The manuscript describes the application of AR-ChaMo, being an additive logistic regression model, to forecast large hail and lightning over Europe. The additive logistic regression model predicts the probability of lightning (hence also of a thunderstorm) and the probability of the occurrence of large hail given a thunderstorm forms. The observational data stem from lightning observations and hail reports from the ESWD. The authors show the performance of their model using a case study and also consider the AUROC-score as a verification metric. They also compare their model to two other classical indices and find their model to give equal or better guidance.

The manuscript touches on a very important topic and I find the use of the additive logistic models to predict lightning and hail very interesting. However, I think that the presentation should be improved and some major aspects should be clarified before the manuscript can be accepted.

General comments:

-----------------

- Lines 14-15: Where do you show the model's ability to reproduce the climatological distribution and the seasonal cycle? 

- Section 2.1: I think the model should be described in more detail in this section. Although you refer to another article that should contain all the details, I advocate to repeat the main aspects here to provide the reader a self-contained article. In addition, I have a some further remarks/questions

(i) Training of the hail-model is based on the hail reports from the ESWD and obviously not all hail occurrences are reported there hence your trainingdata contains many cases where hail occurred but no report was issued. Does this pose a severe problem for the training? How do you mitigate the effect of these non-reported but true hail events?

(ii) If a hail report is within the ESWD, how do you connect the report to the corresponding gridpoints (in space and time)?

(iii) Judging from the figures in other sections, you apply AR-ChaMo per ensemble member. This should be stated explicitely. 

(iv) Given that you also refer to a "Ensemble mean probabilistic forecast" (e.g. captions of figures 3 and 4 and also line 51 or line 170), how do you compute this mean forecast? Is it the mean of the forecasted probabilities of AR-ChaMo on the individual ensemble members?

(v) What exactly would your forecast-product look like, i.e. what should a forecaster see? Do you want to forecast a probability per member or a single probability that is a combined/mean probability derived from all ensemble members? If it is the latter, I suggest to add a figure of that combined probability.

- Section 2.2: Does the ECMWF model change from 46r1 to 47r1 (as mentioned in line 71) have an impact on the training of AR-ChaMo?

- Section 2.3: How many hail reports are within your trainingset?

- Section 3: 

(i) Why do you adapt the model based on ERA5 with this complicated process and not retrain it on the reforecast-data? 

(ii) Where do all these adaptation-formulas come from, i.e. how are they derived?

(iii) Please clearly state the quantities that your lightning model and your hail model try to forecast, e.g.: The probability of occurrence of at least two lightnings within 3 hours and an area of 0.2°x0.2°.

- Line 96: You state that the probabilities are calibrated. Please add some reliability plots. 

- Line 108: Did you also upscale the observations?

- Line 123: How did you upscale the probabilities?

- Lines 166-167: To be more precise, you showed that for this case the forecast was skilful up to 108h in advance but a single case study cannot support such a conclusion in general.

- Section 5: Please add a few sentences that clarify the quality aspect that is covered by the AUROC-score.

- Line 172: In section 2.3 you define a lightning case as a one-hour period. Now you consider a three-hour period. What is the final product? Mixing of both products is misleading.

- Line 183: Are the 1-dimensional logistic regression models trained on the same trainingset (where the appropriate quantities are computed) as AR-ChaMo?

- Appendix B: Maybe I missed it but I think there is no reference to appendix B suggesting that this appendix is not needed.

- Appendic C: The caption of the table says that this is the list of parameters that enter AR-ChaMo based on ERA5. I suggest to extend that table to also give a list of parameters that enter the AR-ChaMo model used in present study.

- Figures 1, 4, 5: I suggest to add a panel that contains the guidance from the other composite indices for comparison.

Technical comments:

-------------------

- Lines 66, 89: I think it should read "Table C1"?

- Line 88: It should read "Battaglioli et al. (2023)".

- Lines 92, 94: There is no Table 2.

- Line 129: It should read "...France. This...".

- Line 138, 139: It should read: "...probabilities for lightning...".

- Line 191: I think it should read "Section 4.3"?

- Lines 194, 197: Prior to this section you used the British version "skilful" instead of "skillful". 

- Line 194: I think it should read "based on the period" instead of "for the period".

1 Overview

In this manuscript, the authors present the performance of an additive logistic regression model called “AR-ChaMo” that is able to predict the probability of lightning and hail occurrence. AR-ChaMO model for hail is compared with two documented convective indices and results show that AR-ChaMo outperforms them.

While the manuscript is well-written and the presented results are very interesting, and of great importance, I would suggest some major revisions to be made before the manuscript is accepted for publication.

2 Major comments

• The study is based on the additive logistic models developed by Rädler et al. (2018) but they were based on the much coarser ERA-Interim reanalysis dataset and in the current manuscript, in Section 3, it mentioned that ERA5 was used to train the models with additional parameters. While in Rädler et al. (2018) it reads in their Section 6 “This model may be improved further by using additional predictor parameters. Such parameters may include low-level moisture, lapse rates, lifted condensation level, or height of the melting level.”, it seems that the authors of the current manuscript have taken into account this suggestion but they do not present their calculations. In addition, Battaglioli et al. (2023) study has not been published yet, and several aspects of the approached used are not currently accessible. The table with some acronyms of the parameters used is not sufficient in order to publish a method and make it reproductible by the scientific community.
• While the study is dealing with both hail and lightning forecasting, the latter is not discussed at all in the Introduction. Please expand this section and use more references for lightning forecasting. Ideally, other approaches for lightning forecasting should be discussed and compared, similar to hail.
• The Significant Hail Parameter (SHP) is able to distinguish environments with very large hail (>=5 cm) and with small hail. How did you compare your model’s result with SHP given that you used ESWD data with hail >=2 cm? Please consider adding a few more graphs either in main text or as an appendix to visually compare examples for large hail forecasts with SHP, CAPESHEAR, and AR-ChaMo.

3 Minor comments

1. Acronyms are used without explicitly defining them, even in the Abstract (i.e., ECMWF, CAPE, etc).
2. Wet Bulb Zero Height is mentioned in the Abstract but it is not mentioned anywhere else. Was it used?
3. In Line 66 Table 1 should be Table 1c.
4. Rädler et al. (2019) should be Rädler et al. (2018).
5. Lighting should be lightning in Lines 80 and 171.
6. I am not totally sure, but the time convention is 00z or 0000 UTC or 00:00 UTC, and not 00 UTC as it is used now. Please double check and modify accordingly.
7. In Line 135: J/kg – J kg^-1

References:

1. Rädler, A. T., P. Groenemeijer, E. Faust, and R. Sausen, 2018: Detecting Severe Weather Trends Using an Additive Regressive Convective Hazard Model (AR-CHaMo). J. Appl. Meteor. Climatol., 57, 569–587, https://doi.org/10.1175/JAMC-D-17-0132.1
2. Battaglioli, F., Groenemeijer, P., Púčik, T., Taszarek, M., Ulbrich, U., and H. Rust, 2023: Logistic modelling of (very) large 315 hail occurrence in Europe and the United States (1950-2021)