The manuscript presents a study of the heat wave characteristics in Germany by using several RCM models and a WRF-based high-resolution downscaling dataset.

The topic is fascinating and fits the journal's aim and scope. Moreover, due to the last-year droughts, a general audience can be interested in such type of study. The manuscript is well-structured and organised. The English is fluent and understandable. Before considering the work for publication, I have the following points to raise with the authors:

1.      I understand the purpose of only comparing RCMs. Still, it might be interesting to exploit the newest version of ECMWF Reanalysis ERA5 as boundary conditions, increasing the spatial resolution to 0.25°. It means waiting until the new EURO-CORDEX simulation is performed, but at least WRF can be run in future studies comparing the previous and latest versions. In general, I would appreciate some discussion about that since one of the conclusions is that the RCM does not significantly improve performance.

2.       Is downscaling with WRF performed with two nested domains due to coarse ERA-Interim resolution (0.50° -> 15km -> 5km)? It should be explained in the data section; I didn't find the reference paper (Wagner and Kunstmann 2016).

3.      Figure 2. I need help seeing all the distribution, especially the reference E-OBS, which is completely missed. I suggest rethinking this.

4. Typos: L530 "are better"; L572 missing comma "(Petrovic et al., 2022), in which"; L611 "an average intensity"

Petrovic et al. present an evaluation study of a set of reanalysis-driven regional climate model simulations with respect to the representation of heat wave characteristics over Germany. The authors employ simulated (EURO-CORDEX evaluation runs and two special WRF experiments) and observed (EOBS) daily maximum temperature in the period 1980-2099 and evaluate the model performance for several heat wave statistics (mean max. temperature, number of heat waves, heat wave duration, heat wave intensity). They find no clear benefit of the increase resolution in one of the WRF experiments and, in general, a mixed picture with no clear best or worse performing model. All models tend to underestimate the observed increase in the spatial extent of heat waves.

The manuscript is nicely written and contains a very good literature review and very appropriately outline the motivation for their study. The authors put great effort into documenting the model performances. However, I do not think that the manuscript is suited for publication in the journal as its novelty and its significance for a wider audience is very limited. The paper has the flavour of a rather technical documentation of model performance and remains very descriptive. Reasons for a good or bad reproduction of heat wave characteristics (large scale flow conditions, surface atmosphere interactions etc.) are discussed but actually not investigated. The employed model ensemble is not special or up-to-date (though not yet outdated) and concerning the EURO-CORDEX sub-ensemble has been validated extensively in previous works. Most recent experiments, for instance, employ the newer ERA5 reanalysis as boundary forcing. The resolution analysis is restricted to the WRF model and hence very limited, and the evaluation is carried out at the coarser 12 km resolution (i.e., the 5 km experiment cannot profit from its potentially finer-scale heat wave patterns in the evaluation). Moreover, most evaluated metrics actually neglect the mean bias (since evaluation is carried out with respect to the simulated 90^th percentile or even as a normalized bias in the Taylor diagrams). That is why I believe that there is not much to learn from the paper. There are also number of minor issues listed below. My suggestion is hence to reject the submission for the time being, but to encourage the authors to produce a more up-to-date and process-based evaluation of heat wave characteristics instead. The topic is certainly highly relevant.

With kind regards

Minor issues:

- The issue of internal variability in different model domain sizes is not discussed at all (relating especially to the 2003 event, i.e. to a singular event). The domains of the WRF experiments are unclear and not shown.

- WRF@5km has been interpolated to the EUR-11 grid, i.e. the resolution has been degraded and any potential benefit in a better representation of finer-scale spatial heat wave patterns is not visible. Also, bilinear interpolation should be avoided when interpolating from a finer to a coarser grid (only a fraction of the high resolution grid cells is employed in this kind of interpolation). Better use a conservative remapping.

- Taylor diagram: You have to consider that the RMSE shown by the green lines is the CENTERED RMSE, i.e. any mean bias is implicitly corrected for. This should at least be mentioned, or even shown for instance by different sizes of the symbols (according to their mean bias). Also compare to Table 1 where the mean bias is shown.

- Figure 2: The shadings are hardly distinguishable from each other. It would be much better to show a simple curve for each model instead of a shaded curve.

- Chapter 4.2 and further: You often speak of a “cold” or “warm” bias here. I would strongly recommend to avoid this, as all statistics are related to the 90^th percentile of each individual simulation. Any mean bias of the 90^th percentile is implicitly subtracted and what you’re looking at is actually the variability (temporal and in terms of magnitude) of all values above the simulated (or observed) 90^th percentile.

- I suggest to speak of “WRF simulations” or “WRF experiments” instead of “WRF domains”.

- Analysis of heat wave trends: You should definitely include a validation of Tmax (and its 90^th percentile) in order to explain deficiencies in simulating heat wave trends.