General comments

This study investigates how well ERA5 model-level wind speeds at different geometric heights represent 80-m mast observations in complex terrain and coastal/flat sites. The work addresses an important practical question for wind-energy applications: at which ERA5 heights are hub-height conditions best represented, and how does this depend on terrain complexity.

Overall, the paper poses a clear scientific question. The scientific approach is sound and the results are clearly presented.  I therefore recommend minor revision, mainly to clarify some key statements, correct small inconsistencies in figures/tables, and explicitly summarize the main uncertainties and limitations in the Conclusions. The manuscript addresses an important and practical question for wind-energy applications and provides useful guidance on the representativeness of ERA5 model-level winds in different terrain types.

Minor comments

1. In Figure 2, I suggest adding the site elevation in parentheses next to each mast name in the legend (e.g. “M1 (3796 m)”, “M2 (3690 m)”), for easier cross-reference with the map.
   
   In addition, it would be very helpful to explicitly include the correlations between the mast observations and the ERA5 10-m and 100-m single-level winds for each site (e.g. additional markers or lines). This would make the comparison between single-level and model-level performance more transparent to the reader.
2. In Figure A3, I only see three coastal sites, while the caption refers to “four Coast sites”. Please check and correct the caption and/or the figure.
   
   Similar to Figure 2, I recommend also indicating the correlations between the mast observations and the ERA5 10-m and 100-m single-level winds for each coastal site, to facilitate comparison.
3. Lines 303–305: I suggest explicitly emphasizing in the Conclusions that, in complex terrain, better correlations are obtained at well-exposed locations (e.g. ridge-top sites) than at sites strongly affected by local flow behaviour (e.g. sheltered hillslopes or valleys).
   
   In addition, the important findings stated in Lines 444–445 (“higher underestimations would be expected for sites with high frequencies of low wind speed”) and in Lines 289–292 (“wind speeds over flat terrain are representative of hub-height conditions at coastal and inner flat sites but are not representative in mountainous areas”) could also be briefly restated in the Conclusions section so that the main messages are more clearly visible to the reader.
4. In the caption of Table 4 it is stated that “The best-performing models … are highlighted in bold”, but no boldface appears in the current version of the table. Please either apply the bold formatting or adjust the caption accordingly.
5. Lines 369–371: M1 shows the smallest discrepancies, whereas M2 exhibits the largest. It would be useful if the authors could briefly discuss the possible reasons for this difference (e.g. differences in exposure, local flow features, representativeness of the ERA5 grid point, or data quality), as this may help readers understand how site characteristics affect model performance.
6. In the Conclusions section, I recommend adding a short paragraph summarizing the main sources of uncertainty and limitations of the study. For example, this could include:
   
   – the limited number of masts and regions considered;
   
   – the representativeness of individual masts for complex terrain;
   
   – the dependence on the chosen ERA5 grid point and the computation of geometric heights from model levels;
   
   – and the relatively short analysis period (2021–2024).
   
   Such a paragraph would help frame how far the results can be generalized to other complex-terrain regions.

The manuscript sets out to answer a clear scientific question, can better hub-height wind speed predictions be made in complex terrain by using other ERA5 model heights than the standard 10 m and 100 m heights, and answers it in a clear and straightforward fashion. Given the simplicity of the question, though, I am not sure of the scientific significance of the work. It seems rather obvious that you can improve wind speed estimates by searching for and selecting the ERA5 height that correlates most with the quantity you want to predict. I would recommend considering the following additions to enhance the impact of the work.

First, I would be interested to see typical model errors expressed for different hub-height wind speeds and directions. Do patterns emerge? For instance, does the error change dramatically when the wind direction is aligned with the ridge as opposed to cases where it is perpendicular to the mountain ridges? This would help to establish the topography as the main reason why upper elevation ERA5 wind speeds correlate so well with the wind speed measurements (even better than any ERA5 heights for the coastal cases).

Second, I would consider methods to remove the need to select the optimal ERA5 height prior to model training. For instance, instead of a single, site-specific optimal height, multiple standard heights identical across all sites could be used. This helps show that the ERA5 data contains useful physical information that the model can utilize to improve wind speed estimates. As it currently stands, the optimal height selection actually includes the testing data as well as the training data. This results in a data leakage that could artificially increase model performance. If the authors decide not to use a multi-height standard model, then the optimal height selection should at least be performed on the training data only. Alternatively, a simple sensitivity analysis could be done to show if the optimal height at each site varies significantly from year to year.

Below are minor technical comments:

1. For figure 4, I would like to also see the PDFs of the training wind speeds. This helps confirm that out-of-sample predictions are not taking place and that there is a different reason for the poor performance at the PDF extremes.
2. Having noted the Cavaiola et al. 2023 study in the introduction (line 69), it would be helpful to give more explanation as to how this study differs from the 2023 study. For instance, more details could be added in the discussion on lines 411-423.

See attached file for comment.