Meso- to microscale modeling of atmospheric stability effects on wind turbine wake behavior in complex terrain

Wise, Adam S.; Neher, James M. T.; Arthur, Robert S.; Mirocha, Jeffrey D.; Lundquist, Julie K.; Chow, Fotini K.

doi:https://doi.org/10.5194/wes-7-367-2022

Articles | Volume 7, issue 1

https://doi.org/10.5194/wes-7-367-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/wes-7-367-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 7, issue 1

Research article

|

17 Feb 2022

Research article |

| 17 Feb 2022

Meso- to microscale modeling of atmospheric stability effects on wind turbine wake behavior in complex terrain

Adam S. Wise, James M. T. Neher, Robert S. Arthur, Jeffrey D. Mirocha, Julie K. Lundquist, and Fotini K. Chow

Download

Final revised paper (published on 17 Feb 2022)
Preprint (discussion started on 10 May 2021)

Interactive discussion

Status: closed

CC1:
'Comment on wes-2021-36', Alfredo Peña, 12 May 2021

Dear authors,
Thanks a lot for this manuscript. The results are really impressive and interesting. I have some questions/comments on your work and I hope you can address them:
1. As far as I understand you do not nudge your simulations. Is there any reason why you don't? I would tend to think that under complex terrain nudging could be good to avoid the runs to drift apart too much. And you mentioned that you chose GFS due to better agreement with ERA5 (or perhaps ERA-interim)... would nudging turn the agreement to be better using ERA5?
2. We recently published a paper where we compare lidar RHI scans of a hydraulic jump at Alaiz and the results of a multi-scale WRF-LES run of the event (Peña and Santos, 2021). Similarly to your results the model was able to capture pretty well the dynamics and features of the hydraulic jump. We did not use a CPM method to trigger turbulence and I wonder whether you have results without using CPM.
3. Since you use CPM (I guess you could comment on how you setup your CPM in a better detail in the corresponding section), I wonder how resolved + SGS turbulence (velocity variances or TKE) compares against the measurements. It would be really interesting to see time series of simulated and observed TKE and TKE vertical profiles of the simulations against those from the masts in the inflow (no wake). TKE transects between simulations and the lidar RHIs could also be really nice to see, although I know of the difficulties in attempting such type of comparison.
Thanks a lot in advance!
Peña, A., & Santos, P. (2021). Lidar observations and numerical simulations of an atmospheric hydraulic jump and mountain waves. JGR Atmos. 126, e2020JD033744. https://doi.org/10.1029/2020JD033744

Disclaimer: this community comment is written by an individual and does not necessarily reflect the opinion of their employer.

Citation: https://doi.org/10.5194/wes-2021-36-CC1
- AC1:
  'Reply on CC1', Adam Wise, 02 Jun 2021
  Dear Dr. Peña,
  Thank you very much for your kind remarks and questions. And thank you for sharing your work for Alaiz, it was a really interesting read. To address each one of your questions:
  You are correct in that we do not nudge our simulations. We felt that nudging was not necessary because there was reasonably good agreement between our results and the observations. We think that nudging is only necessary when there is a serious mismatch with observations. For example, in Arthur et al. 2020, nudging was used because WRF in forecast mode had trouble predicting the dynamics and timing of the frontal passage. In our case, the mesoscale forcing was not drifting from observations. In the runs using ERA5, the height of the jet was severely underestimated and we think that it is unlikely that nudging these runs would have led to better agreement. We could explore nudging more but feel that it is beyond the scope of our work.
  
  Thank you for sharing your work. It was an interesting read with some great figures. We do not have results without CPM. CPM does not add computational cost, and our previous work in this domain, for different cases, suggested that CPM enhanced agreement with tower observations. More detail regarding the CPM methodology can be found in Connolly et al. 2021. In our stable case, CPM only triggers and accelerates any small scale turbulent structures, which are mainly below the jet. However, in complex terrain, the terrain itself induces enough turbulence that it is difficult to determine the exact role of CPM.
  
  Thanks for the suggestions! We are actually currently working on this as a separate paper, including your suggestion and also analyzing the Perdigao case with an additional turbulence closure model. More closely examining the turbulence quantities such as dissipation, TKE, and variances is really important. Connolly et al. 2021 provided some comparisons for total TKE to observations.
  
  Thanks so much for your questions!
  Connolly, A., van Veen, L., Neher, J.M., Geurts, B.J., Mirocha, J., and F.K. Chow. 2021. Efficacy of the cell perturbation method in large-eddy simulations of boundary layer flow over complex terrain. Atmosphere 12(1), 55,1-28. doi.org/10.3390/atmos12010055
  Arthur, R.S., Mirocha, J.D., Marjanovic, N., Hirth, B.D., Schroeder, J.L., Wharton, S. and F.K. Chow. 2020. Multi-scale simulation of wind farm performance during a frontal passage. Atmosphere 11(1), 245, 1-17. doi.org/10.3390/atmos11030245
  
  Citation: https://doi.org/10.5194/wes-2021-36-AC1
RC1:
'Comment on wes-2021-36', Dries Allaerts, 08 Jun 2021
This paper studies the interplay between complex terrain, atmospheric stability, and a wind turbine wake at the Perdigão site by means of large-eddy simulations driven by realistic mesoscale forcing conditions. The paper focuses on two particular case studies with distinct atmospheric stability conditions: a case with stable atmospheric stability conditions in which a mountain wave occurs, and one during convective conditions where a recirculation zone forms in the lee of the ridge with the turbine. The authors used the WRF-LES-GAD multi-scale modelling suite, which couples mesoscale WRF to microscale WRF-LES including a generalized actuator disk model, and they validated the simulations by comparing to field observations with in-situ and remote sensing instrumentation. The paper provides a very detailed description of the terrain-induced flow features and an extensive validation with field data. Furthermore, the interaction between the terrain-induced flow features and the wind-turbine wake is also studied in depth. I believe the paper is well written and will be of interest to the wind energy community, both for demonstrating the viability of WRF-LES-GAD and for explaining the interplay between complex terrain, atmospheric stability and wind-turbine wakes. I do have some scientific questions and a number of minor or and technical comments as listed below.

Scientific questions:

Can you comment on how representative the specific case studies (stable in combination with mountain waves, unstable in combination with a recirculation zone) are for the wind conditions at Perdigão and for complex terrain in general? Do you think the conclusions in terms of wind-turbine wake behavior (terrain following in stable, deflecting upwards due to recirculation) will hold for all stable/unstable wind conditions at this site and for complex terrain in general?

The inverse of the Froude number as defined in Equation 1 is often called the non-dimensional mountain height and it represents the ratio of the mountain height to the (vertical) wave length of the mountain wave. When discussing the case where the mountain wave length is shorter than the width of the mountain (line 133-134), wouldn’t it be more appropriate to use a Froude number based on the width rather than the height of the mountain like for example in the book of Stull (1988, section 14.2.3 Flow over hills). Obviously, this also affects figure 3 and later calculations of the Froude number in section 4.1.

It is not entirely clear how you setup the semi-idealized simulations. What pressure gradient force do you impose, or alternatively what wind speed and direction do you enforce (and at what height)? Do you apply a negative surface heat flux in the stable case? What is the domain height of these simulations? Do you use any damping layers at the top? For how long did you simulate these idealized cases?

Line 182-183: What do you mean with the stratification is self-destructive? Do you mean the stable stratification turns into a constant temperature profile because of turbulent mixing? Does the simulation become unstable due to inertial oscillations? Something else?

Figure 12: Is there any averaging of model results or measurements? Under higher turbulent conditions, does it make sense to compare instantaneous velocity fields with the point measurements of the met towers given the chaotic nature of turbulence?

Line 361 “Differences in 80m mean wind speeds … are less than 1 m/s …”: Is this the difference averaged over the entire time horizon, or the root-mean square, or something else? When I look at figure 13b around 13:55, I obviously see differences of more than 1m/s. Similar for the discussion of the differences in the wind direction and also for the plots for stable conditions. You seem to be talking about a certain average difference being below a certain value, but the instantaneous differences are clearly higher than this value and it is not clear how the averaging is done. Please clarify.

Line 445: The blockage addressed in the papers by Meyer-Forsting, Bleeg and Segalini refers to the global blockage effect far upstream of large wind farms that arises due to the combination of many wind turbines. Here, you only have one turbine and you are only looking at 2D upstream, so I would say that turbine induction might be playing a role, not blockage. I don’t think it is appropriate to refer to these wind-farm blockage studies here, so consider removing these references.

Minor questions and technical comments:

Line 148: Please mention at least once how UTC relates to local time at this particular site. Maybe it is even more interesting to mention what time corresponds to sunrise.

Line 148-149: Please mention what value of the wind direction corresponds to a direction perpendicular to the ridge such that the statement can actually be assessed by the reader.

Line 149 reference to Fig 2a and 2b: I don’t think this reference is correct. Figures 2a and b show wind speed in stable and convective conditions, while the sentence was talking about the wind direction.

Line 155-156 and figure 2d: Wouldn’t it make more sense here to use an actual stability parameter like the Obukhov length L or z/L to assess the atmospheric stability? Especially when talking about moderately convective conditions, how can you assess that based solely on the potential temperature gradient?

Line 158 reference to Fig 2d: Again this reference seems to be a bit misplaced as it refers to the wind direction plot, not the wind speed plot.

Fig2: Please mention hub height of 80m (?) and measurement points for the temperature gradient (10m and 100m?) in the caption for clarity.

End of line 188: surface roughness length should be in units of meters.

Line 291, 356, and many others : You often start a new paragraph without clearly indicating that you will be talking about a new figure. For instance on line 291, are you still referring to figure 7, or are you rather referring to figure 8 or table 2?

Table 2 and 3: the results in these tables are not really discussed it seems. Can you somehow use these results when assessing/discussing the results of related figures?

Line 312: How high above the actual surface is the nose of the jet which you use to calculate the Froude number? The Brunt-Vaisala frequency is based on the temperature gradient between 10m and 100m, but it is not clear whether the nose of the jet is at about the same height above the surface.

Line 320: How high is the inversion layer above the valley floor (600m asl but at what altitude is the valley floor)?

Line 323: When talking about the striation of lower wind speed close to 600 asl in fig 6b and d, in what ranges of x can this be seen?

Line 357 “there is less variability in the potential temperature gradient for the model compared with observations”: I’d say there is more variability in the model output compared with the observations (this is also mentioned later in the text). The statement seems to contradict the results shown in figure 13 g,h,i.

Line 427: The year of the publication by Churchfield and Sirnivas is missing (also in the list of references).
Citation: https://doi.org/10.5194/wes-2021-36-RC1
- AC2: 'Reply on RC1', Adam Wise, 28 Jul 2021
  
  We thank the reviewer for the time spent reviewing our work and for the detailed suggestions and comments. Please see our attached responses.
  
  Citation: https://doi.org/10.5194/wes-2021-36-AC2
RC2:
'Comment on wes-2021-36', Anonymous Referee #2, 25 Jun 2021
Reviewer general comment: The current manuscript presents results of a suite of WRF-LES configurations around the complex topography of the Perdigão field site. Results represent the first time that WRF- LES is used to study the flow interaction in complex terrain and wind turbines. Results of the numerical simulations are qualitatively compared to the experimental measurements. According to the authors, the features of interest for the present work are: (a) mountain waves, (b) recirculation zones, and how these interact with a wind turbine and the corresponding wake. The manuscript is very nicely written, clear, and with superb “candy-to-the-eye” type figures. Therefore one could say that the article is excellent (despite a few typos or unclear elements). However, to this reviewer’s opinion, while the general idea of the work is interesting and filled with interesting challenges and scientific unknowns, the work presented here remains poor in scientific content, and falls short of addressing any of the initial science-based goals. Let me ex- plain, intercomparisons between numerical models and experimental measurements are critical to be able to objectively determine the quality of the simulation results. However, if the effort stops at that, then it only becomes a mere technical work that anyone outside of the academic world could do. There are indeed publications where industry models are intercompared with experimental data, which have been used in the past to provide a sense of confidence or trust to industry. Alternatively, there are also publications from the academic world, where intercomparisons between models and experimental data are done, but in those there is traditionally a critical analysis of the influence of using different numerical schemes, subgrid models, filtering approaches, etc. So at the end there is an evident scientific gain. Unfortunately in the present work the potential scientific gain remains hard to find. What have we learned out of this manuscript? – That we are now entitled to do WRF-LES simulations in complex terrain with turbines? WRF simulations are done on a daily basis around the world, so how will this change what is done on a daily basis? One could have taken advantage of these great simulations to make a more robust objective/quantitative comparison between the simulations and the experimental data. For example, reading that there are wind differences observed at certain times of 1-2 m/s doesn’t really mean much. That could be a 50% difference in a weak mean wind, or a tiny % difference in a strong wind scenario,... Or that the flow looks similar or dissimilar here and there, has little scientific rigor. Once the rigurous/strict comparison of results done, then one could have an additional section with more scientific insight. Research questions that come to mind could be:

1. What is the effect of the surface conditions? – I want to understand that your WRF-LES is using IBM for the topography in regards to the momentum, but is it using the same approach for the ther- mal/moisture field? Based on what I was able to gather from the manuscript it seems that there are no surface conditions for temperature,... does this mean that your flow is insensitive to the surface con- ditions? This is strange since one of the driving mechanisms to thermal stratification is the ground,... but not much is discussed about the impacts or not of the surface conditions on the flow?

2. What is the effect of your turbulence initialization, and is it really worth it to run a suite of mesoscale WRF simulations just to provide time varying boundary conditions? Mesoscale WRF provides an ensemble flow solution that evolves with time, said otherwise, provides a more or less accurate mean flow and thermodynamic conditions of the region. However, the LES simulations provide an instan- taneous realization of the turbulent ABL, the question then could be posed as how does that compare to instead using profiles of experimental data to force the LES? Also, one can only wonder, what is the effect of the cell perturbation method to generate turbulence? Sure enough the mountains will spur some turbulence, but are the incoming turbulent flow conditions representative of all turbulent scales, including large scale perturbations?

3. An alternative, potentially the most interesting research question is how can one use the outcome of the rich simulations to understand flow configurations at other locations? Can results be scaled such that the results become generalizable in terms of stratification, mountain slope, terrain complexity? It would be a lot more interesting if the authors used the rich dataset to come up with generalizable relations that enabled one to extract conclusions at other locations without having to run expensive simulations,...

To this reviewer’s opinion this manuscript provides another missed opportunity by science and the com- munity in general to develop new critical advancements. Instead one could argue that this work is just designed to become another publication like the ones referred in the recently published article in the Atlantic magazine (https://www.theatlantic.com/science/archive/2021/05/xkcd-science-paper-meme-nails-academic- publishing/618810/). It is for all these reasons that at the present time this reviewer can only recommend major revisions, and can not recommend publication of the current manuscript until more scientific value is added to the manuscript.

More Specific Technical Comments

Line42; when talking about ‘length-scales’ one should clarify that these are "turbulent length-scales"

Line 53; it is important to clarify the difference in time scales between the LES and the WRF mesoscale simulations. LES provides an instantaneous realization of the flow at high-temporal frequency, while WRF mesoscale provides results representing the outcome of an ensemble of flow marching in time, but that per the ergodicity definition can not be interpreted at the same frequency than otherwise the LES inout/output.

In line 57; can the authors provide a brief description of the GAD model? Meaning, I don’t need to read the reference to find out whether the models is an actuator disk with/out rotation, etc.

In line 82; the authors mention that “the goal of this work is to model realistic atmospheric conditions and the associated turbulent flow phenomena to better un- derstand wind turbine wake propagation in complex terrain”. At the end of the manuscript, what have we learned about this that maybe of use as a function of thermal stratification, or in other locations, mountains, terrain, etc?

By the end of section 2.1; what about the surface conditions? If they were not measured (besides roughness) how are they taken care of in the simulations? Is z0,t taken as a fraction of z0?

In Figure 2; I could only wonder but what happens to the near surface heat and momentum fluxes? – It is well known in the community the existence of counter gradient heat fluxes, specially in complex terrain. Why not include subpanels with the near surface heat fluxes?

Around line 185; It seems like simulations are initialized with a uniform heat flux. However, it is known also that the heat flux will be rarely uniform in complex terrain. How is that potential effect assessed? Are we left to interpret that the results are indifferent to that surface forcing?

First line in Section 3.3; “Having demonstrated the ability...”; I dare say that to this point not much has been demonstrated besides showing two beautiful figures. Maybe the authors can tone this a bit down.

Line 210; the authors mention that “adequate turbulence is developed”, where is this shown? What is the premised/argument used to judge that? A certain amount of energy at certain scales? A tke scaling of k−5/3? A certain spectral comparison of tke with the field experiment at different heights? – Some of this might be more “objective”.

It is unclear how the Cell Perturbation Method generates different type of turbulence for the stable and unstable stratification.

From line 265 onwards, the text is riddled with subjective comparisons. I would suggest using percentages instead of absolute values when for example comparing wind speeds. Or avoid using comments like:“matches the measurements well”, without an actual metric of it. See line 311; "errors are small” in comparison to what?

Why is the wind speed and direction outputted at different frequen- cies?

Around line 367; the authors comment that the errors/discrepancies observed during the convective periods are larger than during the stable periods, but they don’t provide any comment, hypothesis, argument, detailed inspection, trying to investigate that in details,...

Figure17, great visualization for oral presentations, proposals or others where “candy-to-the-eye” is well accepted; but what is the use of such an image here? What is the intended message, outcome extracted? – I am not suggesting the authors remove the figure but instead include scientific argumentation around it.

In conclusion, I am convinced the data presented in this manuscript is of high quality, and has the potential to become of high value for the community if made publicly available. My only concern is that there is not enough scientific value at presenting the data itself, when additional analysis (some of which does not require extensive work) could be added that would increase the scientific value of the work.
Citation: https://doi.org/10.5194/wes-2021-36-RC2
- AC3: 'Reply on RC2', Adam Wise, 28 Jul 2021
  
  We thank the reviewer for the time spent reviewing our work and for the specific comments and suggestions. Please see our attached responses.
  
  Citation: https://doi.org/10.5194/wes-2021-36-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Adam Wise on behalf of the Authors (28 Jul 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (30 Jul 2021) by Johan Meyers

RR by Dries Allaerts (31 Aug 2021)

RR by Anonymous Referee #2 (01 Sep 2021)

RR by Johan Meyers (20 Sep 2021)

Suggestions for revision or reasons for rejection

The authors present a simulation study of the Perdigao field campign, using WRF with 5 nested domains, with the three finest being in LES mode. Comparisons with the field measurements are discussed for two selected cases, i.e. one stable and one convective. Overall, this simulation study can be of interest to the community. However, as it is presented now, I do not believe it should be accepted for publication. Results are poorly discussed, with a lot of hand waving and qualitative comparisons of instantaneous fields, the paper is overly long, and lacks detail in many instances (see further comments below). The author’s see the merit of their study mainly in the fact that they are the first to use realistic inflow conditions for LES of the Perdigao field campign, but that is not strictly true, since Wagner et al (Atmos. Chem. Phys, 2019) also presented a nested simulation, albeit with only three levels, with only the finest LES. Given all this, I recommend to thoroughly rewrite this paper, and better analyze results.
Comments
1. I do not see the point of the semi-idealized runs. It is obvious that this will not work with periodic BCs. There is also not much connection to the rest of the paper. This has probably been a useful internal verification exercise, but is not worthwhile reporting on. I suggest to remove this part.
2. Discussion is qualitative at best and seems often overly optimistic when reporting own results. For instance: “As seen in Fig. 12, the flow inside the valley and near the two ridges is highly dynamic but very accurately predicted by …” Such a statement can not be simply based on instantaneous snapshots such as shown in Figure 12. Overall, the paper is very lengthy in discussing a lot of these instantaneous figures (e.g. 6, 7, 11, 12, 16, 17). This part could be cut significantly
3. Although there are various mean-field results are shown, I would expect a more careful discussion, and in particular figures showing comparisons between 1h averages simulation and experiment along selected vertical and/or horizontal transects (= a line), including error bars (e.g. using bootstrapping). A side by side comparison of color plots in vertical planes is not really very informative (e.g. Fig. 10, 15, 18). Why are experimental results not added to Figure 19?
4. Figure 7 (lower panel), Figure 8, Figure 13 – this seems to be the long-time evolution of metmast measurements. How much of this is already contained in the d01 and d02 levels. In general, the authors could do a much better job discussing the added benefit of adding LES layers to the prediction quality of their results. E.g. can you quantify a significant reduction of bias per added layer, etc.
5. Discussion on Brunt-Vaisala frequency and gravity waves throughout the paper is very vague and imprecise. In particular: p6 – there are inversion waves and internal waves. Both are determined by a different Froude number (the first based on the inversion step, the second based on the lapse rate). Both relate to phenomena above the ABL. Please properly discuss and define, incl. height and way in which you measure the Fr numbers, velocity scale, length scale (also e.g. in Figure 3). P 12-13. Where/how do you measure the wavelength of the mountain wave event. This seems to be the horizontal wavelength. P 17. The horizontal wavelength of the wave is not influenced by the Froude number, but forced by the terrain feature, only the vertical wavelength is influenced by Fr number. P. 19 Discrepancy between vertical wavelength in model and measurements seems to be related to the fact that your simulation domain and Rayleigh damping layer is not sufficiently high. Properly discuss and justify your set-up or discuss limitations
6. page 8: why these particular cases. Better justify
7. page 12: why set the surface roughness to 0.5m. This seems rather arbitrary. Did you fit this value to obtain the best results?
8. page 13, line 290. This is a very qualitative statement. Also in earlier work (e.g. Berg et al), agreement was quite remarkable. Please try to be objective, and focus on verifiable facts
9. Eq. 3, 4: better explain that these eqs are only valid for the surface layer. Where, at which height to do you measure them?

Hide

ED: Reconsider after major revisions (20 Sep 2021) by Johan Meyers

AR by Adam Wise on behalf of the Authors (21 Nov 2021) Author's response Author's tracked changes Manuscript

ED: Publish subject to technical corrections (24 Dec 2021) by Johan Meyers

ED: Publish subject to technical corrections (29 Dec 2021) by Gerard J.W. van Bussel (Chief editor)

AR by Adam Wise on behalf of the Authors (11 Jan 2022) Author's response Manuscript

Short summary

Wind turbine wake behavior in hilly terrain depends on various atmospheric conditions. We modeled a wind turbine located on top of a ridge in Portugal during typical nighttime and daytime atmospheric conditions and validated these model results with observational data. During nighttime conditions, the wake deflected downwards following the terrain. During daytime conditions, the wake deflected upwards. These results can provide insight into wind turbine siting and operation in hilly regions.