The manuscript describes a statistical (machine learning) approach to derive the uncertainty of wind data from mesoscale model simulations for longer time periods. The topic is relevant for wind energy. Thus, the manuscript should be published after revisions.

The manuscript is clearly written. Nevertheless, reading the manuscript rises a number of issues and questions which should be addressed in more detail in the manuscript.

(1) Lines 37-39: Using marine tower data (e.g. available offshore off the coasts of the Netherlands, Germany or Denmark) should at least be mentioned in the introduction in parallel to the possibility of using floating lidar data. There is more offshore wind data than those from buoys. Please add some references.

(2) Given that there is some tower data (some of them for more than ten years), could testing of the results from this study against long-term time series from the FINO1 or FINO3 platform in the German Bight be an option?

(3) The addressed wind conditions in this manuscript are those in coastal areas. Therefore, the reviewer is wondering why coastal effects in the wind fields are not addressed. This should at least be mentioned in the Introduction.

(4) Some configurations of the mesoscale model WRF have problems representing the atmospheric boundary layer in the transition region between land to open sea (see, e.g., Siedersleben et al., 2018, DOI 10.1127/metz/2018/0900). Has the WRF configuration chosen for this study been checked for the ability to properly simulate the transition from land to sea?

(5) This study addresses uncertainties in wind speed. How does the magnitude of these possible uncertainties in wind speed modelling relate to, e.g., interannual wind speed changes due to long-term oscillations in the global circulation (NAO etc.) or due to climate change?

(6) Figure 7: Why do unstable thermal stratification conditions have such a small share of all cases? Shouldn’t the lagged annual SST variation (due to the much larger thermal inertia of the water) compared to the air temperature variation lead to roughly equally frequent stable and unstable situations (stable in late winter and spring, unstable in late summer and autumn)? Please give an explanation for the found bulk Richardson number distribution.

(7) Why is the 2 m inverse Obukhov length chosen as a parameter at all in Section 2.2? This parameter can only be relevant for the whole rotor area of the future wind turbines in cases with perfect vertical mixing. In stable conditions the depth of the Prandtl layer is much shallower than turbine hub height so that the near-surface Obukhov length is most probably irrelevant for the 100 m wind speed. This is likely to be also true for cases in which internal boundary layers are present (which frequently happens in coastal areas).

(8) Figure 3 demonstrates that the 2 m inverse Obukhov length is not a meaningful parameter, because it is always close to zero. Logically, the weight 0 has been attributed to this parameter in Table 4. So, once again, why has this parameter been included into this study?

(9) Assumed that the inverse Obukhov length has nevertheless some relevance, a not-discussed contradiction is that there seems to be a slight bias of the inverse Obukhov length towards instability in Figure 3 while there is a strong bias of the bulk Richardson number towards stable conditions in Figure 7? This should be explained. It seems to confirm the irrelevance of the inverse Obukhov length stated above.

(10) A major finding of this study is the increase in wind speed uncertainty close to the coast and for non-neutral stratifications. Does this point to weaknesses in the used method or to weaknesses in the used mesoscale model? Or is this a natural feature which comes out of this study? Please discuss this point in the Conclusions.

The manuscript by Bodini et al. on "Assessming Boundary Condition and Parametric Uncertainty in 

Numerical-Weather-Prediction-Modelled, Long-Term Offshore Wind speed Through Machine Learning

and Analog Ensemble" focusses on comparing two different methods on how to extrapolate

the ensemble uncertainty of a short (one year time) series to a long time series for the 

wind conditions along the coast of California. The study is generally well written,

figures are selected meaninfully and readable. However, I have a number of minor comments to

be taking into account before I can recommend publication in Wind Energy Science.

Minor Comments:

- Line 28:  Northern Europe is typically referred to as Scandinavia. I guess you rather mean central Europe here?

- Line 43:  "a continuous, in space and time''  → I suggest: an in space and time continuous

- Line 76:  "prohibitive, and innovative and more computationally''  → remove the "and'' before innovative

- Line 103:  "The 15 WRF ensemble members''  →  I strongly recommend to create a table here with one row per

one of the 16 members and then the settings in the columns. It is hard to read and understand here.

- Line 131:  "mean hourly wind speed''  → How did you compile the mean hourly wind speed? How many timesteps

have you written out from the WRF runs to compute the average?

Line 147:   "the WRF across-ensemble standard deviation''  →  what is the across-ensemble standard deviation. Is this the

standard deviation from the ensemble members?

Line 156-157: Can you explain why you used both the standard deviation from preceding 2 and 6 hours and how these time intervals were

selected? 

Line 161: "However, we found that including all the features''  →  can you explain how you found this? And did you try further

features?

Table 2: It is quite difficult to read column two. Maybe it helps to add a small empty row after each row?

Table 4: I think you should at least try to explain why there is no weight on the inverse Obukhov length but a strong weight on shear.

Line 230 and Line 303-304: "Applying the bias correction proposed in"  →  So, if this would likely reduced the AnEn bias, why didn't you

do it? This should at least be explained. 

Figures 6 and 8: There should be a space between caption text and unit. 

Line 279-280: "For each atmospheric stability class, the values shown are the''  →  Maybe an active sentence is better here, e.g.

for each atmospheric stability class, we show....

Line 293 and the whole Conclusions section: I suggest to replace Weather Research and Forecasting or the abbreviation WRF by "mesoscale'' everywhere in the conclusion, because the results and conclusions should in principle be valid to any mesoscale ensemble dataset, shouldn't they?

Line 319: "deployed in the California OCS very recently''  →  I guess you mean this buoy? https://a2e.energy.gov/data/buoy/lidar.z06.00

Maybe its worth referencing the dataset (including the doi) here.

References: The references are vastly incomplete. Every journal paper should have a doi and every technical report a link or other

accessibility information.