This paper uses roughly 1-year of surface-based doppler lidar wind observations collected at two locations off the west coast of California (Morro Bay and Humboldt Bay) to evaluate the 3-hourly wind speed forecasts from the High-Resolution Rapid Refresh (HRRR) numerical weather prediction model at several different heights from 40 to 240 m above the sea surface.  They have demonstrated that overall, the HRRR’s performance matches the observations pretty well, with better agreement at Morro Bay vs Humboldt Bay, but that there are conditions where the model has significant wind speed errors. 

Generally speaking, this paper reads very well.  The figures are clean, the writing is clear, the results are largely well supported.  I have listed a number of comments / suggestions below that I’d like the authors to consider and address before the paper could be accepted for publication. 

Major comments:

Line 102: what is the accuracy and uncertainty of the GOES derived cloud top heights?  Is there any seasonal dependency to these values? 

Line 102: how are multi-layer cloud systems handled?  Certainly there are cases where there is an upper level cirrus cloud that obscures the lower atmosphere?

Line 111 regarding your “minor discrepancies” comment: could this be a definition and/or sensitivity issue?  For example, the GOES can only identify a cloud layer if the optical depth is above some threshold (which might depend on other atmospheric conditions), whereas the model might define a cloud if there are any hydrometeors in a volume.  How was this handled?  Was an instrument simulator applied to the HRRR’s output to try to mimic the GOES observations?  (I suspect not, but you should at least be clear that this is an issue that could impact these analyses).

Line 134: One of the limitations of this study is that the liquid water path (LWP) cannot be compared between the GOES and model easily.  My sense is that the model has too little LWP in its cloud, and thus in the daytime there is too little absorption of shortwave radiation, and thus too little diabatic heating in the cloud, which is why the cloud did not deepen.  Your suggestion that it could be an issue with longwave radiative cooling is also possible, but again, this connects to errors in getting the diurnal evolution of the LWP in the cloud correct.  So please expand this discussion a bit more.

Line 230: You are hypothesizing, here and later in the paper, that the model is not capturing stability properly and that could be the source of the bias.  However, error in the sea surface temperature (SST), which is a boundary condition for the model, is another possibility.  And it is possible that there are different errors in SST in Morro vs Humboldt Bays, especially in clear skies?  That would affect the stability profile, which would then feed back into the turbulence and wind profiles.  I believe the buoys have SST observations on them, and it would be a pretty straight-forward analysis to determine if SST biases are correlated with the wind biases.  This might result in your need to update the statement at line 350.

Line 253: you state “similarity-based wind speed profile model”.  The way this is worded suggests that the HRRR is using this approach.  Similarly, in line 348, you give the same impression.  The HRRR uses Monin-Obukhov theory in the surface layer; however, above the surface layer the HRRR uses an eddy diffusivity mass flux approach.  See papers by Olson et al (BAMS 2019) and NOAA Tech Memo (https://doi.org/10.25923/n9wm-be49). 

Figure 8 and the discussion starting at line 299: There are both resolved clouds (i.e., where the entire model grid cell is cloudy) and subgrid-scale clouds (i.e., where there is partial cloudiness in a grid cell).  Unfortunately, the HRRR does not save the liquid water content operationally to an output file.  Thus, this figure can only show the resolved cloud liquid clouds.  It is possible that there is significant subgrid-scale cloud liquid between the “gaps” shown in Fig 8a and discussed at line 305, but we just don’t know for sure.  You can get the LWP from the HRRR from the HRRR data archive on the AWS cloud server; the LWP is the vertical integral of both the resolved and subgrid-scale liquid water content profiles.  

Minor comments:

Line 13: “multiple instruments”

Line 14: there are many types of lidar.  Please say “including Doppler lidar”

Line 54: …resource assessments over the United States.  (the HRRR’s domain)

Line 80: Isn’t GOES-17 the only satellite that is relevant for this paper? 

Line 130: “in both locations”  --> “in either location”

Line 131: I believe you have this reversed, as the GOES shows higher values in the daytime in Fig 2

Figure 3: please use the same y-axis range for all panels within the figure, as it would make it much easier to compare the different panels

Figure 4: same comment

Line 173: “overestimates wind shear” – this is not supported by Table 2, which shows the opposite

Line 214: “model consistently underestimates” – I disagree.  Fig 5 top panel shows that the HRRR is essentially unbiased for wind speeds less than 10 m/s at Morro Bay

Figure 6: why didn’t you use the same style of plot as in Figure 5?  It would make the paper more consistent and easier to read

Line 261: “decoupled during the day” – is this always true (i.e., for every cloudy day in your analysis)?  I would be surprised if this was true

Line 275: the differences in the correlations between the two locations are important.  Modify the end of this sentence to indicate the mean values of the correlation coefficient to help strengthen this point

Figure 7: the blue minus black lines in panes c and d do NOT equal the results shown in panels e and f.  Please recompute and update the figure

Synopsis:

The manuscript with the title “Characterization of HRRR simulated Rotor Layer Wind Speeds and Clouds along Coast of California”, which was submitted for publication to the journal Wind Energy Science by the authors Jungmin Leem, Virendra P. Ghate, Arka Mitra, Lee M. Miller, Raghavendra Krishnamurthy and Ulrike Egerer, deals with the evaluation of wind speed and cloud forecasts by the HRRR model for two offshore sites off the coast of California. Data from 3-hour forecast from HRRR are compared with data collected with lidars and additional meteorological sensors on two buoys as well as with satellite data. According to the results presented by the authors the observed seasonal cycle of cloud top height is well reproduced at both sites by the HRRR model. However, in the warm season stratocumulus cloud top heights are underestimated by HRRR, especially at one of the two sites. Another finding by the authors is that clear sky conditions come along with larger wind speeds at the the two sites investigated than cloudy conditions. Clear-sky conditions come also along with a larger bias of the wind speeds predicted by HRRR, although the sign of the bias is different at the two sites for which the authors did the analysis.

Evaluation:  

Understanding how the presence of clouds inside the atmospheric boundary layer changes the wind speed in the atmospheric boundary layer is a topic of high relevance for the wind energy community. The understanding of such processes is a key for modeling these processes correctly and thus enabling also an improved modeling of wind conditions. Thus, in my opinion the topic of the manuscript is interesting for the wind energy community and is basically suitable for being published in Wind Energy Science. However, I have a few comments on the current version of the manuscript that might support sharpening of the clarity of the paper. Moreover, I have a couple of minor comments. Therefore, my recommendation is that in the next step the authors  should apply some corrections to the current version of the manuscript before a decision on its publication in Wind Energy Science can be taken.

Comments on the content:

#1: Abstract: From my point of view the authors could and should present the objectives of the paper more clearly in the abstract.

#2: Abstract: The authors claim that “the findings from this study will potentially inform how to improve the modeling of wind resources off the coast of Northern California”. However, there are no clear advices for the next steps towards improving the modeling of wind resources given in the manuscript. Therefore, I think that either the respective sentence in the abstract should be revised or the conclusions part of the manuscript should be extended with corresponding content.

#3: General comment: The literature review presented in the introduction is rather short. A more extensive introduction e.g. into stratocumulus-topped ABL could be valuable for the reader.  I think e.g. from the paper by Kopec et al. (2016) some relevant information on stratocumulus-topped  ABLs could be presented. Concerning the HRRR model I’m missing some examples of previous studies that evaluated that model (especially for wind energy purposes, but also information on other evaluation studies might be interesting). Moreover, the model itself could be presented in more detail. E.g., what are the initial and boundary conditions used?

Kopec, M. K., Malinowski, S. P., Piotrowski, Z. P., 2016: Effects of wind shear and radiative cooling on the stratocumulus-topped boundary layer, Q. J. R. Meteorol. Soc., 142, 3222-3233, https://doi.org/10.1002/qj.2903  

#4: Line 53/54: “… despite HRRR being one of the most widely used forecasting tools in wind energy resource assessments” Please add references for this statement. Is this forecasting tool applied for wind resource assessment (i.e. derivation of wind speed information for periods with a length of decades)? My understanding is that  mostly models run in hindcast mode are used for wind resource assessment studies. The limited time available for producing a forecast limits e.g. the time that can be spent on the data assimilation process. Moreover, later on it is stated that HRRR is applied for wind power forecasting. I would see this as something different from a wind resource assessment.. HRRR does not output wind power directly. Thus, there might be wind power forecasting systems that make use of HRRR results for generating a wind power forecast, but I think using HRRR alone will not allow you to do a wind power forecast. To summarize, I ask the authors to add clarifications concerning the mentioned points to their manuscript.

#5: Line 94/95: “The HRRR reported wind profiles were linearly interpolated …” Why has a linear interpolation been applied. In the surface layer the wind profile is expected to be logarithmic. Wouldn’t it therefore make more sense to apply a logarithmic interpolation in this area? Another possibility would be to apply a power law profile for the wind speed and use this for interpolation.

#6: Line 97-99: The current description “we extract the corresponding lidar data for each hour” could still be made more precise, e.g. as follows: As HRRR provides data only at a full hour we also use only 10-minute averaged lidar data with the same timestamp as that of the forecast.

#7: How do the authors deal with the fact of HRRR and lidar data being effectively different types of averaged data? E.g. does 10 minute averaged lidar data compare better to HRRR than 30 minute averaged lidar data? Did the authors check for a possible phase shift between the lidar and the model data? Why did the authors decide not to interpolate the model data in the horizontal directions of space to the position of the lidar measurements?   

#8: Figure 3: The following comment is also connected to my impression that the manuscript could benefit from improving the clarity of its objectives. Is HRRR evaluated with respect to its potential for resource assessment or wind power forecasting?  Is the PDF of the wind speed the best quantity to assess the performance of a tool used for forecasting the wind speed? Even if the PDFs of the model and the measurements looked perfectly the same, this would not necessarily mean that it Is well suited for the purpose of wind power forecasting. For that purpose other parameters like the absolute bias would be more relevant. Thus, if the objective were evaluating HRRR for the purpose of wind power forecasting I would recommend to not starting the evaluation with presenting the PDFs. 

#9: Line 172: “This suggests that HRRR model overestimates wind shear at Humboldt Bay location compared to the observations” The authors derive this statement from the PDFs of the wind speeds at different heights. I’m wondering why the authors do not directly investigate and present the wind shear itself before they conclude on it. I think this would strengthen the statements made on the wind shear.

#10: Line 253-254: “As stated in Optis et al. (2016) and the references therein, similarity-based wind speed profile models produce a large bias under strongly stratified boundary layer.” Is it the aim here to provide an explanation for the bias in strongly stratified boundary layers observed by the authors? In that case the link between HRRR and similarity theory should be made clearer.   

#11: Line 269: “Some cases had breaks in the cloud for less than 3 hours that got filled” I think the authors should elaborate on this a little bit further. For me the meaning of this statement was unclear. What was filled with what?

#12: Figure 8b): Is there an explanation why the lidar data, although it is averaged over 10 minutes, is fluctuating quite strongly (and seemingly with some preferred frequency), while the model data is comparatively smooth?

#13: Line 317-319: “Therefore, better understanding the processes within the marine boundary layer bounded by marine stratocumulus clouds is important to better assess and predict wind resources for offshore wind farms in California.” Is it necessary to limit this statement to offshore wind farms off the coast of California?

Minor comments:

#1: Line 20: Please change “… those fields are stronger …” to “ … those parameters have larger values  …”.

#2: Line 40: Please change “… were made from a research flights” to “ … were made from research flights”.

#3: Line 46: Please change “Staring October 2020” to  “Starting in October 2020”.

#4: Line 59: Please change “model wind bias in relation to of various” to “model wind bias in relation to”.

#5: Line 62: Please delete “at the Humboldt location”.

#6: Line 71: Please change “from from 40” to “from 40”.

#7: Line 75: I think the sentence “Temporal cloud mask is estimated to be cloudy” should be rephrased. What is a temporal cloud mask? This term needs to be better introduced.

#8: Section 2.3: The description of HRRR should be extended. In the current text there is no information on the data assimilation process that is used in the model. It would be good to add references to previous evaluation studies that support that HRRR provide forecast with high-fidelity as stated in the description by the authors.

#9: Line 92-94: The corresponding sentence should be revised to avoid a possible misunderstanding.  “… are close to the heights of the lidar measurement, which is at 40, 80, 160 and 240 meters” should therefore be changed to  “… are close to the heights of 40, 80. 160 and 240 m at which data from lidar measurements is available”

#10: Line 116-118: “In December 2020, a large wave event at Humboldt buoy location results in the power outage and lead to a large data gap. The machines on the buoy were back online on May 25^th, 2021, at Humboldt location.” This information had been provided in the manuscript before. Therefore, there is a reduncancy that should be removed.

#11: Line 138, line 142: References are made to a figure S1, however, there is no figure S1 in the manuscript. Please correct these references.

#12: Figure 3: The following comment is also connected to my impression that the manuscript could benefit from improving the clarity of its objectives. Is HRRR evaluated with respect to its potential for resource assessment or wind power forecasting?  Is the PDF of the wind speed the best quantity to assess the performance of a tool used for forecasting the wind speed? Even if the PDFs of the model and the measurements looked perfectly the same, this would not necessarily mean that it Is well suited for the wind power forecast. Other parameters like e.g. the absolute bias would be more relevant. Thus, if the objective were evaluating HRRR for the purpose of wind power forecasting I would suggest not starting with  the presenation of  the PDFs of wind speed. 

#13: Line 260: Please change “As a results …” to “ As a result”.

#14: Line 343: Please change “… wind bias …” to “wind speed bias”.