General comments:

The manuscript “Wind data assessment and energy estimation for potential future offshore wind farm development areas in the Scotian Shelf” by the authors Yongying Ma, Jinshan Xu, Yongsheng Wu, Michael Z. Li, Ryan Stanley, Brent Law and Marc Skinner presents a study on the assessment of the wind potential in the Scotian shelf. Wind speed and wind direction data from four different reanalysis data sets (ERA5, CFSv2, NARR, HRDPS) are compared vs. observational data collected in that region in order to identify those model data sets that show the best agreement with observations (ERA5, HRDPS). Data from these data sets is then used to provide input data for wind farm simulations with the tool PyWake for several potential wind farm sites in the Scotian Shelf. The authors carry out a sensitivity study in which they investigate the dependency of the power output of the simulated wind farms from the distance between two wind turbines in those wind farms.

The manuscript does not supply the reader with new methodologies, but applies for the first-time standard tools and data sets to the assessment of wind resources in the geographic region of the Scotian shelf. While the language and overall structure of the manuscript is fine, in my opinion the readability of the paper could profit considerably from slightly restructuring its contents. My suggestion would be e.g. to organize the report on the performance of the different reanalysis data sets not by the metrics but strictly by the data sets. The reader might be most interested in an overall assessment of the data sets and less in a slightly too detailed  presentation which data set provides the smallest bias, which the smallest RMSE and so on.

I have a couple of specific comments which I ask the authors to consider when revising their manuscript.    

Specific comments:

Lacking wind measurements at larger height:

According to table 1 the largest measurement height from which data was accessible to the authors was 16 m. Modern wind turbines operate in much larger heights. I’m wondering how transferable the results on the quality of the different reanalysis data sets assessed by the authors for heights close to the ground actually are to larger heights. In my opinion the authors should discuss this point when assessing their results.

Lacking inclusion of atmospheric stability:

The vertical wind profile, the turbulence in the marine atmospheric boundary layer and the wake recovery are in practice dependent on the atmospheric stability. However, this parameter is not discussed at all in the manuscript. The authors should at least explain why they have excluded this parameter from their analysis and what the non-consideration of atmospheric stability means for the uncertainty of the results presented by the authors.

Moreover, I’m lacking a discussion on the impact of atmospheric stability on the seasonal changes of the wind conditions in the Scotian Shelf area. Can seasonal changes in the error metrics be related to lacking consideration of atmospheric stability in the analysis?

Interpolation of NARR data in time:

My suggestion would be to compare the different data sets for a temporal resolution of three hours with each other. The interpolation in time might introduce another uncertainty that is not in the original NARR data itself. It should be possible to quantify the impact of the interpolation in time by comparing error metrics for the original NARR data in the gap-filled NARR data with each other.

Filtering for wind speeds between 2 m/s and 17 m/s:

In my opinion it would be also an important criterion whether a reanalysis data set gives the right number of events with wind speeds above cut-out wind speed. I suggest to add such an analysis to the existing analysis. Or is the number of such events too low to have an impact on the calculation of the energy yield in the end?

PyWake:

In my opinion the current description of the wind farm model does not contain all the information that would be required by the user to repeat the calculations of the authors. Therefore, I ask the authors to extent the description of the setup of their PyWake runs. E.g., how has the background turbulence intensity been considered in these simulations? Is the model applicable also for calculations of wind turbines that operate in the near wake of other wind turbines? With the smallest turbine distances assumed in the sensitivity study of the authors they might already be in the near-wake range. This is an important comment e.g. for the accuracy of the results presented in  figure 9.

Error metrics for the wind direction:

As averaging of wind directions is often not made correct I encourage the authors to sensitize the readers and present more details in how they handled the jump of the wind direction at 360°/0° in their analysis.

Page 3, line 81: What is meant by characteristic wind speed in this context?

Table 2: I’m wondering whether this table is actually required. E.g., the information on the time range and the spatial coverage is not of importance for this manuscript.

Page 7, line 174: What is a “naturally-stable” atmospheric condition?

Page 8, line 175: Power law exponent 1/7. Don’t ERA5 and HRDPS provide data on other heights as 10 m? If they provide such data, I suggest to determined the power law exponent from the reanalysis data sets. Or is there a special reason why the authors trust more in the 10 m wind speeds than in the wind speeds from other heights in these data sets?

Section 4.1: I’m wondering whether the wind farm simulations for just the seasonal mean wind speed are sufficient here. What does this tell us concerning the energy yield to be expected when the power in the wind is actually depending on the cube of the wind speed?

Table 5: The explanation of x_m and x_t should be presented before table 5 is presented. I had difficulties to interpret these parameters without having read the information on these parameters in the text.

Technical corrections:

Abstract, line 1: “The Scotian Shelf is one of the top wind regimes in the world.” Regimes should be replaced by regions.