Review of the manuscript “Estimating Long-Term Annual Energy Production of a Large Offshore Wind Farm from Large-Eddy Simulations: Methods and Validation with a 10-Year Simulation” by Bernard Postema, Remco Verzijlbergh, Pim Van Dorp, Peter Baas, and Harm Jonker submitted for publication in Wind Energy Science.

 In the manuscript “Estimating Long-Term Annual Energy Production of a Large Offshore Wind Farm from Large-Eddy Simulations: Methods and Validation with a 10-Year Simulation” the authors present a numerical study focused on estimating long-term wind power production of an offshore wind farm based on mesoscale simulations and limited high-resolution, large-eddy simulations (LES).

General Remarks

While the numerical study presented in the manuscript is unique because it uses a long-term (10-year) LES, the authors did not present a convincing motivation and justification for their numerical study that uses coarse LESs to estimate offshore wind speed and simulate wind farm power production by representing operating wind turbines with an actuator disk model. It is not clear that a coarse LES can deliver any benefit in estimating hub height or rotor equivalent wind speed compared to significantly less costly mesoscale simulations. The LES at resolution of 120 m does not improve wind speed prediction compared to mesoscale simulations (> 1 km) or even ERA5 reanalysis as shown in the manuscript. Coarse LES does not resolve turbulent eddies in the inertial range considering: offshore conditions, effective resolution, and the need for inflow turbulence development not described in the manuscript.

The study presents a Bayesian downscaling approach that combines LES and reanalysis output. While this approach is interesting and the use of LES can be considered as something that has not been explored before, similar methodologies have been used extensively for regional climate downscaling (e.g., Holthuijzen et al., 2021, J. App. Meteorolo. Climat. and references therein). The authors did not provide any references to research that preceded their study, so based on the way the approach is presented, it would seem that this approach is completely novel. The novel part is exploring three scenarios for long-term correction, and this is possibly the most relevant contribution.

The study uses an actuator disk model that is not appropriate for the coarse LES resolution used in the study, i.e. 120 m in horizontal and ~30 m in vertical taking, with the effective resolution that is even coarser. For example, Calaf et al. (2010, also Meyers and Meneveau, 2010, cited in the paper) resolved the rotor with ~10 vertical grid cells. Bass et al. (2023) conducted a sensitivity study with grid cell size of 120 m and 60 m and similar vertical grid as present study and argued that the 120 m grid cell size is adequate, however, their estimates of aerodynamic losses were about 20% larger with coarser resolution – this is significant. In addition, the actuator disk model was implemented in the ASPIRE LES model, a version of a commercial, GPU-based model, which is not publicly available. It is not clear how ASPIRE differs from other versions of the GPU model (e.g., GRASP used in Bass et al., 2023, study) and if ASPIRE was validated in any way.

Although, to achieve greater realism and therefore relevance, the numerical study includes coupling between coarse reanalysis and LES simulated is a hypothetical wind farm which means that the results of the study cannot be validated. The use of a hypothetical wind farm is of a very limited value considering that it is now possible to obtain wind power production data for existing wind farms in the North Sea (see, e.g., https://rave-offshore.de/en/data.html).

Taking all the above in the consideration I do not recommend the manuscript “Estimating Long-Term Annual Energy Production of a Large Offshore Wind Farm from Large-Eddy Simulations: Methods and Validation with a 10-Year Simulation” in Wind Energy Science.

Specific Remarks

• Abstract – The abstract is confusing and poorly written, e.g., no motivation for the use of LES is provided.
• Line 28 – It is stated that “’real- weather’ LES has been demonstrated to be viable… to explicitly model the interactions between wind farms and the atmosphere (Baas et al., 2023)” however the cited study by Bass et al. (2023) does not include a validation that would demonstrate viability.
• Line 68 – ASPIRE model is used in the study and a reference is made to the Dutch Atmospheric Large Eddy Simulation (DALES) model and the GPU version, however, there are no previous references to the ASPIRE model and its validation.
• Line 119 – It is stated that “In the practice of LES modelling, it is often found that the wind speed displays a mean bias of O(0.1 m s−1)…” but no reference is provided for this statement.
• Line 121 – The sentence: “Such a free stream run has no turbines included, or it has turbines that exert no force on the flow, and therefore leave it undisturbed.” can be removed since is nonsensical. Turbines that exert no force on the flow are not turbines, since even turbines at rest exert force on the flow.
• Line 131 - It is stated that “…because of explicit representation of all fluid dynamical… effects…” First, it is not true that all fluid dynamical effects are represented since the rotation of the turbine is not accounted for, and second, at the coarse resolution used in the study it cannot be said that the fluid dynamical effects are represented well.
• Line 160 – How is ASPIRE different from or related to GRASP. Has it been validated?
• Line 164 – Given the resolution, it is not clear how turbulence develops on the inner LES domain.
• Line 165 – It is stated that “…the core LES domain is nested in a coarser mesoscale-type simulation with a resolution of 1.5 km,…” and “This coarser simulation has the same model formulation as the LES…,” however, it is not clear what is the exact model formulation. It is only mentioned that the turbulence parameterization is different, but not how is it different, what other parameterizations were used: radiative transfer, microphysics, land surface, etc. For the study to be reproducible this information should be provided.
• Line 185 – If turbines do not exert any force they are not included in the simulation, so there is no need to mention turbine.
• Line 221 (also Firgure 2) – It is not clear what is the benefit of LES offshore when it results in larger errors and turbulence is not resolved with 120 m grid cell size.
• Line 230 – As shown by Baas et al. (2023) a coarse LES result in overestimation of aerodynamic losses.

Review of “Estimating Long-term annual energy production of large offshore wind farm from large-eddy simulations: methods and validation with a 10-year simulation.”

The authors present an approach for a long-term correction of the power production of a hypothetical wind farm. They use a ten-year long timeseries of modelled power production as their verification data and test how different sampling methods together with a joint probability density between wind farm production coming from an LES and the ERA5 wind speed can approximate the long-term mean. They further include observation data and MCP to correct for the wind speed bias of the simulation.

The topic is relevant to the application of computationally expensive higher-fidelity modelling for long-term energy yield assessment and the general approach is interesting.

However, my main problem with the manuscript is mentioned by the authors themselves in l.325: [the methods] are not limited to LES but can also be applied to other (wake) models, numerical weather prediction models, or observation data.

The manuscript is focussing way too much on the method how the joint probability density between wind farm power and wind speed is created by LES, while this has basically no impact on the applicability of the method. The wind farm power can in principle be replaced by any other (meteorological) variable that has a dependency on wind.

The joint probability density between wind farm power and wind speed represents a probabilistic power curve of the wind farm. It might be the case, that the LES used in this study represents the best method to create this power curve, but this is not the topic of this manuscript as also mentioned by the authors in the introduction. Furthermore, by making the wind farm power curve only dependent on the wind speed the authors ignore other dependencies, for example on the wind direction and other meteorological variables. The approach thus might work for this particular wind farm, but most real wind farms usually show at least a dependency of their efficiency on the wind direction due to their geometry or turbine spacing.

I would only recommend a publication in WES if the authors address the following main requests:

• Title and scope of the manuscript should be changed to be centred around the methodology of the long-term correction. The method to create the wind farm power curve should be clearly defined as an example application. This would make it also more robust against any criticism about the coarse LES methodology (because this is not the topic of the paper)
• The generalisability of the approach to irregular layouts should be explored or well explained. If this is not possible, it should be clearly mentioned that the manuscript only explores this specific case of a regularly-layouted wind farm