Dear Authors,

Thank you for a well-written paper. You have collected a comprehensive dataset of wind measurements (62 sites) and production data (55 small wind turbines) and used it to evaluate three wind resource models (WIND Toolkit, ERA5, and MERRA-2) and three simulation tools (SAM, Wind Report, and MyWindTurbine.com). It is an ambitious analysis. The comprehensive evaluation is the strength but also the weakness of the paper. There are many uncertainties associated with energy yield assessment, wind measurements and small wind turbine performance (e.g. SAM and MyWT assumes 18% & 7% loss), but the manuscript can not detail all aspects. I would like to see a general discussion about the prediction error and whether it is within the expected uncertainty - does the simulation tools provide added value, or are wind measurements needed for proper energy yield assessments?

Wind Report and SAM make estimates by interpolating within coarsely resolved wind resource grids (2 and 5km). MyWT is fundamentally different in that it uses high-resolution micro-scale and obstacle flow models (<1m) to downscale the large-scale forcing found from the wind resource model (50km in GWA1). The paper hints that the poor performance of MyWT is due to the coarse resolution of the wind resource model. However, as the sheltering effect of nearby trees and buildings can be significant for small turbines (like the 8.9 kW turbine), they should be included in the analysis. MyWT requires the user to input the exact location and type of obstacles; these are not defaults. The paper does not mention if obstacles were included, but if not, it should be noted as the probable cause for the high wind speeds predictions.

The Global Wind Atlas 3 (GWA3) is another easy-to-use tool to estimate wind speed and the energy yield of wind turbines. The GWA3 is based on large-scale forcing from WRF simulations (3km) with global coverage down-scaled to 250m resolution. It would be interesting to compare GWA3 (3km) and GWA1 (50km) as this could indicate if the prediction error is caused by the resolution of the wind resource model or the lack of an obstacle model.

General Comments:

The paper reports on a well-reasoned and performed piece of work related to wind speed and wind energy assessment for small, distributed wind turbines.  The approach taken, data used, and results presented are useful and needed, and the work represents a step forward.  Identification of the source data is useful, especially for any that would like to use the wind simulation data from the sources listed.  Typical proposed small wind turbine installations do not have good met tower or weather data from nearby sources, so the use of simulation data is very practical, and thus the some understanding of its ultimate accuracy in predicting AEP is critical.

Specific Comments:

The wind speed bias errors are only positive for the smallest wind speed bin plotted in Fig. 5 (0-5 m/s), yet the wind turbine capacity factors are almost all overpredicted (Fig. 7).  This seems to imply that the wind speed spends most of its time in the lowest wind speed bin. However, most of the wind power curves shown in Fig. 2 show little production in this wind speed range. Would it be possible to add to Fig. 5 the percent of time that the wind speed was in each wind speed bin?  That could help in interpreting the results.  Moreover, if they possess sufficient time resolution in the wind turbine production data, it would be helpful if the authors provided an explanation for the overestimate of capacity factor… at which wind speeds (or wind speed bins) does the overestimate of energy occur?

The (simulated – actual) capacity factors for 55 turbines are presented in Fig. 7, split up by region of the country. Another interesting categorization would be by “surface roughness” or something similar that provides an indication of whether the area in the vicinity of the turbine is forest, fields, etc.  It would also be nice to sort by complex terrain vs non-complex, and to provide some suggestion of how to distinguish between the two.  Have the authors considered looking at the data in these ways?

Fig. 10 shows availability statistics for 9 turbines of varying ages.  Why were only nine turbines used?  Also, did the authors sort the availably by turbine age? 

For the conclusions: Do the authors have any recommendation about how a person whose job is estimating small wind turbine AEP/capacity factor should account for interannual variability, given its significant influence on capacity factor?   

Technical Corrections:

Page 6 Line 148:  Should read “2.1 kW to 56 kW”  NOT “8.9 kW to 56 kW”

I believe the Southwest WindPower Skystream is now owned by Xzeres.  Perhaps some mention should be made of this when introducing the turbine?