The paper introduces the Phy-RF model, an innovative approach combining physical wind profile models with machine learning to extend wind profile estimations beyond the surface layer. Through comprehensive analysis and validation, the study demonstrates that the Phy-RF model offers a more accurate estimation of wind speeds at 100 meters, outperforming both the traditional power law method (PLM) and random forest (RF) machine learning algorithm alone. In general, this paper is well written. I recommend the publication of this manuscript, while I have some comments below for the authors to address.

Specific comments:

1. The paper compares the seasonal mean of ERA-5 and Phy-RF.  It could also benefit from a direct comparison between the Phy-RF model outputs and ERA-5 wind speed data at 100 meters using scatter plots. Since the model uses ERA-5 as the input, such a comparison would be beneficial for evaluating the Phy-RF model's performance.

2. While the study incorporates data from several ARM sites, there appears to be insufficient evaluation using these datasets. Since averaged profiles may not capture the full extent of discrepancies between the Phy-RF model estimates and observations, relying on mean profiles from lidar cannot accurately represent accuracy.

3. More detailed statistical analysis, such as examining the mean absolute error (MAE), could enhance our understanding of the model's performance. I suggest including the MAE to evaluate the differences between Phy-RF, ERA-5, and Lidar.

4. I noticed there are notable discrepancies in the mean profiles of Phy-RF and Lidar over the SGP. What factors lead to such biases? Meanwhile, Are such biases also included in the ERA-5?

5. The paper may include its discussion on the limitations of the Phy-RF model, particularly how it performs under extreme events.

6. Despite a worse performance compared to Phy-RF, the traditional PLM also seems good (Figure 7). It would be beneficial to discuss more clearly the potential applications and advantages of the Phy-RF approach, particularly in scenarios where the PLM and ERA-5 may fall short.

Please find my comments in the PDF supplement.