolution and will contain none of smaller scale variability that a realistic continuously running assimilation cycle with a CPM would have.

This deficiency is addressed to a small extent by an assimilation cycle using surface observations; however, as one might expect, this proves only partially effective and experiments are performed essentially by adding random noise at the grid-scale to establish whether convection forms and organises.

As a numerical experiment, this paper is interesting and worth publishing. However, it’s physical motivation is poorly justified and not well explained. They claim that they have designed a CI scheme and “Its idea is to use near-grid-scale temperature perturbations, possibly resembling those physically developing in CBL, and allow the model to explicitly represent further upscale growth of the perturbations.”

They recognise that such schemes already exist but theirs has very different properties that are not explained. In particular, their scheme is not just a stochastic perturbation scheme!

They use “temperature perturbations with a realistic amplitude of about 1.0-1.5 °C” – indeed ”The amplitude ΔT of the temperature perturbation is drawn from the Gaussian distribution with a mean of 1.25 °C and standard deviation of 0.5 °C”. Thus it is biassed! They are essentially correcting a bias of 1 degree before even adding random noise. This is not just a stochastic perturbation scheme but also a bias correction scheme. Sub-filter perturbations must have zero mean!

Furthermore the amplitude of their perturbations is at least an order of magnitude larger  (or even two) than published schemes. This is not justified – as explained by Kober and Craig  (2016) and Clark et al (2021), perturbations “physically developing in CBL” must scale on the convective temperature scale, which is typically O(0.1 °C) – this is the typical perturbation of a thermal. When filtered to a larger scale, the amplitude will be even smaller. Clark et al (2021) show that, in deep convective situations, the continuous application of realistic CBL perturbations can lead to variability on the near grid scale that is indeed O(0.5 °C) after 12-36 h of upscale growth, largely through moist convection.

Thus, the authors are modelling the variability that might exist in the initial state, but then go on to change them every 10 minutes (eddy turnover time). There is no physical justification for this. Others (some cited, many not) have used perturbations of this magnitude to study predictability, adding such noise to the initial state or at some critical time, but to refresh them so frequently is unphysical – either continuously refresh the perturbations with a physically realistic amplitude and let them grow for an appropriate period, or simply start with appropriate variability. Doing both is unphysical.

Scientific significance:

The results are of interest, though some major revision is required to put the perturbations used into context. I believe they are unphysical and also do more than randomly perturb the state. The fact that the pertubations are combined with a bias correction seems a major failing. The CI scheme could not be scientifically justified for use in a modern NWP system; the scientific question addressed is really the slightly more mundane 'a low-resolution CPM does suppresses initiation - how hard do we have to kick it to get the result we want'? The answer is in line with other studies, but tells us nothing about where that variability comes from or how, therefore, it should be represented.

Scientific quality:

Most of the paper is scientifically well executed and well argued, but the basis of the CI scheme needs to be revised, claims that it is ‘realistic’ justified, and the role of bias correction explained.

Presentation quality:

On the whole the quality is acceptable, but I found navigating the maps difficult, because what landmarks there are (mainly country borders) are very hard to see,

Minor comments

P2

39: Some more comment should be made on the deficiencies RKW. It is certainly not a complete theory.

50: Have Bow echoes only been successfully simulated over US and Poland?

64: At most 0.25 km grid spacing, not at least.

P3

84: ‘the scheme’ – PSP is not one scheme.

86 ‘Random representations of CI’ not yet explained.

103 Novelty of using ERA 5 with DA. Explain the motivation more, and discuss the scales of variability addressed by the DA.

110 Explain the meaning of  ‘realistic environmental conditions’.

P5

Fig 1: Hard to see maps, especially as a-c have different domains compared with c-f.

P10

258: Single moment microphysics. How important is this to organisation?

P11

No mention of cloud scheme. Probably Sommeria-Deardorff, which may slow initiation.

P12

280: “the realistic 280 evolutions of 2-m T and Td good indicators of realistic temperature and humidity profiles across the CBL”. Some aspects, perhaps, but probably not a good indicator of cloud-base conditions. Level 2.5 MY schemes have issues with entrainment.

P13

299: How are soil temperature measurements corrected for surface altitude in the Cressman analysis?

What about soil moisture errors?