Two-way coupling of gravity waves and wind farm wakes: a reduced-order boundary-layer model

arXiv:2605.30077 · physics.flu-dyn · Submitted 2026-05-28 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Two-way coupling of gravity waves and wind farm wakes".

Mira: Here's the gist of the paper. This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms because wake evolution and gravity-wave dynamics constitute a single coupled problem.

Kai: Model Development The framework starts with the non-hydrostatic Boussinesq equations, which are simplified to yield separate governing equations for the boundary layer and the overlying stratified free atmosphere.

Mira: These regions are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow.

Kai: The model employs a mixed pseudo-spectral finite-difference discretisation to be computationally efficient while retaining vertical boundary-layer structure.

Paper summary: Kai: So, looking at this paper, "Two-way coupling of gravity waves and wind farm wakes: a reduced-order boundary-layer model," it’s about developing a framework that models the two way coupling between gravity waves and turbulent wakes in large scale wind farms.

Mira: The main claim is that starting from the nonhydrostatic Boussinesq equations, simplifying them for the boundary layer and free atmosphere yields separate governing equations coupled by a dynamic boundary condition at the capping inversion.

Lev: From an engineering standpoint, what this means is they’ve created a mid fidelity tool that’s computationally efficient compared to high fidelity simulations but still captures important farm scale flow physics.

Kai: They show good agreement with les data in both internal wind farm flow and large scale gravity wave response, capturing things like upstream blockage and wake recovery driven by favourable pressure gradients.

Mira: The authors suggest this approach is suitable for parameterizing wind farm effects in weather and climate models because its computational cost is far below that of high fidelity simulation, which is ideal for many evaluations.

Lev: They are focused on using this single framework to bridge the gap between physics based wake and gravity wave coupling with a computational cost far below high fidelity simulation, which makes it well suited for applications that require many evaluations.

Kai: The paper’s limitations are that they didn't include Coriolis force and boundary layer stratification, so things like wind veer and the supergeostrophic jet near the inversion aren't represented in this version.

Mira: So, overall, this work provides a single framework for understanding how these phenomena interact across scales when you need a computationally efficient way to get that information.

Conclusion: Kai: So, this paper is about modeling how gravity waves and wind farm wakes talk to each other simultaneously using a simplified numerical setup.

Mira: Yeah, they took these nonhydrostatic equations and split them into separate problems for the boundary layer and the free atmosphere, linking them at the capping inversion layer.

Lev: From a hardware standpoint, that means they’ve built a framework that’s fast enough to run on real systems while keeping most of the physics intact.

Kai: They show good results when comparing their model to high-fidelity simulations for both the inside of the farm and those big gravity wave effects.

Mira: What's interesting is how they capture things like that upstream blockage and how wakes recover faster because of favorable pressure gradients downstream.

Lev: That kind of coupling in one framework, with a computational cost much lower than a full simulation, makes it useful for running lots of tests on weather models.

Kai: The authors say this gives us a mid-fidelity tool to bridge the gap between super expensive simulations and simpler engineering models.

Mira: They're focusing on how this can be used for parameterizing wind farm effects in climate models.

Lev: The model uses just one free tuning constant, ck, which they keep fixed across all their different tests.

Kai: It’s a neat way to get good agreement with the high-end LES data even with this reduced-order approach.

Mira: They reproduce the row by row structure of the velocity deficit inside the farm and that shift in where maximum inversion displacement happens under different wave conditions.

Lev: But they did admit a simplification, like clamping that correction factor beta n to a minimum of one for certain flows, which is important for real hardware testing because it cuts down on unnecessary complexity.

Kai: So this framework gives us a way to predict how these two big flow phenomena interact without needing the massive computational power of those high-fidelity runs.

Mira: The authors point out that while they did keep things like Coriolis force out, and boundary layer stratification, the core idea is coupling wake dynamics with gravity waves efficiently.

Lev: It’s a solid starting point for parameterizing wind farm effects where you need many evaluations of different scenarios.

Kai: We're going to look at how those specific results compare across different atmospheric regimes next.

Department of Mathematical Sciences, Durham University · Department of Engineering, Durham University

physics.flu-dyn

Submitted: 2026-05-28

Updated: 2026-11-12

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 77/100

The gist: This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms because wake evolution and gravity-wave…

Key concepts

Non-hydrostatic Boussinesq equations
These are governing equations used to describe fluid flow where density variations are small but vertical motion is significant. They simplify the complex physics by separating the flow into a boundary layer near the ground and a free atmosphere above, allowing for separate modeling.
Two-way coupling
The model treats gravity waves and turbulent wakes as interacting systems. Gravity waves affect the boundary layer flow, which in turn influences how wind farm wakes evolve. This feedback loop is captured by linking the equations governing both regions at the capping inversion.
Wake correction factor ($eta_n$)
This factor adjusts the calculated wind farm forcing based on upstream row positions. It accounts for how the local wake deficit changes due to neighboring turbines, allowing for a more accurate representation of flow interaction within the farm.

Terminology

Summary

This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms because wake evolution and gravity-wave dynamics constitute a single coupled problem. p. 0

Where the paper says it

bastankhah@durham. uk Received: XX 2020; Revised: XX XX 2020; Accepted: XX XX 2020 Keywords: wind farm wakes; gravity waves, reduced-order numerical model Abstract This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms.

Model Development

The framework starts with the non-hydrostatic Boussinesq equations, which are simplified to yield separate governing equations for the boundary layer and the overlying stratified free atmosphere. These regions are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow. p. 0

Where the paper says it

Starting with the non-hydrostatic Boussinesq equations and introducing simplifications appropriate to the boundary layer and the overlying stratified free atmosphere yield separate governing equations for the two regions. These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow.

The model employs a mixed pseudo-spectral finite-difference discretisation to be computationally efficient while retaining vertical boundary-layer structure. p. 0
Where the paper says it

These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow. A mixed pseudo-spectral finite-difference discretisation yields a computationally efficient model while retaining vertical boundary-layer structure.

Governing Equations and Coupling

The governing equations retain the full nonhydrostatic momentum equations in both regions, meaning the horizontal velocity, vertical velocity and pressure remain dynamically coupled throughout both regions. In the boundary layer, eliminating pressure and horizontal velocity reduces the system to a single equation for the vertical velocity, which retains the continuous vertical structure of the boundary layer. p. 8

Where the paper says it

Allaerts and Meyers, 2019]. As mentioned earlier, in these models, the vertical velocity is not dynamically coupled to the pressure and horizontal velocity fields within the boundary layer and therefore does not enter the interface coupling in the same manner as in the present formulation.

The free atmosphere is described by a Helmholtz equation whose solution is selected by the radiation condition. p. 7
Where the paper says it

After eliminating 𝑢, 𝑝 and 𝑏 in favour of 𝑤 in (16), we obtain 𝜕 2𝑤 𝜕𝑥2 + 𝜕 2𝑤 𝜕𝑧2 = −(𝑁 2 /𝑈 2 H ) 𝑤, (17) Taking the Fourier transform of (17) with respect to 𝑥 (denoted by hat), the well-known Helmholtz equation used in mountain-wave theory [Gill, 2016] can be obtained: 𝜕 2𝑤ˆ (𝑘, 𝑧) 𝜕𝑧2 + 𝑚 2𝑤ˆ (𝑘, 𝑧) = 0, with 𝑚 2 = 𝑁 2 /𝑈 2 H − 𝑘 2 , (18) where 𝑘 is the horizontal wavenumber. The solution is 𝑤ˆ (𝑘, 𝑧) = 𝑤ˆ + (𝑘, 𝐻) e 𝑖𝑚(𝑧−𝐻) , 𝑚 =    𝑖 √︃ 𝑘 2 − 𝑁2/𝑈2 H , |𝑘 | ≥ 𝑁/𝑈H sgn(𝑘) √︃ 𝑁2/𝑈2 H − 𝑘 2 , |𝑘 | < 𝑁/𝑈H.

The two regions are joined at the capping inversion through kinematic and dynamic conditions. p. 8
Where the paper says it

The ABL requires an upper boundary condition at 𝑧 = 𝐻, while the free-atmospheric solution depends on the unknown interfacial amplitude 𝑤ˆ + (𝑘, 𝐻). These two missing ingredients are supplied by coupling the two regions across the capping inversion through kinematic and dynamic conditions.

Wind Farm Forcing

The turbine forcing enters the governing equations as a momentum sink acting on the resolved flow field, obtained by laterally averaging a three-dimensional actuator-disc representation of the farm. The lateral average of the three-dimensional turbine force becomes the forcing term f in equation (10). p. 4

Where the paper says it

The lateral heterogeneity of the flow is nonetheless retained in the turbine forcing introduced later in this section. 2 are therefore understood as laterally averaged quantities, with the angle brackets omitted for brevity in other parts of this paper, and the forcing 𝑓 is the lateral average of the three-dimensional turbine force derived below.

A crucial step involves estimating a correction factor βn that depends only on the farm layout, which is determined solely by the relative positions of the upstream rows. p. 6
Where the paper says it

∑︁𝑛−1 𝑗=1 𝑋 −1 𝑗𝑛 . (15) According to (15), 𝛽𝑛 is determined solely by the relative positions of the upstream rows: the lateral offset 𝑦𝑛 − 𝑦 𝑗 sets how much of each upstream wake is felt at the rotor, while the weight 𝑋 −1 𝑗𝑛 ensures that nearer rows contribute more strongly.

Results and Validation

The model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects when compared with Large Eddy Simulations (LES). The model captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients. p. 0

Where the paper says it

Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects. It captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients.

Quantitative comparisons with LES data show good agreement in both the internal wind-farm flow and the large-scale gravity-wave response. p. 0
Where the paper says it

For the semi-infinite wind farm considered here, a complete solution is obtained in a matter of seconds on a standard laptop. Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects.

The model captures the shift in the location of maximum inversion displacement between supercritical and subcritical conditions. p. 0
Where the paper says it

These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow. A mixed pseudo-spectral finite-difference discretisation yields a computationally efficient model while retaining vertical boundary-layer structure.

Conclusions and Outlook

The developed numerical framework provides a mid-fidelity tool for bridging the gap between computationally expensive high-fidelity simulations and highly parametrised engineering models. Its computational efficiency makes it suitable for parameterisations of wind-farm effects in weather and climate models. p. 0

Where the paper says it

The developed numerical framework provides a mid-fidelity tool for bridging the gap between computationally expensive high-fidelity simulations and highly parametrised engineering models of wind-farm–atmosphere interaction. More broadly, its computational efficiency and physics-based representation of wake and gravitywave coupling make it a suitable candidate for parameterisations of wind-farm effects in weather and climate models, while also enabling efficient assessment of wind-farm performance and farm–farm interactions.

The combination of physics-based wake and gravity-wave coupling in one single framework with a computational cost far below that of high-fidelity simulation makes this approach well suited to applications that require many evaluations. p. 0
Where the paper says it

The developed numerical framework provides a mid-fidelity tool for bridging the gap between computationally expensive high-fidelity simulations and highly parametrised engineering models of wind-farm–atmosphere interaction. More broadly, its computational efficiency and physics-based representation of wake and gravitywave coupling make it a suitable candidate for parameterisations of wind-farm effects in weather and climate models, while also enabling efficient assessment of wind-farm performance and farm–farm interactions.

The model introduces no free parameters aside from the tuning constant ck entering the layout correction. p. 6
Where the paper says it

Flow E1-7 where 𝜗3 [𝜁, 𝑞] = 1 + 2 Í∞ 𝑙=1 𝑞 𝑙 2 cos(2𝑙𝜁) is the Jacobi theta function. The constant 𝑐𝑘 is a tuning parameter and a value of 𝑐𝑘 = 1.

It is important to note a crude simplification when we applied βn in (11) to the full perturbation velocity. p. 6
Where the paper says it

(15) According to (15), 𝛽𝑛 is determined solely by the relative positions of the upstream rows: the lateral offset 𝑦𝑛 − 𝑦 𝑗 sets how much of each upstream wake is felt at the rotor, while the weight 𝑋 −1 𝑗𝑛 ensures that nearer rows contribute more strongly. It is important to note a crude simplification when we applied 𝛽𝑛 in (11) to the full perturbation velocity.

p. 6
Where the paper says it

(15) According to (15), 𝛽𝑛 is determined solely by the relative positions of the upstream rows: the lateral offset 𝑦𝑛 − 𝑦 𝑗 sets how much of each upstream wake is felt at the rotor, while the weight 𝑋 −1 𝑗𝑛 ensures that nearer rows contribute more strongly. It is important to note a crude simplification when we applied 𝛽𝑛 in (11) to the full perturbation velocity.

The farm is taken to be semi-infinite in the lateral direction, so that the flow is solved in two dimensions. p. 2
Where the paper says it

The feedback of gravity waves on the ABL flow is captured through the boundary condition imposed at the top of the ABL. Here, for simplicity we focus on a semi-infinite wind farm (finite in the streamwise direction and infinite in the lateral direction), but the developed numerical approach can be extended to finite wind farms in future works.

Extending the framework to finite farms and incorporating these effects are the natural next steps. p. 2
Where the paper says it

The feedback of gravity waves on the ABL flow is captured through the boundary condition imposed at the top of the ABL. Here, for simplicity we focus on a semi-infinite wind farm (finite in the streamwise direction and infinite in the lateral direction), but the developed numerical approach can be extended to finite wind farms in future works.

The eddy viscosity is evaluated from the background shear alone, and may therefore under-represent the enhanced mixing generated within densely spaced farms. p. 4
Where the paper says it

Retaining its full form would arguably be more realistic, since it lets the eddy viscosity respond to the perturbed shear, but it makes 𝜏 quadratic in the perturbation and complicates the solution procedure. We therefore adopt the linear form, while noting that it may under-represent the enhanced mixing within dense farms.

The Coriolis force and boundary-layer stratification are not included, so that wind veer and the supergeostrophic jet near the inversion are not represented. p. 1
Where the paper says it

A key driver of these phenomena is the thermal stratification within the atmosphere. Even when the boundary layer itself is neutral, stratification is typically present above the ABL in the form of an abrupt temperature jump at the capping inversion and a more gradual variation in the free atmosphere above (figure 1).

The paper is 2605. p. 0
Where the paper says it

0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited. arXiv:2605.

30077. The two are connected by a lateral average, which we introduce here since it also determines the form of the forcing term f appearing in (1a). p. 4
Where the paper says it

2 describe a two-dimensional flow in the (𝑥, 𝑧) plane, whereas the flow through a wind farm is inherently three-dimensional. The two are connected by a lateral average, which we introduce here since it also determines the form of the forcing term 𝑓 appearing in (1a).

The steady two-dimensional non-hydrostatic Boussinesq equations yields U(z)∂u/∂x + u∂u/∂x + wU' + w∂u/∂z = −∂p/∂x + ∂τ/∂z + f, (1a). The presence of wind farm leads to a displacement of the inversion layer denoted by η (shown schematically in figure 1). p. 2

Where the paper says it

We assume that (1) holds in both the ABL and the lower free atmosphere. The presence of wind farm leads to a displacement of the inversion layer denoted by 𝜂 (shown schematically in figure 1).

Considering small η, we approximate the ABL with a rectangular domain capped at z = H. p. 2
Where the paper says it

The presence of wind farm leads to a displacement of the inversion layer denoted by 𝜂 (shown schematically in figure 1). Considering small 𝜂, we approximate the ABL with a rectangular domain capped at 𝑧 = 𝐻.

However, we take the dynamical effects of this displacement into account leading to the coupling of the ABL and free atmosphere (discussed later in §2. p. 3
Where the paper says it

E1-4 Hossein A. Kafiabad and Majid Bastankhah displacement into account leading to the coupling of the ABL and free atmosphere (discussed later in §2.

5).

Each turbine is represented as an actuator disc of zero streamwise thickness, uniformly loaded over its frontal area, so that rotation and the radial variation of the loading are not represented. The total thrust of a single turbine of the n th row is Tn = 1/2 ρ0 C'T,n U 2 n nπR squared, where U n is the local disc-averaged velocity of WT ns (i.e., turbines in the nt h row). p. 4

Where the paper says it

For a semiinfinite wind farm with 𝑁𝑟 rows of turbines, let 𝑛 ∈ {1, 2, . , 𝑁𝑟 } denote the row number increasing in the streamwise direction, 𝑥𝑛 the streamwise location of the 𝑛 th row, 𝑧ℎ the hub height, and 𝑦𝑛 + 𝑞𝑆𝑦 with the integer 𝑞 labelling turbines within the row.

Distributing this force uniformly over the disc area πR squared gives the three-dimensional force per unit mass exerted by the farm, F(x, y, z) = −1/2 ∑ Nr n=1 C'T,n U 2 n δ(x - xn) ∑ q=-∞ H R − r n,q, r n,q = √[(y - yn - qSy) squared + (z - zh) 2] (9). By averaging (9) laterally based on (8), the laterally averaged force per unit mass exerted by the farm f becomes f = ⟨F⟩ = −1/Sy ∑ Nr n=1 C'T,n U 2 n δ(x - xn) √[R squared − (z - zh) 2], z - zh < R, 0, z - zh ≥ R (10).

--- Page 6 ---

E4-4 Wind farm forcing

The vertical profile of forcing in (10) is therefore simply the half-chord of the rotor disc scaled by the lateral inter-turbine spacing, i.e. the local solidity of the turbine row. p. 5

Where the paper says it

By averaging (9) laterally based on (8), the laterally averaged force per unit mass exerted by the farm 𝑓 becomes 𝑓 = ⟨F ⟩ =    − 1 𝑆𝑦 ∑︁ 𝑁𝑟 𝑛=1 𝐶 ′ 𝑇,𝑛 𝑈 2 𝑛 𝛿(𝑥 − 𝑥𝑛) √︁ 𝑅2 − (𝑧 − 𝑧ℎ) 2 , |𝑧 − 𝑧ℎ | < 𝑅, 0, |𝑧 − 𝑧ℎ | ≥ 𝑅, (10) The vertical profile of forcing in (10) is therefore simply the half-chord of the rotor disc scaled by the lateral inter-turbine spacing, i. the local solidity of the turbine row: the force is largest at hub height and tapers to zero at the rotor tips.

For numerical implementation, the forcing term is smoothed in both the streamwise and vertical directions using a Gaussian kernel. p. 5
Where the paper says it

the local solidity of the turbine row: the force is largest at hub height and tapers to zero at the rotor tips. For numerical implementation, the forcing term is smoothed in both the streamwise and vertical directions using a Gaussian kerne Since the governing equations are solved for the laterally averaged perturbation field 𝑢(𝑥, 𝑧), they carry by construction no information about the lateral variation of the flow within the farm.

Since the governing equations are solved for the laterally averaged perturbation field u(x, z), they carry by construction no information about the lateral variation of the flow within the farm. p. 5
Where the paper says it

the local solidity of the turbine row: the force is largest at hub height and tapers to zero at the rotor tips. For numerical implementation, the forcing term is smoothed in both the streamwise and vertical directions using a Gaussian kerne Since the governing equations are solved for the laterally averaged perturbation field 𝑢(𝑥, 𝑧), they carry by construction no information about the lateral variation of the flow within the farm.

The thrust force in (10), however, depends on the velocity actually incident on the rotors, which differs from its laterally averaged counterpart. p. 5
Where the paper says it

For numerical implementation, the forcing term is smoothed in both the streamwise and vertical directions using a Gaussian kerne Since the governing equations are solved for the laterally averaged perturbation field 𝑢(𝑥, 𝑧), they carry by construction no information about the lateral variation of the flow within the farm. The thrust force in (10), however, depends on the velocity actually incident on the rotors, which differs from its laterally averaged counterpart.

We therefore write U n = Uh + βn u(xn, zh), where Uh = U(zh) and βn is the ratio between the local wake deficit induced by upstream turbines and its laterally averaged value at the same streamwise location. p. 5
Where the paper says it

Without a relation linking the two, the model would incorrectly return identical predictions for the two layouts. We therefore write 𝑈𝑛 = 𝑈ℎ + 𝛽𝑛 𝑢(𝑥𝑛, 𝑧ℎ), (11) where 𝑈ℎ = 𝑈(𝑧ℎ) and 𝛽𝑛 is the ratio between the local wake deficit induced by upstream turbines and its laterally averaged value at the same streamwise location.

Where the gains come from

The correction factor βn is determined solely by the relative positions of the upstream rows. It is important to note a crude simplification when we applied βn in (11) to the full perturbation velocity. The flow perturbation associated with local blockage, or with the global blockage induced by gravity waves do not need this correction. p. 6

Where the paper says it

In reality, only the component of velocity perturbation induced by upstream wakes requires this correction. The flow perturbation associated with local blockage, or with the global blockage induced by gravity waves do not need this correction.

We therefore clamp βn to a minimum of one, so that these contributions are not spuriously reduced, and the first row, unaffected by upstream wakes, likewise takes β1 = 1. p. 6
Where the paper says it

The flow perturbation associated with local blockage, or with the global blockage induced by gravity waves do not need this correction. We therefore clamp 𝛽𝑛 to a minimum of one, so that these contributions are not spuriously reduced, and the first row, unaffected by upstream wakes, likewise takes 𝛽1 = 1.

Pruning redundant reasoning tokens

The model relies on a single free parameter, c k discussed in §2.3, whose value is kept fixed across all presented results in this paper. p. 6

Where the paper says it

The constant 𝑐𝑘 is a tuning parameter and a value of 𝑐𝑘 = 1. 75 provides acceptable predictions against the three LES datasets presented in §3.

Comparisons with three LES datasets spanning different atmospheric regimes show good agreement in both the internal wind-farm flow and the large-scale gravity-wave response. p. 0
Where the paper says it

For the semi-infinite wind farm considered here, a complete solution is obtained in a matter of seconds on a standard laptop. Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects.

The model reproduces the row-by-row structure of the velocity deficit within the farm, the upstream blockage associated with the adverse pressure gradient ahead of the farm, and the accelerated wake recovery driven by the favourable pressure gradient within and downstream of it. p. 0
Where the paper says it

Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects. It captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients.

It also captures the shift in the location of maximum inversion displacement between supercritical and subcritical conditions. p. 0
Where the paper says it

Starting with the non-hydrostatic Boussinesq equations and introducing simplifications appropriate to the boundary layer and the overlying stratified free atmosphere yield separate governing equations for the two regions. These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow.

p. 0
Where the paper says it

Starting with the non-hydrostatic Boussinesq equations and introducing simplifications appropriate to the boundary layer and the overlying stratified free atmosphere yield separate governing equations for the two regions. These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow.

p. 0
Where the paper says it

Starting with the non-hydrostatic Boussinesq equations and introducing simplifications appropriate to the boundary layer and the overlying stratified free atmosphere yield separate governing equations for the two regions. These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow.

--- Page 9 ---

E5-5 Comparison of normalised velocity perturbation at hub height with the LES data of Zhu et al. [2025]. (b) Comparison of normalised velocity perturbation vertically averaged over z = (0, 2zh) with the LES data reported in Allaerts and Meyers [2019]. p. 3

Where the paper says it

Allaerts and Meyers, 2019, Devesse et al. , 2024], which incorporated non-hydrostatic effects through the freeatmospheric wave dispersion relation while retaining the vertically-averaged hydrostatic description of the ABL.

Overall, the model shows very good agreement with the LES data given the simplicity of the reduced-order framework. p. 0
Where the paper says it

bastankhah@durham. uk Received: XX 2020; Revised: XX XX 2020; Accepted: XX XX 2020 Keywords: wind farm wakes; gravity waves, reduced-order numerical model Abstract This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms.

While Q00 experiences a considerably stronger blockage in the upwind region, its wake recovers faster than S1 due to the strong favourable pressure gradient created by gravity waves downstream as qualitatively discussed earlier in figure 4. In the strongly stratified Q00 case, the model overpredicts farm-wake recovery, resulting in a slightly positive u. p. 0
Where the paper says it

Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects. It captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients.

--- Page 12 ---

E6-6 Row number, n

Row Number Analysis

The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching β2 ≈ 4 for an aligned farm with Sx = 5D. p. 6

Where the paper says it

5, respectively, and by the periodicity of 𝜗3 these two limits bracket all possible layouts of this type. The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching 𝛽2 ≈ 4 for an aligned farm with 𝑆𝑥 = 5𝐷, and decays further downstream as the flow within the farm becomes more homogenous.

p. 6
Where the paper says it

5, respectively, and by the periodicity of 𝜗3 these two limits bracket all possible layouts of this type. The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching 𝛽2 ≈ 4 for an aligned farm with 𝑆𝑥 = 5𝐷, and decays further downstream as the flow within the farm becomes more homogenous.

The value of β also reduces with increasing streamwise spacing, since the wakes reaching a given row have then travelled farther. p. 6
Where the paper says it

The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching 𝛽2 ≈ 4 for an aligned farm with 𝑆𝑥 = 5𝐷, and decays further downstream as the flow within the farm becomes more homogenous. The value of 𝛽 also reduces with increasing streamwise spacing, since the wakes reaching a given row have then travelled farther.

The correction is confined to a comparatively narrow band of near-aligned layouts, Δy/Sy ≲ 0. p. 6
Where the paper says it

The value of 𝛽 also reduces with increasing streamwise spacing, since the wakes reaching a given row have then travelled farther. The correction is confined to a comparatively narrow band of near-aligned layouts, Δ𝑦/𝑆𝑦 ≲ 0.

13 for Sx = 5D.

The model reproduces the row-by-row structure of the velocity deficit within the farm, the upstream blockage associated with the adverse pressure gradient ahead of the farm, and the accelerated wake recovery driven by the favourable pressure gradient within and downstream of it. It also captures the shift in the location of maximum inversion displacement between supercritical and subcritical conditions.

--- Page 13 ---

E7-7 Figure 5(a) shows streamwise variations of u/U h at the hub height for both our reduced-order model and the LES data of Zhu et al. [2025], which we refer to as the JH case in Table 1. p. 1

Where the paper says it

This structure is called a conventionally neutral boundary layer (CNBL). In this case, the wind farm’s interaction with the overlying stratified layers generates large-scale gravity waves.

Within the farm the model reproduces the LES well, capturing both the row-by-row structure of the deficit and its cumulative growth along the farm. p. 0
Where the paper says it

Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects. It captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients.

Upstream of the farm, however, the model predicts a deceleration that extends considerably further than that seen in the LES. p. 5
Where the paper says it

The thrust force in (10), however, depends on the velocity actually incident on the rotors, which differs from its laterally averaged counterpart. For example, in an aligned farm the turbines operate in the wakes of the upstream rows and therefore experience a deficit considerably larger than the laterally averaged one, whereas in a staggered farm the two (i.

This difference is likely due to the streamwise extent of the LES domain, which reaches only about 14D upstream of the first turbine row, whereas resolving the global blockage response requires domains extending much further ahead of the farm.

The model reproduces the row-by-row structure of the velocity deficit within the farm, the upstream blockage associated with the adverse pressure gradient ahead of the farm, and the accelerated wake recovery driven by the favourable pressure gradient within and downstream of it. It also captures the shift in the location of maximum inversion displacement between supercritical and subcritical conditions.

--- Page 16 ---

E8-8 Figure 2 shows the variation of βn for farms in which each row is displaced laterally by a constant offset Δy relative to the row immediately upstream, so that yn - yj = (n - j)Δy. Aligned and staggered farms correspond to the limiting cases Δy/Sy = 0 and Δy/Sy = 0. p. 6

Where the paper says it

Figure 2 shows the variation of 𝛽𝑛 for farms in which each row is displaced laterally by a constant offset Δ𝑦 relative to the row immediately upstream, so that 𝑦𝑛 − 𝑦 𝑗 = (𝑛 − 𝑗)Δ𝑦. Aligned and staggered farms correspond to the limiting cases Δ𝑦/𝑆𝑦 = 0 and Δ𝑦/𝑆𝑦 = 0.

p. 6
Where the paper says it

Figure 2 shows the variation of 𝛽𝑛 for farms in which each row is displaced laterally by a constant offset Δ𝑦 relative to the row immediately upstream, so that 𝑦𝑛 − 𝑦 𝑗 = (𝑛 − 𝑗)Δ𝑦. Aligned and staggered farms correspond to the limiting cases Δ𝑦/𝑆𝑦 = 0 and Δ𝑦/𝑆𝑦 = 0.

5, respectively, and by the periodicity of θ3 these two limits bracket all possible layouts of this type. p. 6
Where the paper says it

Aligned and staggered farms correspond to the limiting cases Δ𝑦/𝑆𝑦 = 0 and Δ𝑦/𝑆𝑦 = 0. 5, respectively, and by the periodicity of 𝜗3 these two limits bracket all possible layouts of this type.

p. 6
Where the paper says it

Aligned and staggered farms correspond to the limiting cases Δ𝑦/𝑆𝑦 = 0 and Δ𝑦/𝑆𝑦 = 0. 5, respectively, and by the periodicity of 𝜗3 these two limits bracket all possible layouts of this type.

The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching β2 ≈ 4 for an aligned farm with Sx = 5D.

--- Page 17 ---

E9-9 A. Time-dependent free-atmospheric waves and radiation condition

The two regions are joined at the capping inversion through kinematic and dynamic conditions. The former requires continuity of w across the inversion, while the latter, obtained from the pressure jump associated with the displaced interface, yields a Robin boundary condition (30) for the vertical velocity at the top of the ABL. p. 8

Where the paper says it

The ABL requires an upper boundary condition at 𝑧 = 𝐻, while the free-atmospheric solution depends on the unknown interfacial amplitude 𝑤ˆ + (𝑘, 𝐻). These two missing ingredients are supplied by coupling the two regions across the capping inversion through kinematic and dynamic conditions.

The gravity-wave feedback on the boundary layer is therefore expressed through a single upper boundary condition. p. 0
Where the paper says it

Wake evolution and gravity-wave dynamics therefore constitute a single coupled problem. The present work addresses this coupling by retaining the vertical structure of the boundary layer together with the non-hydrostatic dynamics of the free atmosphere, joined through a dynamic condition at the inversion.

--- Page 10 ---

E2-2 Governing equations

We consider a background flow with horizontal velocity that only varies with height U(z) and no vertical velocity. The steady two-dimensional non-hydrostatic Boussinesq equations yields U ∂u/∂x + u ∂u/∂x + wU' + w∂u/∂z = −∂p/∂x + ∂τ/∂z + f, (1a).

We retain the full nonhydrostatic momentum equations in both regions, meaning the horizontal velocity, vertical velocity and pressure remain dynamically coupled throughout both regions. It is also worth noting that the non-linear advection terms are retained in (1). p. 3

Where the paper says it

Consequently, in the present formulation the horizontal velocity, vertical velocity and pressure remain dynamically coupled throughout both regions, making the vertical acceleration an integral part of the model. It is also worth noting that the non-linear advection terms are retained in (1).

Previous reduced-order numerical works [e. p. 3
Where the paper says it

It is also worth noting that the non-linear advection terms are retained in (1). Previous reduced-order numerical works [e.

g. Ott, 2011, Ebenhoch et al. p. 2
Where the paper says it

A separate line of work has instead used reduced-order numerical models, and in particular linearised CFD approaches such as FUGA [Ott, 2011] and other works [e. , Ebenhoch et al.

, 2017] usually linearise the governing equations by dropping all products of perturbation quantities. p. 3
Where the paper says it

Ott, 2011, Ebenhoch et al. , 2017] usually linearise the governing equations by dropping all products of perturbation quantities.

We therefore retain the non-linear advection and forcing terms when solving for the flow within the boundary layer (§2.2). The turbine forcing is also non-linear for the same reason, since the thrust force scales with (U + u) squared. p. 3

Where the paper says it

Such a simplification is formally valid only where the perturbations remain small compared with the background flow, which may not be necessarily the case near the turbines. The turbine forcing is also non-linear for the same reason, since the thrust force scales with (𝑈 + 𝑢) 2 .

--- Page 10 ---

E5-6 Numerical implementation

Numerical Implementation

We employ a spectral discretisation in the horizontal direction, which provides a computationally efficient treatment of the governing equation. It also facilitates the implementation of our key vertical boundary condition (30), which is non-local and naturally expressed in Fourier space. p. 5

Where the paper says it

the local solidity of the turbine row: the force is largest at hub height and tapers to zero at the rotor tips. For numerical implementation, the forcing term is smoothed in both the streamwise and vertical directions using a Gaussian kerne Since the governing equations are solved for the laterally averaged perturbation field 𝑢(𝑥, 𝑧), they carry by construction no information about the lateral variation of the flow within the farm.

The solver was run for six iterations using an under-relaxation factor of 0. p. 4
Where the paper says it

Flow E1-5 the background shear, leaves the perturbation stress 𝜏 = 2𝜈𝑡(𝑧) 𝜕𝑢 𝜕𝑧 , (6) where 𝜈𝑡 = 𝑙 2 𝑚 d𝑈/d𝑧 is the turbulent viscosity of the background profile. The factor of two therefore arises from linearising the quadratic closure, and 𝜈𝑡 depends on the background shear alone, so the farm’s contribution to the mixing is not represented.

5.

At each iteration, the right-hand side is known and (31) can be solved independently for each horizontal wavenumber k. The nonlinear terms are evaluated pseudo-spectrally, with products formed in physical space and dealiased using the two-thirds rule. p. 3

Where the paper says it

(2) where ∇ 2 = 𝜕 2 𝜕𝑧2 + 𝜕 2 𝜕𝑥2 and N (𝑤) represents the nonlinear terms: N (𝑤) = −∇2𝑤 ∫ 𝑥 0 𝜕𝑤 𝜕𝑧 (𝑠, 𝑧) d𝑠 + 𝑤 𝜕 2𝑤 𝜕𝑥𝜕𝑧 + ∫ 𝑥 0 𝜕 3𝑤 𝜕𝑧3 (𝑠, 𝑧) d𝑠 , (3) with 𝑠 being a dummy variable for integration. The shear stress is modelled using the Boussinesq turbulent-viscosity hypothesis with a mixing-length closure [Belcher et al.

At each iteration, the inverse Fourier transform of wˆ is already computed in order to evaluate the nonlinear terms in physical space. p. 7
Where the paper says it

These simplifications reduces (1) to 𝑈H 𝜕𝑢 𝜕𝑥 = − 𝜕 𝑝 𝜕𝑥 , (16a) 𝑈H 𝜕𝑤 𝜕𝑥 = − 𝜕 𝑝 𝜕𝑧 + 𝑏, (16b) 𝜕𝑢 𝜕𝑥 + 𝜕𝑤 𝜕𝑧 = 0, (16c) 𝑈H 𝜕𝑏 𝜕𝑥 + 𝑁 2𝑤 = 0, (16d) where 𝑈H = 𝑈(𝐻). After eliminating 𝑢, 𝑝 and 𝑏 in favour of 𝑤 in (16), we obtain 𝜕 2𝑤 𝜕𝑥2 + 𝜕 2𝑤 𝜕𝑧2 = −(𝑁 2 /𝑈 2 H ) 𝑤, (17) Taking the Fourier transform of (17) with respect to 𝑥 (denoted by hat), the well-known Helmholtz equation used in mountain-wave theory [Gill, 2016] can be obtained: 𝜕 2𝑤ˆ (𝑘, 𝑧) 𝜕𝑧2 + 𝑚 2𝑤ˆ (𝑘, 𝑧) = 0, with 𝑚 2 = 𝑁 2 /𝑈 2 H − 𝑘 2 , (18) where 𝑘 is the horizontal wavenumber.

The flow field typically converges within 4–7 iterations. p. 2
Where the paper says it

To address these limitations, this work develops a new reduced-order numerical framework that resolves the continuous vertical structure of the ABL while capturing the two-way coupling between wake flows and gravity waves within a single set of governing equations. The flow inside the ABL is obtained from a simplified numerical model in which the turbine forcing acts directly on the resolved flow field, so that no engineering wake model and iterative coupling between micro- and meso-scale components are needed.

--- Page 11 ---

E3-2 Boundary layer

Boundary Layer

Within the boundary layer, we assume constant density, so N = b = 0. Rearranging the remaining terms in (1) and eliminating u and p gives the following equation for w: U∇ 2w − U''w + N(w) = −∂ 2τ/∂z squared + ∂f/∂z, (2). The shear stress is modelled using the Boussinesq turbulent-viscosity hypothesis with a mixing-length closure. p. 3

Where the paper says it

(2) where ∇ 2 = 𝜕 2 𝜕𝑧2 + 𝜕 2 𝜕𝑥2 and N (𝑤) represents the nonlinear terms: N (𝑤) = −∇2𝑤 ∫ 𝑥 0 𝜕𝑤 𝜕𝑧 (𝑠, 𝑧) d𝑠 + 𝑤 𝜕 2𝑤 𝜕𝑥𝜕𝑧 + ∫ 𝑥 0 𝜕 3𝑤 𝜕𝑧3 (𝑠, 𝑧) d𝑠 , (3) with 𝑠 being a dummy variable for integration. The shear stress is modelled using the Boussinesq turbulent-viscosity hypothesis with a mixing-length closure [Belcher et al.

Substituting (6) into (2) and using the continuity equation (1c) yields U∇ 2w − U''w − 2∂ 2τ/∂z squared / ∂z squared ∫ x 0 w(s, z) ds + 2τ(z)/∂x ∂2w/∂z cubed + N(w) = −∂f/∂z, (7).

The four boundary conditions wˆ = ∂z wˆ = 0 at z = z0 and ∂zzwˆ - k 2wˆ = 0 and (30) at z = H are incorporated into the finite-difference discretisation using one-sided differences or ghost points. Once w is obtained, the remaining fields u, p, and η follow from (1c), (1a), and (20), respectively. p. 8

Where the paper says it

As mentioned earlier, in these models, the vertical velocity is not dynamically coupled to the pressure and horizontal velocity fields within the boundary layer and therefore does not enter the interface coupling in the same manner as in the present formulation. We next derive the dynamic condition by comparing the pressure distribution before and after displacing the capping inversion by 𝜂, which are denoted by 𝑃 (background pressure) and 𝑃 + 𝑝 (background plus perturbation), respectively (figure 3).

--- Page 16 ---

E8-8 Figure 2 shows the variation of βn for farms in which each row is displaced laterally by a constant offset Δy relative to the row immediately upstream, so that y n - y j = (n - j)Δy. Aligned and staggered farms correspond to the limiting cases Δy/Sy = 0 and Δy/Sy = 0.5, respectively, and by the periodicity of θ3 these two limits bracket all possible layouts of this type. The correction is largest immediately downstream of the first row, where the wakes are narrowest relative to the lateral spacing, reaching β2 ≈ 4 for an aligned farm with Sx = 5D.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI can perform parameter sweeps over atmospheric conditions because of its computational efficiency and physics-based representation of wake and gravitywave coupling, making it suitable for such tasks while remaining orders of magnitude below that of LES.

  2. The improved system can be used for assessing wind farm performance and farm to farm interactions by providing a mid-fidelity tool for bridging the gap between computationally expensive high-fidelity simulations and highly parametrised engineering models.

  3. The AI system can serve as a candidate for parameterisations in weather and climate models because of its computational efficiency and physics-based representation of wake and gravitywave coupling, enabling efficient assessment of these effects.

  4. The system can capture the shift in the location of maximum inversion displacement between supercritical and subcritical conditions, which is important for understanding how different atmospheric regimes affect wind farm behavior.

  5. The AI can reproduce the row by row structure of the velocity deficit within a farm and its cumulative growth along the farm, matching LES results for cases like JH case.

Abstract

This paper develops a reduced-order numerical framework for modelling the two-way coupling between gravity waves and turbulent wakes in large-scale wind farms. Starting with the non-hydrostatic Boussinesq equations and introducing simplifications appropriate to the boundary layer and the overlying stratified free atmosphere yield separate governing equations for the two regions. These are coupled through a dynamic boundary condition at the capping inversion, which directly captures the feedback of gravity waves on the boundary-layer flow. A mixed pseudo-spectral finite-difference discretisation yields a computationally efficient model while retaining vertical boundary-layer structure. For the semi-infinite wind farm considered here, a complete solution is obtained in a matter of seconds on a standard laptop. Comparisons with large-eddy simulations (LES) confirm the model successfully reproduces both internal wind-farm flow and large-scale gravity-wave effects. It captures the upstream blockage induced by adverse pressure gradients, as well as the accelerated wake recovery within and downwind of the farm, driven by favourable pressure gradients.

Related papers