Renewable Energy 180 (2021) 806e828

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Numerical modelling of neutral atmospheric boundary layer flow

through heterogeneous forest canopies in complex terrain (a case
study of a Swedish wind farm)
Hamidreza Abedi a, c, *, Saptarshi Sarkar b, c, Håkan Johansson b, c
a
Division of Fluid Dynamics, Department of Mechanics and Maritime Sciences, Chalmers University of Technology, SE-412 96, Gothenburg, Sweden
b
Division of Dynamics, Department of Mechanics and Maritime Sciences, Chalmers University of Technology, SE-412 96, Gothenburg, Sweden
c
Swedish Wind Power Technology Centre (SWPTC), SE-412 96, Gothenburg, Sweden

a r t i c l e i n f o a b s t r a c t

Article history: This paper exposes the risk of generalization of wind conditions from a single met-mast measurement to
Received 9 June 2021 be representative of the actual flow field in a wind farm situated in complex terrain. As a case study,
Received in revised form Large-Eddy Simulation (LES) of the neutral Atmospheric Boundary Layer (ABL) flow for a mid-western
9 August 2021
Sweden wind farm is performed. The site-specific complex topography and the forest properties like
Accepted 11 August 2021
Available online 18 August 2021
the Plant Area Density and the tree heights are extracted from the Airborne Laser Scanning (ALS) 3D data,
thus the forest is heterogeneous. To emphasize the impact of the local topography and surface roughness
on the wind field, the wind turbines are not included in the numerical simulations. The predicted wind
Keywords:
Complex terrain
speeds using LES are compared to wind speed from the nacelle-mounted anemometers taken from the
Large-eddy simulation wind farm's turbine SCADA data, focusing on the wake-free turbines. A sufficient degree of match is
Atmospheric boundary layer flow observed, supporting the accuracy of the numerical simulations. The results show that inflow variables
Heterogeneous forest i.e., mean wind speed, shear exponent and turbulence intensity vary at each wind turbine location
Short-term damage equivalent load justifying the need for turbine-specific assessment of the wind resource in a wind farm located in
Bearing life forested complex terrain. The inter-turbine (between turbines in the wind farm) differences in wind
resource is quantified in terms of the difference in turbine-specific structural and mechanical loads by
running wind turbine mechanical simulations using the extracting the wind fields predicted by the LES.
The results show that not only inter-turbine loads varying significantly in the wind farm, but the turbine
loads also differ significantly if a homogeneous assumption is made for the forest. Most importantly, it
was found that the homogeneous forest assumption predicted a higher turbulence intensity compared to
a heterogeneous forest resulting.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction turbulence [1,2]. Therefore, the prediction of the flow field is

extremely important to ensure proper choice of turbine and oper-
Wind power is known as one of the most environment-friendly ation, to reduce maintenance costs and increase turbine's life.
sources of renewable and clean energy which has been a pioneer- The Atmospheric Boundary Layer (ABL) flow consists of various
ing renewable technology in recent decades. The development and spatial scales ranging from millimeters to several hundred meters
operation of wind farms requires wind resource assessment which determining the flow pattern over the entire wind farm. In ABL
is the core concept of the economic feasibility of a wind turbine modeling, among various number of meteorological phenomena
farm project. Wind turbines always operate within the Atmo- such as Coriolis force, buoyancy forces and heat transport, the
spheric Boundary Layer (ABL) and are subjected to atmospheric impact of ground topography is taken into account as surface
roughness, dominating the near-surface wind flow. The ground
topography is mainly characterised by the complex terrain [3] and
* Corresponding author. Division of Fluid Dynamics, Department of Mechanics
the vegetation [4]. The complex terrain affects the motion of large-
and Maritime Sciences, Chalmers University of Technology, SE-412 96, Gothenburg, scale turbulent structures [5] whereas the motion of small-scale
Sweden. turbulent structures is governed by the vegetation. Therefore, the
E-mail address: [email protected] (H. Abedi).

https://doi.org/10.1016/j.renene.2021.08.036
0960-1481/© 2021 Elsevier Ltd. All rights reserved.
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

airflow in a wind farm strongly depends on local topography and than RANS. However, LES is a computationally expensive and re-
surface roughness [6]. The local topography and surface roughness quires more attention in the near wall region for flows with very
significantly change the flow patterns in and around wind farms high-Reynolds number such as ABL flow [18] to avoid mismatching
which in turn increase the uncertainty of on-site wind resource between the modeled near-wall region and the resolved outer
assessment. Moreover, the flow features significantly affect the region.
aerodynamic loads on rotor blades and annual power production Initial attempts for CFD modelling of the air flow over complex
for a wind farm located in a complex terrain. topographies were limited to the flat terrain and 2D/3D hills
Because of the great inhomogeneity of the flow features in a [19e23] and have been continuously developed until now [24e27].
complex terrain, the on-site measurement [7] cannot afford a full A review of wind flow over hills, escarpments and valleys using
map of the wind field over the entire wind farm. In addition, the RANS has been summarized by Bitsuamlak [28].
uncertainty of local field measurement to provide a detailed In 2000s, several studies have been carried out to simulate
description of a wind resource due to the topographic complexity airflow over complex terrains [29e31] using RANS turbulence
[8] along with the time-consuming and expensive process of col- models [32e39]. In addition, the field measurement campaigns of
lecting measured data by means of on-site meteorological masts two isolated hills, Askervein [40] and Bolund [41,42] have been
motivates performing numerical modelling to predict the wind widely used for validation of complex terrain flow models using
field over any complex terrain wind farm [9]. Therefore, developing steady-state RANS equations with k-ε and k-u turbulence closures
accurate computational models [10] for the flow field for wind [43e48].
farms located in hilly regions is quite important for design In complex terrain, the highly variable ground elevations in-
purposes. crease the complexity of the air flow. Furthermore, the wake in-
Historically, the ABL modelling over complex terrain (mainly teractions between the turbines and the local ground topography
2D/3D low hill) was performed using analytical theory [11e13] make the flow complexity greater so that the attached boundary
showing a reliable flow prediction approach in the region without layer flow assumption is no longer valid. The separated flow and re-
separation. Advances in computational power and development of circulation regions increase the unsteadiness of the airflow and the
numerical algorithms for non-linear equations enabled to simulate RANS-based turbulence models are no longer capable of predicting
ABL using Computational Fluid Dynamics (CFD). Nowadays, CFD the flow accurately. Hence, for a very high-Reynolds-number tur-
has become a common numerical simulation tool to predict the bulent boundary layer such as ABL flow, the LES based turbulence
atmospheric boundary layer turbulence structure ranging from model is a promising approach to simulate the air flow over com-
mesoscale to microscale modeling. Apart from successfully usage of plex terrains.
mesoscale models [14] for wind resource mapping of countries and Successful applications of LES in ABL flow have been reported
regions with a typical model grid resolution of 1e5 km, the limi- for simulating the flow over flat terrain [49e59]. Breton et al. [60]
tations of mesoscale models (due to a very coarse horizontal res- reviewed the current state-of-the-art Large-Eddy Simulation (LES)
olution and resolved time scales) make them improper tools for approach for wind farm aerodynamics under various atmospheric
wind farm and wind turbine design. This propels the use microscale and terrain conditions. Similar to the RANS turbulence models,
models for micro-siting even though they most often rely on local various studies have been performed to validate LES turbulence
measurements. In microscale models, CFD can cover 10e20 km model against the field measurement over the Askervein hill
area and is run with 2e20 m model grid resolution which makes it [61e64] and Bolund hill [65e71]. In addition to the Askervein and
possible to capture topographic effects of complex terrains and Bolund as isolated hills, there are few studies on ABL flow over
vegetation. Hence, the capability of advanced CFD of including naturally complex terrain [72e80].
topographical phenomena in ABL simulation flow provides a more Like terrain complexity, the vegetation as a part of ground
realistic wind farm flow prediction. However, it requires large topography also has a significant impact on ABL flow. Apart from
computational resources for wind resource assessment over a the practical risks associated with wind turbines in forest regions
complex terrain. Moreover, the predicted wind farm flow by CFD is [81], a higher turbulence level and wind shear due to the forest
highly affected by the computational grid resolution, the specified canopies have been reported in various numerical and experi-
boundary conditions, the heterogeneous surface roughness and the mental studies [36,76,82e84]. In the past twenty years, successful
choice of turbulence models. numerical modellings of flow over and inside horizontally homo-
Two common CFD methodologies which have been extensively geneous forest canopies using LES have been performed
used to simulate the atmospheric boundary layer flow are the [58,76e78,80,85e90]. Horizontally homogeneous forest canopies
Reynolds-Averaged Navier-Stokes (RANS) and Large-Eddy Simula- are presented by either Leaf Area Density (LAD) or Plant Area
tion (LES). The RANS models only predict the mean flow where Density (PAD) varying in the vertical direction and considered as
Reynolds stresses are entirely modeled using different turbulence drag in the ABL flow. Leaf Area Density (LAD) or Plant Area Density
closures with reasonable computational costs. They include con- (PAD) have been previously been used to quantify air-vegetation
stant numbers which are calibrated by either fundamental flows exchange of momentum, latent/sensible heat and carbon dioxide
(e.g., channel flow) or measurement data which in turn increase [91]. They are defined as one-sided plant/leaf area per unit of
uncertainties of RANS simulations. Moreover, at very high- horizontal layer volume [92]. The difference between the PAD and
Reynolds number turbulent flow such as ABL flow over the com- LAD is that the PAD includes both leaf and wood surface areas
plex terrain including non-equilibrium region such as adverse whereas the LAD only includes leaf surface area. The distribution of
pressure gradients [15] and three dimensional separation [16], the plant/leaf elements from ground to top of canopy is defined by a
reliability and accuracy of RANS models to predict turbulence vertical distribution profile of PAD/LAD. In addition, the Plant Area
characteristics is low. On the other hand, LES is able to predict Index (PAI)/Leaf Area Index (LAI), obtained by the vertical integra-
unsteady flow features and chaotic motion of turbulent structures tion of the PAD/LAD profiles, are dimensionless parameters that are
at very high-Reynolds number flows where the large-scale flow usually used to characterize the density of canopies. The assump-
structures are dominant. Unlike RANS, in LES the large flow struc- tion of forest homogeneity increases the uncertainties of flow
tures, the so-called grid-scales, are resolved whereas the small flow simulation where some of the extreme events such as wind gusts
structures, the so-called subgrid-scales, are modeled using a sub- and sudden change in wind speed cannot be well-predicted.
grid model [17]. Therefore, it can provide more accurate flow field Therefore, inclusion of forest heterogeneity in numerical flow
807
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

modelling provides more realistic simulation framework. However, density 0.5e1 points per square metre for the terrain and 1e2
obtaining accurate canopy properties is a difficult task [93]. Lately, points per square metre for the vegetation (forest). A commercial
Boudreault [94] and Arnqvist [91] presented numerical algorithms software called Global Mapper was used to derive the complex
to extract the detailed distribution of Plant Area Density (PAD) terrain coordinates with the specified horizontal resolution of 7  7
profiles and Plant Area Index (PAI) from airborne laser (lidar) scan m2. The coordinates were imported into STAR-CCMþ in the STL
data for large areas with higher accuracy than the previous format to generate the computational grid for the numerical sim-
methods [95]. ulations.The forest properties, Plant Area Density (PAD) and trees
Contrary to the horizontally homogeneous forest, in heteroge- height, have been extracted from the ALS data with the certain
neous forest, LAD/PAD profile varies in both horizontal and vertical resolution of 20  20 m2 and 2 m in the horizontal and vertical
directions. This inhomogeneity has a significant impact on the directions, respectively. The software to extract the forest proper-
variability of wind field [93,94,96] and it is more pronounced for ties from the ALS data is available at software repository GitHub
wind farms located in a complex terrain. Recent studies by Bou- [91,99].
dreault et al. [96] and Ivanellet al. [39] have shown the importance Fig. 2 shows the terrain elevation with respect to sea level (left)
of local heterogeneity of forest canopies in wind resource assess- and forest properties (right) in a local domain of 14  16 km2
ment where the forest edges and local variation of PAD distribution surrounding the wind farm. The red dots indicate the location of
have large impacts on the airflow inside and above the canopies. eight wind turbines of the wind farm where all are placed in the
Additionally, Ross [97] demonstrated that highly varying forest highest elevation of the neighbourhood region. Moreover, the black
density results in acceleration and deceleration of the flow into and dot displays the location of the met mast.
out of the canopy. However, to the best of authors’ knowledge,
there are very few studies comparing the flow field characteristics 2.1. Measurement data
extracted from LES using heterogeneous forest assumption to
operational or met mast data at several locations. The on-site meteorology mast data at Ro €bergsfja
€llet site has
In this study, numerical simulation of the neutral ABL flow been equipped with two heated cup and vane wind measurement
through heterogeneous and homogeneous forest canopies in stations designed for arctic conditions (Vaisala Wind Set WA25) at
complex terrain are performed. For homogeneous forest modeling, 40 and 60 m above ground. It determines the wind speed and di-
the average PAD profile over computational domain is used. The rection as well as the shear exponent. It is located at the clearing
results for the neutral ABL flow through homogeneous forest region of the North of the wind farm, close to the wind turbine 1
modeling are not presented and are used for comparison purposes (see Fig. 1 c). These data are used for validation of the numerical
only. The neutral ABL flow assumption helps to study the turbu- modelling. The measurement data include a full year mean and
lence generation mechanism only by means of local topography standard deviation of the wind speed measured in 2006 with the
and surface roughness [57] where the impact of the buoyancy frequency of an hour.
forces are neglected. The focus of the study is to employ a high- According to Fig. 3, at 60 m above the ground the dominant
fidelity CFD method - the so-called Large-Eddy Simulation (LES) - wind direction is 216 with respect to the North. The 10 m/s mean
to model the airflow inside and over complex terrains and around wind speed, corresponding to the dominant wind direction, is
each wind turbine in a mid-western Swedish wind farm while the chosen for the numerical simulation.Comparison between the full-
difference between the heterogeneous and homogeneous forest day mean wind speed measurement with those measured only in
assumptions on the dynamic response of the wind turbines in a the day-time and night-time can be observed in Fig. 4. The mean
wind farm is also investigated. A commercial software (STAR- values for the full-day, day-time and night-time time intervals are
CCMþ) is used to predict the atmospheric turbulence and wind approximately the same and they are equal to 6.8, 6.6 and 7.1 m/s,
profile over the wind farm. The results are compared against the respectively. The small differences between these values are taken
SCADA (Supervisory Control And Data Acquisition) data of the wind as support for adopting neutral atmospheric boundary layer flow
farm. To study and compare the impact of homogeneous and het- assumption in this study.In addition, the turbulence intensity can
erogeneous forest properties on the dynamic response of the wind be estimated from the measurement data. Fig. 5 shows the varia-
turbines installed in the wind farm, the computed flow field by tion of the turbulence intensity with respect to the mean wind
STAR-CCMþ is supplied to the aeroelastic wind turbine simulator speed measured at the met mast point (60 m above ground) for 1-h
FAST [98]. FAST is an open-source CAE tool for predicting the power averages over the year. As illustrated, the 0.9 quantile values of
production and simulating the structural and system response of turbulence intensity for the measured data is below the class C of
wind turbines developed by NREL. international standard IEC 61400e1 [100] for Normal Turbulence
Model (NTM) wind condition. In other words, the turbulence in-
2. Complex topography tensity for turbine classification at the Ro€bergsfj€
allet site is asso-
ciated with the class C of the IEC standard [100].
The Ro € bergsfja
€llet wind farm, located at Ro € bergskullen in
southern part of Vansbro municipality in Sweden (60160 49.8”N, 3. Simulation set-up
14120 59.6”E), is used as the complex terrain where it is partly
covered by heterogeneous forest with some clearings (see Fig. 1). The numerical simulations are done over the complex terrains
The farm was built in 2007 with the highest point at 543 m above covered by the heterogeneous forest. For this purpose, a high-
sea level (a.s.l) where the difference between the highest and fidelity CFD approach, Large-Eddy Simulation (LES) is employed
lowest points of the wind farm is dyg x 284 m (see Fig. 2). The wind over a rectangular computational domain to predict the neutral
farm consists of eight Vestas V90 horizontal axis wind turbines atmospheric turbulence and time-varying wind profile for a period
(referred to as WT1-WT8) with a hub height of 90 m and a rotor of 90 min with output sampling of 10 Hz. The computational
diameter of 90 m, each with a capacity of 2 MW. domain has dimensions 9Hx6HxH where H denotes the average
The complex topography of Ro €bergsfj€
allet wind farm was height of H x 1.0 km (app. four times higher than the difference
extracted from Airborne Laser Scanning (ALS) 3D-data delivered on between the highest and lowest points of the wind farm, i.e., dyg x
May 2017 by Swedish University of Agricultural Sciences, SLU 284 m) in the vertical direction (y). In other words, LES simulations
(www.slu.se). The ALS data contains a point cloud with a point are performed over a rectangular box of size L ¼ 9.0
808
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

€bergsfja
Fig. 1. The Ro €llet wind farm, located at Ro
€ bergskullen in southern part of Vansbro municipality, Sweden. The numbers indicate the location of eight wind turbines.

km  W ¼ 6.0 km in horizontal plane with average height of i.e., 300 m. All sides of the computational domain are parallel to
H x 1.0 km. each other; and they are flat except the floor which is made up of
The size of the computational domain and simulation run-time the topography of the area extracted from the ALS data.
are limited based on the available computational resources. Fig. 6 The computational grid spacing is constant in the horizontal
shows the layout of the computational grid and the location of directions with Dx ¼ Dz ¼ 17 m where x and z denote the local
the wind turbines inside the farm.As seen, the computational streamwise and spanwise directions, respectively. The extracted
domain (enclosed by the black solid lines) has been chosen to be inhomogeneous PAD profiles over the wind farm, varying in both
aligned with the dominant wind direction, 216 w.r.t. to the North, horizontal and vertical directions with a maximum trees height of
at the site. The rotated domain has the advantage of perpendicular 40 m, is used to take the impact of vegetation on the airflow into
inflow at inlet boundary condition. Therefore, the length and width account. In case of using the Monin-Obukhov similarity relation-
of the domain are associated with the local streamwise (x) and ships with a prescribed physical roughness length [83], as recom-
spanwise (z) directions, respectively. In addition, the wind farm is mended by Richards and Hoxey [101], the first cell height in the
situated in the middle of the computational domain allowing tur- grid mesh must be at least ten times larger than the physical
bulent structures to be developed within the distance between the roughness length at the ground prescribed by z0 ¼ 0.001h where h
inlet boundary and the wind farm. Because of the varying ground is the canopy height [83]. In the heterogeneous forest, the physical
elevation in the computational domain, the maximum and mini- roughness length varies spatially, so the first cell height can be
mum distance from the ground level are 1600 m and 900 m, specified based on the maximum trees height in the computational
respectively. The minimum distance is equal to three times the domain.
difference between the highest and lowest points of the wind farm In STAR-CCMþ, the LES solver can only handle the ground as the

809
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

€bergsfj€allet region (red dots indicate the location of eight wind turbines of the wind farm), (b) Trees height of the Ro
Fig. 2. (a) Topographic map of the Ro € bergsfja
€llet region. (The
red dots and the black dot display the location of eight wind turbines and the met mast, respectively.).

€ bergsfja
Fig. 3. Wind rose of the mean wind speed measurement in 2006 at 60 m above the ground for Ro €llet site.

smooth wall. To avoid increasing the number of grid cells and to 3.1. Forest properties
maintain enough resolution near the wall for the ABL flow simu-
lation, the first cell height is considered equal to 2 m while thirteen Fig. 7 presents the distribution of trees height and Plant Area
grid cells are used to discretize the highest canopy height in the Index (PAI) within the local region and the computational domain.
vertical direction. In addition, thirty grid cells are vertically As stated in section 2, the forest properties were extracted from ALS
stretched until 210 m to refine the wall region. Above the height of data delivered on May 2017. They are almost identical where the
210 m, a constant grid spacing of Dy ¼ 17 m is used. Because of the average of the trees height are in the range of 10e20 m and the
complex topography over the entire domain, the number of grid maximum of trees height is restricted to 40 m. However, there are
cells in the vertical direction vary between 78 and 91 resulting in a many scattered clearing zones in the local region violating the
total number of approximately 16 million grid cells for the entire homogeneity assumption for the Ro € bergsfja
€llet region.
computational domain. In addition, the average value of the PAI distribution is about one
(below the averaged PAI values for the coniferous and deciduous

810
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

€ bergsfja
Fig. 4. Mean wind speed time series with the frequency of 1 h at Ro €llet wind farm, (a) full-day, (b) day-time, (c) night-time. The red line indicates the annual averaged value.

forests in global temperate ecosystems [102]) which reveals that

the wind farm is not been covered by a dense forest [86]. The
average PAD profile can be also computed from the vertical distri-
bution of PAD over the entire local and computational domains as
presented in Fig. 8.
Both profiles are rather similar except for the first 2 m near the
ground. The mean PAD profile of the computational domain (blue
line in Fig. 8) is used for the precursor and homogeneous forest
simulations.

3.2. Governing equations

In LES, the turbulent eddies are divided into large, resolvable

scales and small, subgrid scales (SGS) where the small scales are
modeled. The distinction between large and SGS scales is done
implicitly by the grid, referred to the grid filtering. The incom-
Fig. 5. Variation of turbulence intensity with the mean wind speed, Blue dots: the pressible, grid-filtered, Navier-Stokes equations are expressed as
measurement data, Black dashed-line: IEC 61400e1, NTM- Class C, Red circles: 90%
quantile of turbulence intensity for the given mean wind speed. e
vv i
¼0 (1a)
vxi

811
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

€bergsfj€allet wind farm, (b) Canopies height covered the Ro

Fig. 6. Layout of the computational domain, (a) Topographic elevation (above sea level) around the Ro € bergsfja
€llet wind
farm. (The red dots and the black dot indicate the location of eight wind turbines and the met mast, respectively in the wind farm.).

Fig. 7. (a) Distribution of canopies height within the computational domain (blue) and the local region (red), (b) Distribution of PAI within the computational domain (blue) and the
local region (red).

1 e
tij  tkk dij ¼ nsgs s ij (3a)
e
e e  3
vv i
þ v vi vj
vt ! (1b) e e
e e e vv i vv j
vxj ¼ vp
1r vx þ v
vxj n vv i
vxj  tij þ Ff ;i s ij ¼ þ (3b)
i vxj vxi

e e
e e 1=2
where v i , r, p and n denote the velocity component in the xi-di- nsgs ¼ ðLÞ2 2s ij s ij (3c)
rection (xi ¼ {x, y, z}), the air density, the pressure and the kinematic
viscosity, respectively and the overbars implies grid-filtered
L ¼ fv minfkd; Cs Dg (3d)
quantities. Inclusion of forest canopies in Eq. (1b) is done through
the source terms Ff,i added to the momentum equations as
yþ 
fv ¼ 1  exp (3e)
A
ee
Ff ;i ¼ CD af  v  v i (2) D ¼ ðDxDyDzÞ1=3 (3f)

where CD ¼ 0.15 denotes the forest drag coefficient [103], af is the where dij, nsgs, sij, fv, k, d, D and yþ denote Kronecker delta, turbulent
e
vertical Plant Area Density (PAD) of the forest and  v  is the velocity viscosity, strain rate tensor, van Driest damping function [105,106],
magnitude. In Eq. (1b), the Smagorinsky subgrid model [104] is von Karman constant, wall distance and filter-width computed by
used to model the small scales given by the local grid size, respectively. Cs and A are the constant model
812
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

simulation as described above. The turbulent fluctuations at

inlet are generated by an in-built function in STAR-CCMþ the so-
called Synthetic Eddy Method (SEM) [107]. The derived Rey-
nolds stresses from the precursor simulation are specified as
input into the SEM to provide the correlation function required
by it. Since the flow is incompressible, the SEM scales the inflow
fluctuations to maintain constant mass flow rate across the
domain. The prescribed inlet boundary condition profiles are
displayed in Fig. 9.
 Outlet: The Pressure Outlet boundary condition with zero gauge
pressure is chosen at the outlet boundary condition.
 Top and Sides: Symmetry boundary condition is specified for the
top and the sides boundaries. It must be noted that the sym-
metry boundary condition at the top boundary may lead to an
artificial flow acceleration [108] because of the short distance
between the highest point of the complex terrain and the top
boundary. Therefore the minimum height of the domain has
been chosen long enough to reduce the impact of the acceler-
ated flow on the wind farm flow. Moreover, because of the
Fig. 8. Comparison between the mean PAD profile covered the computational domain imposed symmetry boundary conditions, the Coriolis force is
€bergsfja
(blue) and the local region (red) at Ro €llet site. not taken into account.

A constant time step of t ¼ 0.1 s is used in the simulation to

ensure that the Courant number is below one over the entire
coefficients assumed to be 0.10 and 25.0, respectively. domain. However, for a few highly skewed grid meshes (due to the
For the spatial and temporal discretization of the governing ground complex topography), there are negligible instances that
equations, a second-order bounded-central differencing scheme the Courant number becomes greater than one. The simulation is
and a second-order time integration scheme are used, respectively. carried out for 120 min. The atmospheric turbulence and time-
varying wind profile for the last 90 min of the simulation over
the rotor swept area of all eight turbines are then extracted and
3.3. Boundary conditions exported to FAST for aerodynamic and aeroelastic analysis.

Generally, the turbulence quantities inside the computational

domain are dominated by the inlet conditions. Thus, the prescribed
inlet profile must be cohered with upstream flow characteristics.
To supply the inlet boundary condition profiles for the complex 3.4. Aeroelastic simulation of wind turbine
terrain simulation, a precursor simulation is performed. The pre-
cursor simulation is carried out over a flat rectangular domain with In this paper, a comparison is made between homogeneous and
periodic boundary conditions in the streamwise and spanwise di- heterogeneous terrain by evaluating the aero-servo dynamic loads
rections. A symmetry boundary condition is employed at the top of experienced by wind turbine under the two different assumptions.
the computational domain. Moreover, the ground is covered by The aim is to understand the effect of the terrain composition on
homogeneous forest where the properties of the homogeneous the flow field and subsequently on the turbine loads giving us
forest is taken from the averaged vertical PAD profile computed insight into the effect the terrain has on the turbine loads. The
from the heterogeneous forest distribution (see Fig. 8). Ro€ bergsfja
€llet wind farm is equipped with 8 Vestas V90e2.0 [109]
The computational domain consists of four different types of wind turbines that have a nominal electrical capacity of 2 MW. In
boundary conditions: this paper, a generic 2 MW wind turbine model is developed to
mimic the behavior/power production of a Vestas V90 machine.
 Wall: No-slip wall boundary condition is set for the ground. The The state-of-the-art aeroelastic simulation tool FAST [98] is used for
ground surface is assumed to be a smooth wall without any aeroelastic simulations of the wind turbine in this paper. The
correction usually done by specifying the momentum flux from generic 2 MW wind turbine model is developed by down-scaling
standard similarity theory [86] based on roughness length. The the NREL 5 MW reference wind turbine [110] guided by Vestas
extracted spatial varying PAD profiles with a horizontal resolu- V90e2.0 [109] power curve and the available SCADA data. For
tion of 20 m  20 m is specified at each grid point of the brevity, the step-by-step down-scaling procedure is not provided in
computational domain using the nearest-neighbour interpola- this paper. It must be noted that the generic 2 MW wind turbine
tion scheme. model used here is not aimed to be an exact representation of a
 Inlet: The inlet boundary condition is specified as Velocity Inlet. It Vestas V90 wind turbine, but serves as a reference/baseline 2 MW
is chosen so that the calculated mean wind speed and the tur- wind turbine model that mimics the performance of a 2 MW Vestas
bulence intensity at the met mast point (located at 60 m above V90 [109] turbine as shown in Fig. 10.
the ground, in the vicinity of wind turbine 1 (WT1) with 5.2 km Two important categories of turbine loads are investigated, the
distance from the inlet boundary condition and in the mid-span along-wind loads on the blades and the tower and the high-speed
of the computational domain) are 10 m/s and 0.15, respectively shaft bearing loads. The loads on the blades and the tower are
similar to the on-site measurement. The inlet boundary condi- compared using the equivalent short-term damage equivalent
tion consists of the sheared mean velocity profile superimposed loads and the bearing loads are compared using the equivalent
to turbulent fluctuations that is required in LES approach. The bearing lives. In the following subsections a brief description of
sheared mean velocity profile is taken from the precursor these quantities is presented.
813
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 9. The Prescribed inlet boundary condition profiles obtained from the precursor simulation.

Fig. 10. Comparison of SCADA data against FAST model, (a) Generator power, (b) Rotor speed.

3.5. Estimation of short-term damage equivalent turbine loads critical component of a wind turbine [114]. The schematic diagram
of the high-speed shaft can be seen in Fig. 11. Typically, the high-
The along-wind loads on the blades and the tower is used for speed shaft bearings have the highest failure rates [115]. Hence, it
comparison as their magnitude is higher than the cross-wind loads is important to understand and predict the behaviour of the high-
experienced by the turbine. The loads considered here are: Blade 1 speed shaft bearing loads under the two different types of forest
root flapwise bending moment (named Bl1RootFlpMom in FAST); assumptions. The high-speed shaft cylindrical roller bearing life is
Tower base fore-aft bending moment (TwrBsFAMom); and, Tower used in this paper for comparison. The High-Speed Shaft Cylindrical
top yaw bearing moment (TTYawMom). These measurements are Roller Bearing (HSS-CRB) life is estimated using the high-speed
available directly as outputs from FAST [98]. However, for appro- shaft cylindrical roller bearing radial force (Fr). As this quantity is
priate comparison, the structural loads are compared using the not available directly as an output from FAST [98], a approximate
short-term damage equivalent loads (DELs) estimated from a procedure of estimating the radial bearing force from the high-
90 min time history prediction. The short-term fatigue damage speed shaft torque (available as a direct output from FAST [98]) is
equivalent loads (DELs) [111,112] are calculated on the basis of the presented next.
output times series, which, for a given mean wind speed, is deter- On the high-speed shaft in Fig. 12, from torque balance it can be
mined by observed that
!1=m
1 X T ¼ Ft r (5)
DEL ¼ n ðDFi Þm (4)
Neq i i
where, T is the high-speed shaft torque, Ft is the tangential force, r is
the working pitch radius of the pinion.
where ni is the number of load cycles with range DFi in a time series,
The helical gear's transmission/gear-pair force, Fn, which is
i is the fatigue cycle index, m is the Wholer exponent, and Neq ¼ feqT
normal to the tooth surface, can be resolved into a tangential
where is feq the DEL frequency, and T is the elapsed time of the time-
component, F1, and a radial component, Fr, as
series under consideration. The short-term DELs have been esti-
mated using Mlife [113] distributed by the National Renewable
Energy Laboratory (NREL). F1 ¼ Fn cos an
(6)
Fr ¼ Fn sin an

3.6. Estimation of high-speed shaft bearing life where, an is the normal pressure angle. The tangential component,
F1, can be further resolved into circular sub-components, Ft, and
The drive-train/gearbox of the wind turbines are one of the most axial thrust sub-component, Fx as
814
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 11. Schematic diagram of the high-speed shaft.

Fig. 12. Forces on a helical gear mesh.

good estimation of a radial bearing forces from high-speed shaft

Ft ¼ F1 cos b torque. This estimated radial bearing forces can be used now to
(7)
Fx ¼ F1 sin b estimate the bearing life.
The fatigue life of an individual bearing is the number of revo-
where, b is the helix angle. Then, using simply supported beam lutions (or the number of operating hours at a constant speed) that
assumption the resultant bearing forces on the tapered roller the bearing operates before the first sign of metal fatigue (rolling
bearing (TRB) and cylindrical roller bearing (CRB) as shown in contact fatigue (RCF) or spalling) occurs on one of its rings or rolling
Fig. 12 can be estimated as elements. The rating life L10 is the fatigue life that 90% of a suffi-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  ciently large sample of identical bearings operating under identical
a conditions can be expected to attain or exceed. For a given bearing,
F TRB ¼ F 2r þ F 2t
L its fatigue life can be predicted using simplified equations as shown
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (8) in Ref. [116]. These predictions although not highly accurate,
b
F CRB ¼ F 2r þ F 2t together with engineering experience and judgement provide a
L
good basic for bearing selection.
while it is assumed that the axial forces are taken up by the TRB For applications like wind turbines the operating conditions,
such as the magnitude and direction of loads, speeds, temperatures
alone. These equations, although approximate, provides us with a
815
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

and lubrication conditions, are continually changing. In these cases, 4. Results

the load spectrum can be reduced to a histogram plotting constant-
load blocks. Each block characterizes a given percentage or time- 4.1. Validation against met mast
fraction during operation. Heavy and normal loads consume
e
bearing life at a faster rate than lighter loads. Therefore, it is The time-averaged grid-filtered axial wind velocity Cv 1 D and
important to have peak loads well represented in the load histo- e02
0:5
e
gram, even if the occurrence of these loads is relatively rare and of streamwise turbulence intensity TIu ¼ Cv 1 D =Cv 1 D at the met mast
relatively short duration. Under variable operating conditions, location, obtained from the simulation for a period of 90 min is
bearing life can be rated using compared against the measurement data in Fig. 13. The - and the
error bars display the mean value and the min/max of the measured
quantities at 40 m and 60 m heights above the ground in the
1 dominant wind direction i.e., 216 w.r.t. North, respectively. The
L10m ¼ P Ni
(9)
e
i Li
10m
power spectral density of axial wind velocity (v 1 ) at 60 m height
above the ground at the met mast location is also presented in
where Fig. 13(c). It follows a  5/3 decay and drops after the cut-off fre-

L10m ¼ SKF½116 rating life ðat 90% reliabilityÞ ½million revolutions

Li10m ¼ SKF½116 rating lives ðat 90% reliabilityÞ under constant condition i ½million revolutions
P
Ni ¼ life cycle fraction under condition i; Ni ¼ 1

quency of f ¼ 0.1 Hz limited by the grid resolution in the simulation.

4.2. Validation against SCADA

The LES predicted wind field is compared to wind speed as

Fig. 13. (a) Time-averaged axial wind velocity profile, (b) Streamwise Turbulence intensity at the met mast location compared with the measurement ( ), (c) Power spectral density
of axial wind velocity at 60 m height above the ground at the met mast location, ( )  5/3 law.

816
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

recorded by nacelle anemometer and collected through the wind by LES. As seen from Fig. 16(d), turbines WT1, WT5, WT7 and WT8
turbine's SCADA system on WT1-WT8. For the site under consid- have no immediate other turbine upstream. Since no effect from
eration, 1 Hz SCADA data is available (with some losses) from 20th turbine on the wind field is considered in the LES simulations in
June 2017 to 3rd February 2019. The LES simulations are carried out this study (for instance using actuator disk model) it can be
at 10 m/s inlet velocity, 216 to north, under assumption of neutral assumed the TI level predicted by LES underestimates the actual TI
atmosphere. Also, the LAD/PAD was measured when trees had for WT2, WT3, WT4 and WT6. This assumption is confirmed by the
leaves. Therefore, 10-min samples of SCADA data are extracted comparison in Fig. 15. No attempt of using SCADA to compare to LES
when WT1 (which is adjacent to the met mast) wind speed satisfies predicted wind shear is made.
the following conditions: mean wind is 10.32 ± 1 m/s, turbine yaw
direction is 216 ± 7 to north, time is 11:00e17:00 h and is outside 4.3. Flow characteristics over the entire wind farm
the period Oct 1st to May 1st. For comparison, a less restrictive set
of samples is collected where the daytime and summer time Fig. 16(a) and (b) display the iso-surface of the horizontal mean
criteria are ignored, and only wind speed and direction is consid- qffiffiffiffiffi2 2
e e
ered. A few instances of samples where at least one turbine was out wind speed (s ¼ Cv 1 D þ Cv D) and turbulent kinetic energy,
3
of operation and did not report data were discarded. e0 e0
The anemometer data is likely disturbed by the rotor and it is defined as k ¼ 12 Cv i v i D, at hub height (i.e., 90 m above the ground)
assumed that wind speed data is similarly disturbed at all turbines. for a period of 90 min. All turbines have been located at the highest
Moreover, to avoid bias from the distribution of wind speed sam- elevation compared to the surrounded area. As seen, wind turbine
ples within the chosen wind speeds, the difference in wind speed WT1, WT7 and WT8 have greater hub-height mean wind speed
between turbines are chosen as evaluation criteria over the abso- than the other turbines, mostly related to their elevations level. On
lute values. Level of turbulence is compared using the turbulence the other hand, WT2, WT6 and WT4 experience higher inflow
intensity (TI) values since TI is dimensionless. The results form turbulent kinetic energy at hub height than the others. This moti-
comparison are presented in Fig. 14 and Fig. 15. vates the study of inflow properties at each wind turbine station in
In Fig. 14 the difference in wind speed of turbine WT2-WT8 more detail.
compared to WT1 is presented together with the difference as
predicted by the LES in Table 1. Fig. 15 compares turbulence in- 4.4. Flow characteristics at turbines’ location
tensity from the same SCADA samples with TI as predicted by LES
simulations in Table 1. Fig. 17(a) shows the time-averaged grid-filtered wind velocity
In terms of mean wind speed differences, as displayed in Fig. 14, qffiffiffiffiffi2 2
e e
SCADA data reports a lower wind speed (negative differences) for (s ¼ Cv 1 D þ Cv D) along the vertical axis (y) for all eight turbines
3
all turbines than LES, most prominent for turbines WT3, WT4 and extracted for a period of 90 min. As seen, the mean wind profiles
WT7. It is interesting to note, that a more restrictive set of wind vary w.r.t. the turbines’ location because of the terrain complexity
conditions that are more likely to target neutral atmospheric con- and forest canopies heterogeneity. In addition to the mean wind
ditions (black) is closer to the LES-predicted wind speeds (red). profile, each turbine encounters a different level of turbulence
From the comparison in Fig. 15, a good agreement is obtained for represented by the turbulent kinetic energy (k) in Fig. 17(b). A large
turbines WT5, WT7 and WT8, while poorer agreement is obtained difference in mean wind velocity and turbulent kinetic energy
for the remaining turbines. For all turbines but WT7, the more profiles between the lower section (from 45 m to 90 m) and the
restrictive samples (black) have a mean closer to the TI as predicted upper section (from 90 m to 135 m) of the rotor plane make

Fig. 14. Relative frequency of 119 10-min samples of wind speed collected at WT1 and difference to corresponding samples of WT2-WT8. Black mean value of sample's difference.
Green mean of 1271 samples less likely from neutral conditions. Red is difference predicted by LES according to Table 1. Yellow is difference predicted by LES with homogeneous
forest assumption.

817
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 15. Relative frequency level of turbulence intensity (TI) of the 119 10-min samples collected at WT1-WT8. Black mean value of 119 samples. Green mean of 1271 samples less
likely from neutral conditions. Red TI predicted by LES according to Table 1. Yellow TI predicted by LES with homogeneous forest assumption.

additional undesirable cyclic forces and moments variations. impact of wind shear over the rotor swept area. According to IEC
Fig. 17(c) and (d) display 90 min time-averaged wind veer (4) and 61400-12e1:2017 standard (Wind energy generation systems-Part
yaw (g) angles along the vertical axis (y) for all eight turbines, 12e1: Power performance measurements of electricity producing
respectively. Although the inlet inflow profile is veer and yaw-free wind turbines) [117], the rotor equivalent axial wind speed is
and the Coriolis force is not taken into account, the mean wind determined as
direction varies both vertically (veer) and horizontally (yaw).
Contrary to the mean wind veer angle (4) which is almost constant 3 !1=3
along the (y) axis, the mean wind profile tends to turn in the ver- Xn
e Ai
UEq ¼ v (10)
tical direction. This reveals that the turbines located in a complex i¼1
A
1;i
terrain covered by a heterogeneous forest are exposed to a non-
uniform yawed and inclined flows over the rotor area. e
where n, v 1;i and Ai denote the number of vertical segments, grid-
As mentioned in sub-section 2.1, the on-site measurement was
filtered instantaneous axial wind speed at height i, the rotor
done by means of a meteorological mast with two cup anemome- e
ters at 40 m and 60 m above the ground. Because of the terrain swept area and the area of ith segment associated with v 1;i .
complexity and canopies heterogeneity, the flow field would vary In the neutral ABL, the wind shear exponent mainly varies with
around each wind turbine in the wind farm. Therefore, one mete- altitude and surface roughness. The impact of the vertical wind
orological mast is not sufficient to accurately assess the wind shear due to the planetary boundary layer is more pronounced
resource across the wind farm. because of the cyclic behaviour of the rotor at high mean wind
To evaluate how much the flow field at the turbines’ location speed. A large vertical wind shear makes a considerable imbalance
differ from the met-mast location, the mean wind speed and tur- air loads acting on rotor blades resulting in significant fatigue loads
bulent kinetic energy profiles of each turbine are normalized by the and component failures, such as gearbox bearing [118].
corresponding profiles at the met-mast location i.e., sMM and kMM. The hub height and rotor radius of the existing wind turbines at
As can be seen in Fig. 18(a), except for the wind turbines WT1, WT7 Ro€ bergsfja
€llet site are 90 m and 45 m, respectively. For a turbine
and WT8, the normalized mean wind speed (s/sMM) along the (y) with large rotor diameter exposed to an unsteady and complex flow
axis is lower than the met-mast location. The normalized turbulent field, the wind profile across the rotor does not smoothly follow the
kinetic energy (k/kMM) shown in Fig. 18(b) demonstrates that all the power law. Therefore, following the IEC 61400-12e1:2017 standard
turbines, apart from WT1, are confronted by a higher turbulence [117], three heights has been chosen to compute the wind shear
than the met mast location. exponent over the entire rotor area, i.e., the bottom of rotor plane at
Generally, the wind shear and turbulent kinetic energy across 45 m, the hub height at 90 m and the top of the rotor plane at 135 m
the rotor have significant impact on aero-structural dynamics of a above the ground.
e
wind turbine. Apart from various atmospheric parameters, they are The instantaneous longitudinal velocity (v 1 ) is used to calculate
substantially driven by the terrain complexity and canopies het- the instantaneous wind shear exponent for the lower (a1) and the
erogeneity. Remarkable variation in vertical direction reveals that upper (a2) sections of the rotor plane on the basis of the power law
more detailed analysis must be done to quantify the impact of the equation given by [119].
mean wind speed and shear as well as the inflow turbulence on
power production and structural response of each turbine in a wind
e e 
a1 ¼ ln v 1;45 v 1;90
farm. Instead of single-point evaluation (e.g., hub height), the axial (11a)
lnðy45 =y90 Þ
mean wind speed for a period of 90 min is computed by the time-
integration of the rotor equivalent wind speed (UEq) to include the

818
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 16. Iso-surface (at hub height, 90 m above the ground) of (a) horizontal mean wind speed and (b) turbulent kinetic energy over the entire computational domain for the
€ bergsfja
Ro €llet wind farm, (d) and (e) zoom over the wind farm. (c) The wind farm elevation a.s.l and (f) trees height covered the wind farm. The blue dots represent the wind
turbine's location. MM denotes Met-Mast location too.

below 1% at each turbine location. Therefore, the hub height can be

e e  taken as the reference point.
a2 ¼ ln v 1;90 v 1;135
(11b) According to the logarithmic wind profile, the wind shear
lnðy90 =y135 Þ exponent (a) is a function of effective surface roughness length (y0)
Table 1 presents a summary of the axial mean wind speed at hub and elevation above the surface (y) given by a ¼ 1=lnðy=y0 Þ. The
e effective surface roughness length in the neutral ABL is strongly
height Cv 1 Dhub , the mean equivalent axial wind speed UEq, wind
shear exponents (a1 and a2) and the rotor equivalent turbulence affected by the ground topography and surface roughness element
intensity TIUEq statistics over the entire rotor for a period of 90 min. such as heterogeneous forest structure characterized by the trees
height and density [120]. In addition, the power law exponents,
Except WT7 which is the third from last (antepenultimate) in terms
collected for different types of terrains [121], vary from 0.1 for
of the altitude, UEq is related to the higher mean wind speed at
smooth surface and open water to 0.25 for many trees and hilly
higher elevation.
terrain. As observed, the mean shear exponents vary at each wind
The difference between the axial mean wind speed at hub
e turbine location justifying the impact of upstream ground
height Cv 1 Dhub and the mean equivalent axial wind speed UEq is
819
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

€bergsfj€allet site. (a) Mean wind speed profile, (b) Turbulent kinetic energy profile, (c) Mean veer angle
Fig. 17. Vertical variation of some wind characteristics for the turbines at Ro
profile and (d) Mean yaw angle profile. The rotor swept area limits and hub height at 90 m are illustrated by (,,,).

€bergsfja
Fig. 18. Vertical variation of (a) Normalized mean wind speed and (b) Normalized turbulent kinetic energy for the turbines at Ro €llet site. The rotor swept area limits and hub
height at 90 m are illustrated by the dotted line (,,,).

complexity indicated by effective surface roughness. The minimum rotor equivalent turbulence intensity TIUEq for all turbines are below
mean shear exponent occurs at the location of WT1 whereas WT5 13% which corresponds to the turbulence intensity of class C in IEC-
has the maximum one. The absolute difference between the mean 61400-1 standard [100] as shown in Fig. 5.
shear exponent at the lower and upper sections of the rotor area
defined as Da ¼ |a1  a2| may induce additional bending moment 4.5. Flow Evolution
acting on rotor blades. WT7 and WT2 are subjected to the highest
and lowest Da among all other turbines, respectively. The potential Wind turbines are operating within the Atmospheric Boundary
impact of Da on the bearing life of the turbines will be studied in Layer (ABL) ranging from ~ 150e1500 m [120]. Normally, the lowest
section 4.9. part of the ABL, the so-called surface layer extends up to ~10% of the
Except for WT1 and WT7 with minimum and WT2 with ABL height where the wind speed vary with height [122]. In neutral
maximum rotor equivalent turbulence intensity, other turbines are ABL, the air flow within the surface layer is characterized as a high-
exposed to the same level of inflow turbulence. Furthermore, the Reynolds number turbulent flow generated by large velocity
820
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 19. The horizontal mean wind speed magnitude SN (solid lines) and turbulent kinetic energy kN (dashed-lines), normalized by the associated values at 1000 m upstream the
rotor plane at three different heights above the ground. blue: bottom of the rotor plane (45 m), red: middle of the rotor plane (90 m) and black: top of the rotor plane (135 m). At
each wind turbine location, the ground elevation yN is also normalized by the altitude at 1000 m upstream the rotor plane.

821
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

gradients in the vertical direction. The generated turbulence is 4.7. Impact of surface roughness on turbulence
transported to the air stream through the upward surface layer
momentum flux. As previously mentioned in sub-section 2.1, in the In the neutral ABL, the turbulence inside the wind farm is mainly
neutral ABL, the mean wind speed profile and turbulent kinetic generated by the surface roughness affecting largely the aero-
energy strongly depend on the surface roughness i.e., terrain dynamics and structural dynamics of the rotor blades. However, the
complexity and canopies heterogeneity. In other words, the surface prediction of airflow and development of turbulent flow structures
roughness plays a major role in the motion of turbulent flow over the complex topography covered by non-uniform vegetation
structures inside the neutral ABL. and forest canopies (characterized by vertically varying Plant Area
Fig. 19 displays the variation of flow field at three different Density (PAD) and height) is challenging [94]. In neutral ABL can-
heights above the ground for the wind turbines’ at Ro€ bergsfja
€llet opy flow, the acceleration and deceleration of the mean flow
qffiffiffiffiffi2 2 mainly taking place at the forest edge is considered as the primary
e e
site.The horizontal mean wind speed (s ¼ Cv 1 D þ Cv D) and tur- source of the turbulent kinetic energy budget [124].The importance
3
bulent kinetic energy (k) at each elevation is normalized by the of the turbulent structures generated at the forest edge is due to the
associated values at 1000 m upstream the rotor plane as SN ¼ s/s1000 fact that they are convected to the upper layers of the air flow. Their
and KN ¼ k/k1000, respectively. It can be seen that SN rapidly react to size and strength are steadily grown up to satisfy the equilibrium
the terrain topography at low elevation up to the hub height at between the airflow and surface roughness which accordingly
90 m. affect the flow passing through the rotor area. Moreover, in neutral
The impact of the windward slope measured between the tur- ABL flow over the hilly terrain covered by the heterogeneous forest,
bines’ location and 1000 m upstream on SN is obvious. Apart from the horizontal variation of PAD, the maximum PAD height and the
WT2 and WT4 with negative windward slope, WT1, WT3, WT5 and separated flow by the ground topography increase the difficulties of
WT6 are located on a very gentle slope showing a smooth speed up. numerical modelling of wind flow inside and above the forest
The maximum speed up occurs for WT7 and WT8, respectively especially on the turbulence generation mechanisms. Fig. 21
because of their moderate windward slopes. demonstrates the turbulence generation caused by the complex
The steep-sided valleys, located upstream WT4, WT5 and WT6 terrain and heterogeneous forest at each wind turbine's location
and downstream of WT8, decrease SN at low elevation. This within the Ro €bergsfj€
allet wind farm. As seen, the turbulent kinetic
reduction can be partially recovered by a very strong slope after a energy below the forest edge is low and grows rapidly towards the
steep-sided valley. Furthermore, recirculation regions occur at edge of the forest. Moreover, in some locations, the turbulence
these steep-sided valleys. These regions are characterized by low generation is much higher where they are mostly associated with
mean wind speed (s) and high turbulent kinetic energy (k). Unlike the denser forest edges [125] and steep-sided valleys. At the denser
the mean wind speed, the sensitivity of turbulent kinetic energy to forest edge, the strong shear due to the acceleration and deceler-
the terrain topography extends above the hub height denoting the ation of the mean flow is the main source for the production of the
motion of large-scales turbulent structures inside the wind farm. In turbulent kinetic energy. In steep-sided valleys corresponding to
addition to the terrain complexity, the heterogeneity of forest the recirculation region, the production of turbulent kinetic energy
canopies has a considerable effect on the turbulence generation occurs at separation zone which is slightly before the valley trough.
mechanism especially at the forest edge. For instance, the sudden
change in kN, as seen in Fig. 19(e) and (g), occurs at a constant 4.8. Comparison of short-term damage equivalent turbine loads
ground elevation and it can be translated to the abrupt change in
surface roughness caused by the forest properties. A summary of the short-term damage equivalent loads (DELs)
for Blade 1 root flapwise bending moment (Bl1RootFlpMom),
Tower base fore-aft bending moment (TwrBsFAMom) and Tower
top yaw bearing moment (TTYawMom), obtained from numerical
simulations, are provided in Table 2. These quantities are obtained
4.6. Deflection and inclination of mean flow from 90 min time-history predictions obtained from FAST. It can be
observed that the damage equivalent loads are higher, on an
The airflow within a wind farm is distorted by the surface average, under the homogeneous forest assumption. The main
roughness heterogeneity including complex topography and reason for this increase is the higher turbulence in the flow-field
ground's heterogeneous vegetation. The deflected airflow have a under the homogeneous forest assumption. Generally, in the ho-
large impact on the aerodynamics of rotor blades. Fig. 20 presents mogeneous forest assumption, the ground is uniformly covered by
the streamlines of the vertical and lateral mean wind velocities the forest canopies with the same PAD profile. Therefore, additional
around the rotor planes. The colorbar indicates the magnitude of turbulence due to forest canopies are generated everywhere in the
the peripheral mean wind speed normalized by the axial mean computational domain through the drag term in the momentum
wind speed. As seen, the peripheral mean wind speed magnitude is equation (Eq. (1b)). However, in the heterogeneous forest
the largest at WT2 and the lowest at WT7. The ratio between the assumption, because of the scattered forest canopies with varying
peripheral and axial mean wind speeds is limited by 0.12, however PAD profiles, the drag term in the momentum equation (Eq. (1b)) is
variation in turbulent flow structures can be clearly identified. In more likely to be smaller than the homogeneous forest which in
particular, the flow structure over the swept rotor of WT4, WT6 and turn produce less turbulence. It must be noted here that although
WT7 dramatically change in both horizontal and vertical directions the DELs reported here are short-term, they provide an indication
which makes a flow turning across the rotor. Generally, the hori- of structural load trends presented by the two different terrain
zontal and vertical misalignment of the incoming flow with respect assumptions. It is also important to note that these results are site
to the rotor axis, the so-called yawed flow and inclined flow, specific. For another site, the turbulence intensity associated with
respectively induce additional load and moments at rotor blades heterogeneous forest (in case of higher forest density) assumption
[123]. But the mean flow turning over the entire rotor (like WT4, may be higher than homogeneous forest. This only emphasises the
WT6 and WT7) causes the upper and lower halves of the rotor importance of site-specific analysis and heterogeneous modelling
blades to expose to an undesirably imbalanced load and moments of the terrain to achieve more accurate prediction of structural
distribution. loads and service life.
822
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

e e
Fig. 20. Streamlines ofvertical Cv22 Dand spanwise Cv 3 D mean wind speed around the rotor plane perpendicular to the dominant wind direction (marked by €bergsfj€allet
) at Ro
e2 e 0:5 e e
wind farm colored by Cv D þ Cv D =Cv 1 D. The axial mean wind speed Cv 1 D is perpendicular to the rotor plane.
2 3

4.9. Comparison of high-speed shaft cylindrical roller bearing life unlike the structural loads, the bearing life predictions are higher
with a homogeneous forest assumption.The primary reason re-
A summary of the high-speed shaft cylindrical bearing lives for mains the same. Due to the increased turbulence intensity pre-
the 8 turbines are provided in Table 3. It can be observed here that dicted by the homogeneous forest, the turbine spends less time at

823
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Fig. 21. Turbulent kinetic energy (k) at each turbine's location shown by the colorbar on the left, sub-figures (a) to (h) are associated with WT1 to WT8 respectively. : Rotor
plane and : Vertical edge of the trees. The highest peaks (limited to 40 m) on the lines represent the wind turbine towers detected by the Airborne Laser Scanning
(ALS) measurements as obstruction/vegetation on the terrain.

Table 1
Mean and Standard Deviation (STD) of wind profile properties, over the rotor swept area and at hub height.

Item Elevation [m] Mean STD TIUEq

e e
UEq [m/s] Cv 1 Dhub [m/s] a1 [] a2 [] UEq [m/s] Cv 1 Dhub [m/s] []

WT 1 527.1 10.32 10.31 0.12 0.14 1.07 1.30 0.10

WT 2 508.3 9.74 9.73 0.19 0.21 1.29 1.57 0.13
WT 3 507.8 9.61 9.66 0.28 0.23 1.11 1.44 0.12
WT 4 519.7 9.74 9.77 0.21 0.18 1.13 1.42 0.12
WT 5 523.6 9.76 9.81 0.35 0.23 1.17 1.44 0.12
WT 6 526.6 9.79 9.81 0.23 0.20 1.15 1.45 0.12
WT 7 510.2 10.42 10.50 0.23 0.10 1.06 1.26 0.10
WT 8 531.1 10.25 10.35 0.18 0.12 1.21 1.32 0.12

824
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

Table 2
Comparison of short term damage equivalent loads on wind turbines.

Wind Turbine # Bl1RootFlpMom DEL [kNm] TwrBsFAMom DEL [kNm] TTYawMom DEL [kNm]

Het. Hom. % diff. Het. Hom. % diff. Het. Hom. % diff.

WT 1 520 546 4.8 2210 2190 0.9 250 313 20.1

WT 2 511 585 12.6 2320 2430 4.5 290 355 18.3
WT 3 498 560 11.1 2180 2320 6.0 289 357 19.1
WT 4 538 522 3.1 2170 2240 3.1 272 343 20.7
WT 5 610 607 0.5 2180 2430 10.3 335 409 18.1
WT 6 500 592 15.5 2140 2490 14.1 291 360 19.2
WT 7 553 559 1.1 2360 2250 4.9 262 303 13.5
WT 8 544 576 5.6 2260 2280 0.9 249 294 15.3

Table 3 Comparison between Table 1 and Table 3 reveals that a higher

Comparison of bearing life. equivalent mean wind speed reduces the life time of the turbine
Wind Turbine # Bearing life [million revolutions] bearings. Moreover, a higher wind shear exponent for the upper
sections of the rotor plane (a2) also has a negative impact on the life
Het. Hom. % diff.
time of the turbine bearings. By putting Fig. 20 beside Table 3, it
WT 1 1056 1259 16.1
seems that horizontal flow misalignment (yawed flow, as in WT5)
WT 2 1364 1881 27.5
WT 3 1473 1619 9.0 has lesser impact than the vertical flow misalignment (inclined
WT 4 1330 1848 28.0 flow, as in WT2 and WT4) on the life time of the turbine's bearing
WT 5 1377 1718 19.9 knowing that WT2 and WT4 have a lower equivalent mean wind
WT 6 1342 1788 24.9 speed than WT5.
WT 7 1008 1132 11.0
WT 8 1113 1239 10.2
5. Conclusion

The results show that for a forested hilly terrain turbine-specific

structural and drive-train loads can be quite different within a wind
farm. Therefore, a single met mast data might not be a real repre-
sentative for the wind field of entire wind farm.
Variation of inflow variables such as mean wind speed, shear
exponent and turbulence intensity at each wind turbine location
justifies the need for high-fidelity CFD method - the so-called
Large-Eddy Simulation (LES) - to accurately model the airflow in-
side and over complex terrains and around each wind turbine in a
wind farm, despite the expensive computational cost.
The nacelle-mounted anemometer data from SCADA reports a
lower mean wind speed for all turbines than LES. However, for a
more restrictive set of wind conditions, that better represents
neutral atmospheric conditions, the SCADA data are closer to the
predicted wind speeds by LES. The more restrictive samples of
SCADA data also have a mean value closer to the turbulence in-
tensity (TI) predicted by LES. The difference between the predicted
TI by LES and SCADA data is larger for the turbines that are placed in
the wake of upstream wind turbine. This shows the necessity of
Fig. 22. Histogram of WT 7 HSS-CRB radial bearing force for the heterogeneous and inclusion of wind turbine model (for example, actuator disk or
homogeneous forest modelings. actuator line model) to capture more detailed description of wind
field within a wind farm, especially for aeroelastic calculation.
For the wind farm considered in this study, the predicted tur-
high operating load ranges as can be observed in Fig. 22 for wind
bulence intensity under the homogeneous forest assumption is
turbine 7. All the other turbines experience similar features hence
greater than the heterogeneous forest assumption. This leads to
the radial bearing force of only wind turbine 7 is shown here for
higher damage equivalent loads, computed for the blades and
demonstrative purposes. As mentioned before, bearings consume
tower for the homogeneous forest assumption. Hence, the impor-
their remaining life at a much faster rate during high load operation
tance of site-specific analysis and heterogeneous forest modelling
compared to lower operational loads. Therefore, a modest decrease
of the terrain to achieve more accurate prediction of structural
in time spent at high operating speeds/loads results in a significant
loads and service life is highlighted. On the other hand, the bearing
increase in remaining bearing life and vise-versa. It must again be
life predictions are higher for the homogeneous forest assumption.
noted that these results are site-specific and provide only an indi-
This is because the wind turbine, exposed to a higher turbulence
cation of the trends. It becomes clear that wrongful estimation of
intensity, spends less time at high operating ranges resulting in an
atmospheric turbulence can have a big impact on bearing lives. As
increase in remaining bearing life. Note here that the bearing life
we observed, turbine located at places with little vegetation and a
model used is idealized and does not take load transients into ac-
more uniform flow will consume their bearing life at a faster rate
count. In addition, it seems that horizontal flow misalignment
compared to turbines located in forests. This again emphasises the
(yawed flow) has a less negative impact than the vertical flow
importance of terrain-specific modelling of the flow in a wind farm.
misalignment (inclined flow) on the life time of the turbine's
825
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

bearing. This claim, however, requires further study and opens up a european wind atlas, Phil. Trans. Math. Phys. Eng. Sci. 375 (2091) (2017)
20160101.
new research question.
[8] M.K. Liu, M.A. Yocke, Siting of wind turbine generators in complex terrain,
The investigation into the turbine loads presented in this paper J. Energy 4 (1) (1980).
shows that there is a considerable difference in the turbine loads [9] R.J.A.M. Stevens, C. Meneveau, Flow structure and turbulence in wind farms,
and especially the HSS bearing life between the two types of forest Annu. Rev. Fluid Mech. 49 (1) (2017) 311e339.
[10] N. Wood, Wind flow over complex terrain: a historical perspective and the
assumptions. There is also significant inter-turbine differences in prospect for large-eddy modelling, Boundary-Layer Meteorol. 96 (1) (2000)
the loads experienced by the turbines in the same wind farm. These 11e32.
differences are attributed to the terrain around the individual tur- [11] P.S. Jackson, J.C.R. Hunt, Turbulent wind flow over a low hill, Q. J. R. Meteorol.
Soc. 101 (430) (1975) 929e955.
bines that can vary significantly in the same farm. Thus empha- [12] P.J. Mason, R.I. Sykes, Flow over an isolated hill of moderate slope, Q. J. R.
sising the fact that proper modeling of the flow considering the Meteorol. Soc. 105 (444) (1979) 383e395.
heterogeneous terrain is required for better estimation of turbine- [13] P.J. Mason, J.C. King, Measurements and predictions of flow and turbulence
over an isolated hill of moderate slope, Q. J. R. Meteorol. Soc. 111 (468)
specific fatigue life of its components as turbine-specific fatigue life (1985) 617e640.
can vary significantly from the average wind turbine fatigue life in a [14] J. Sanz Rodrigo, R.A. Ch avez Arroyo, P. Moriarty, M. Churchfield, B. Kosovi
c,
farm. P.E. Re thore, K.S. Hansen, A. Hahmann, J.D. Mirocha, D. Rife, Mesoscale to
microscale wind farm flow modeling and evaluation, WIREs Energy and
Environment 6 (2) (2017) e214.
CRediT authorship contribution statement [15] R.E. Britter, J.C.R. Hunt, K.J. Richards, Air flow over a two-dimensional hill:
studies of velocity speed-up, roughness effects and turbulence, Wind Energy
107 (451) (1981) 91e110.
Hamidreza Abedi: Conceptualization, Methodology, Software, [16] R.L. Simpson, Turbulent boundary-layer separation, Annu. Rev. Fluid Mech.
Validation, Formal analysis, Data curation, Writing e original draft, 21 (1) (1989) 205e232.
Visualization. Saptarshi Sarkar: Conceptualization, Methodology, [17] L. Davidson, Fluid Mechanics, Turbulent Flow and Turbulence Modeling,
Lecture Notes; Department of Mechanics and Maritime Sciences, Division of
Software, Validation, Formal analysis, Data curation, Writing e Fluid Dynamics, Chalmers University of Technology, 2020.
original draft, Visualization. Håkan Johansson: Conceptualization, [18] S. Feng, X. Zheng, R. Hu, P. Wang, Large eddy simulation of high-Reynolds-
Methodology, Software, Validation, Formal analysis, Data curation, number atmospheric boundary layer flow with improved near-wall correc-
tion, Boundary-Layer Meteorol. 41 (2020) 33e35.
Writing e original draft, Writing e review & editing, Visualization. [19] H.W. Detering, D. Etling, Application of the e-e turbulence model to the at-
mospheric boundary layer, Boundary-Layer Meteorol. 33 (2) (1985)
Declaration of competing interest 113e133.
[20] O. Zeman, N.O. Jensen, Modification of turbulence characteristics in flow over
hills, Q. J. R. Meteorol. Soc. 113 (475) (1987) 55e80.
The authors declare that they have no known competing [21] D.D. Apsley, I.P. Castro, A limited-length-scale k-e model for the neutral and
financial interests or personal relationships that could have stably-stratified atmospheric boundary layer, Boundary-Layer Meteorol. 83
(1997) 75e98.
appeared to influence the work reported in this paper. [22] D. Xu, P.A. Taylor, An E-εel turbulence closure scheme for planetary
boundary-layer models: the neutrally stratified case, Boundary-Layer
Acknowledgments Meteorol. 84 (2) (1997) 247e266.
[23] J. O'Sullivan, Modelling Wind Flow over Complex Terrain, PhD thesis, Uni-
versity of Auckland, New Zealand, 2012.
This project is financed through the Swedish Wind Power [24] S. Dupont, Y. Brunet, J.J. Finnigan, Large-eddy simulation of turbulent flow
Technology Centre (SWPTC). SWPTC is a research centre for the over a forested hill: validation and coherent structure identification, Q. J. R.
Meteorol. Soc. 134 (636) (2008) 1911e1929.
design of wind turbines. The purpose of the centre is to support [25] A. El Kasmi, C. Masson, Turbulence modeling of atmospheric boundary layer
Swedish industry with knowledge of design techniques as well as flow over complex terrain: a comparison of models at wind tunnel and full
maintenance in the field of wind power. The Centre is funded by the scale, Wind Energy 13 (8) (2010) 689e704.
[26] S. Shamsoddin, F. Porte -Agel, Large-Eddy simulation of atmospheric
Swedish Energy Agency, Chalmers University of Technology as well boundary-layer flow through a wind farm sited on topography, Boundary-
as academic and industrial partners. Layer Meteorol. 163 (2017) 1e17.
The computations were enabled by resources provided by the [27] M. Cindori, I. D zijan, F. Jureti
c, H. Kozmar, The atmospheric boundary layer
above generic hills: computational model of a unidirectional body force-
Swedish National Infrastructure for Computing (SNIC) at Chalmers driven flow, Boundary-Layer Meteorol. 176 (2) (2020) 159e196.
Centre for Computational Science and Engineering (C3SE) partially [28] G.T. Bitsuamlak, T. Stathopoulos, C. Be dard, Numerical evaluation of wind
funded by the Swedish Research Council through grant agreement flow over complex terrain: Review, J. Aero. Eng. 17 (4) (2004) 135e145.
[29] R.J. Barthelmie, E.S. Politis, J.M. Prospathopoulos, S.T. Frandsen, K. Rados,
No. 2018e05973.
O. Rathmann, K.S. Hansen, W. Schlez, D. Cabezon, J.D.A. Phillips, A. Neubert,
Software to convert.las data was provided by J, Arnqvist, and is S. Pijl, J.G. Schepers, Flow and Wakes in Large Wind Farms in Complex
available at https://github.com/johanarnqvist/ALS2PAD.git. Terrain and Offshore, European Wind Energy Conference, Brussels, Belgium,
2008.
[30] P. Brodeur, C. Masson, Numerical site calibration over complex terrain, J. Sol.
References Energy Eng. 130 (3) (2008).
[31] S.J. Wakes, T. Maegli, K.J. Dickinson, M.J. Hilton, Numerical modelling of wind
[1] A.S. Monin, The atmospheric boundary layer, Annu. Rev. Fluid Mech. 2 (1) flow over a complex topography, Environ. Model. Software 25 (2) (2010)
(1970) 225e250. 237e247.
[2] J.C. Wyngaard, Atmospheric turbulence, Annu. Rev. Fluid Mech. 24 (1) (1992) [32] E. Berge, A.R. Gravdahl, J. Schelling, L. Tallhaug, O. Undheim, Wind in Com-
205e234. plex Terrain . A Comparison of WAsP and Two CFD-Models, Proceedings

[3] M.A.C. Teixeira, D.J. Kirshbaum, H. Olafsson, P.F. Sheridan, I. Stiperski, The EWEC, Athens, Greece, 2006.
atmosphere over mountainous regions, Front. Earth Sci. 4 (2016) 84. [33] D.C. Martínez, K.S. Hansen, R.J. Barthelmie, Analysis and Validation of Cfd
[4] Y. Xi, C. Meneveau, Modelling turbulent boundary layer flow over fractal-like Wind Farm Models in Complex Terrain. Effects Induced by Topography and
multiscale terrain using large-eddy simulations and analytical tools, Philos Wind Turbines, 2010.
Trans A Math Phys Eng Sci 375 (2091) (2017) 20160098. [34] E.S. Politis, J. Prospathopoulos, D. Cabezon, K.S. Hansen, P.K. Chaviaropoulos,
[5] M. Lehner, M. Rotach, Current challenges in understanding and predicting R.J. Barthelmie, Modeling wake effects in large wind farms in complex
transport and exchange in the atmosphere over mountainous terrain, At- terrain: the problem, the methods and the issues, Wind Energy 15 (1) (2012)
mosphere 9 (7) (2018) 276. 161e182.
[6] J. Ruel, D. Pin, K. Cooper, Effect of topography on wind behaviour in a [35] L. Li, P.W. Chan, L. Zhang, F. Hu, Numerical simulation of a lee wave case over
complex terrain, Forestry: An International Journal of Forest Research 71 (3) three-dimensional mountainous terrain under strong wind condition, Adv.
(1998) 261e265. Meteorol. 2013 (2013), https://doi.org/10.1155/2013/304321.
[7] J. Mann, N. Angelou, J. Arnqvist, D. Callies, E. Cantero, R.C. Arroyo, [36] J.S.R. R Cha vez Arroyo, P. Gankarski, Modelling of atmospheric boundary-
M. Courtney, J. Cuxart, E. Dellwik, J. Gottschall, S. Ivanell, P. Kühn, G. Lea, layer flow in complex terrain with different forest parameterizations,
J.C. Matos, J.M.L.M. Palma, L. Pauscher, A. Pen~ a, J.S. Rodrigo, S. So
€ derberg, J. Phys. Conf. 524 (2014), 012119.
N. Vasiljevic, C.V. Rodrigues, Complex terrain experiments in the new [37] B. Blocken, A. van der Hout, J. Dekker, O. Weiler, CFD simulation of wind flow

826
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

over natural complex terrain: case study with validation by field measure- Layer Meteorol. 141 (2) (2011) 245e271.
ments for Ria de Ferrol, Galicia, Spain, J. Wind Eng. Ind. Aerod. 147 (2015) [66] M. Diebold, C. Higgins, J. Fang, A. Bechmann, M. Parlange, Flow over hills: a
43e57. large-eddy simulation of the Bolund case, Boundary-Layer Meteorol. 148
[38] B.W. Yan, Q.S. Li, Y.C. He, P.W. Chan, RANS simulation of neutral atmospheric (2013) 177e194.
boundary layer flows over complex terrain by proper imposition of boundary [67] V. Vuorinen, A. Chaudhari, J.P. Keskinen, Large-eddy simulation in a complex
conditions and modification on the k-ε model, Environ. Fluid Mech. 16 hill terrain enabled by a compact fractional step openfoam® solver, Adv. Eng.
(2016) 1e23. Software 79 (2015) 70e80.
[39] S. Ivanell, J. Arnqvist, M. Avila, D. Cavar, R.A. Chavez-Arroyo, H. Olivares- [68] A. Chaudhari, Large-Eddy Simulation of Wind Flows over Complex Terrains
Espinosa, C. Peralta, J. Adib, B. Witha, Micro-scale model comparison for Wind Energy Applications, PhD thesis, Lappeenranta University of
(benchmark) at the moderately complex forested site, Wind Energy Science Technology, Finland, 2014.
3 (2) (2018) 929e946. [69] A. Chaudhari, A. Hellsten, J. Ha €m€al€
ainen, Full-scale experimental validation
[40] P.A. Taylor, H.W. Teunissen, The Askervein hill project: overview and back- of large-eddy simulation of wind flows over complex terrain: the Bolund hill,
ground data, Boundary-Layer Meteorol. 39 (1) (1987) 15e39. Adv. Meteorol. 2016 (2016) 1e14, https://doi.org/10.1155/2016/9232759.
[41] A. Bechmann, J. Berg, M.S. Courtney, H.E. Jørgensen, J. Mann, N.N. Sørensen, [70] B. Conan, A. Chaudhari, S. Aubrun, J. van Beeck, J. Ha €ma
€l€
ainen, A. Hellsten,
The Bolund Experiment: Overview and Background. Technical Report, Risø- Experimental and numerical modelling of flow over complex terrain: the
R-1658(EN); Risø DTU, National Laboratory for Sustainable Energy, Technical Bolund hill, Boundary-Layer Meteorol. 158 (2) (2016) 183e208.
University of Denmark, 2009. [71] Y. Ma, H. Liu, Large-Eddy simulations of atmospheric flows over complex
[42] J. Berg, J. Mann, M.S. Courtney, H.E. Jørgensen, P.E. Re thore, The Bolund terrain using the immersed-boundary method in the weather research and
experiment, part i: flow over a steep, three-dimensional hill, Boundary-Layer forecasting model, Boundary-Layer Meteorol. 421 (3) (2017) 421e445.
Meteorol. 141 (2) (2011) 219e243. [72] A. Maurizi, J. Palma, F. Castro, Numerical simulation of the atmospheric flow
[43] G.D. Raithby, G.D. Stubley, P.A. Taylor, The Askervein hill project: a finite in a mountainous region of the north of Portugal, J. Wind Eng. Ind. Aerod.
control volume prediction of three-dimensional flows over the hill, Bound- 74e76 (1998) 219e228.
ary-Layer Meteorol. 39 (3) (1987) 247e267. [73] H.G. Kim, V. Patel, C.M. Lee, Numerical simulation of wind flow over hilly
[44] S.L.A. Castro F Palma J, Simulation of the Askervein flow. part 1: Reynolds terrain, J. Wind Eng. Ind. Aerod. 87 (1) (2000) 45e60.
averaged Navier-Stokes equations(k-ε turbulence model), Boundary-Layer [74] F.K. Chow, A.P. Weigel, R.L. Street, M.W. Rotach, M. Xue, High-resolution
Meteorol. 107 (3) (2003) 501e530. large-eddy simulations of flow in a steep alpine valley. Part I: methodology,
[45] K.J. Eidsvik, A system for wind power estimation in mountainous terrain. verification, and sensitivity experiments, J. Appl. Meteorol. Climatol. 45 (1)
prediction of Askervein hill data, Wind Energy 8 (2) (2005) 237e249. (2006) 63e86.
[46] O. Undheim, H.I. Andersson, E. Berge, Non-linear, microscale modelling of [75] A.P. Weigel, F.K. Chow, M.W. Rotach, R.L. Street, M. Xue, High-resolution
the flow over Askervein hill, Boundary-Layer Meteorol. 120 (3) (2006) large-eddy simulations of flow in a steep alpine valley. Part II: flow structure
477e495. and heat budgets, J. Appl. Meteorol. Climatol. 45 (1) (2006) 87e107.
[47] D. Cavar, P.E. Re thore
, A. Bechmann, N.N. Sørensen, B. Martinez, F. Zahle, [76] J.C.P.L. da Costa, Atmospheric Flow over Forested and Non-forested Complex
J. Berg, M.C. Kelly, Comparison of openfoam and ellipsys3d for neutral at- Terrain, PhD thesis, Universidade do Porto, Portugal, 2008.
mospheric flow over complex terrain, Wind Energy Science 1 (1) (2016) [77] Y. Han, M. Stoellinger, J. Naughton, Large eddy simulation for atmospheric
55e70. boundary layer flow over flat and complex terrains, J. Phys. Conf. 753 (2016),
[48] J. Prospathopoulos, E. Politis, P. Chaviaropoulos, Application of a 3d rans 032044.
solver on the complex hill of Bolund and assessment of the wind flow pre- [78] LES study on the turbulent flow fields over complex terrain covered by
dictions, J. Wind Eng. Ind. Aerod. 107e108 (2012) 149e159. vegetation canopy, J. Wind Eng. Ind. Aerod. 155 (2016) 60e73.
[49] P.J. Mason, D.J. Thomson, Stochastic backscatter in large-eddy simulations of [79] X. Yang, M. Pakula, F. Sotiropoulos, Large-eddy simulation of a utility-scale
boundary layers, J. Fluid Mech. 242 (1992) 51e78. wind farm in complex terrain, Appl. Energy 229 (2018) 767e777.
[50] P.P. Sullivan, J.C. McWilliams, C.H. Moeng, A subgrid-scale model for large- [80] J. Berg, N. Troldborg, N.S.E.G.P, P.P. Sullivan, Modelling of atmospheric
eddy simulation of planetary boundary-layer flows, Boundary-Layer Mete- boundary-layer flow in complex terrain with different forest parameteriza-
orol. 71 (3) (1994) 247e276. tions, J. Phys. Conf. 854 (2017), 012003.
[51] -Agel, C. Meneveau, M. Parlange, A scale-dependent dynamic model
F. Porte [81] P. Enevoldsen, Onshore wind energy in northern european forests: review-
for large-eddy simulation: application to a neutral atmospheric boundary ing the risks, Renew. Sustain. Energy Rev. 60 (2016) 1251e1262.
layer, J. Fluid Mech. 415 (2000) 261e284. [82] A two-equation turbulence model for canopy flows, J. Wind Eng. Ind. Aerod.
[52] F. Wan, F. Porte -Agel, R. Stoll, Evaluation of dynamic subgrid-scale models in 35 (1990) 201e211.
large-eddy simulations of neutral turbulent flow over a two-dimensional [83] R.H. Shaw, U. Schumann, Large-eddy simulation of turbulent flow above and
sinusoidal hill, Atmos. Environ. 41 (13) (2007) 2719e2728. within a forest, Boundary-Layer Meteorol. 61 (1992) 47e64.
[53] A. Brown, J. Hobson, N. Wood, Large-Eddy simulation of neutral turbulent [84] A. Lopes, J.M.L.M. Palma, J.V. Lopes, Improving a two-equation turbulence
flow over rough sinusoidal ridges, Boundary-Layer Meteorol. 98 (2) (2001) model for canopy flows using large-eddy simulation, Boundary-Layer
411e441. Meteorol. 149 (2) (2013) 231e257.
[54] T. Allen, A. Brown, Large-Eddy simulation of turbulent separated flow over [85] E. Mueller, W. Mell, A. Simeoni, Large eddy simulation of forest canopy flow
rough hills, Boundary-Layer Meteorol. 102 (2) (2002) 177e198. for wildland fire modeling, Can. J. For. Res. 44 (12) (2014) 1534e1544.
[55] T. Tamura, S. Cao, A. Okuno, LES study of turbulent boundary layer over a [86] B. Nebenführ, L. Davidson, Prediction of wind-turbine fatigue loads in forest
smooth and a rough 2d hill model, Flow, Turbul. Combust. 79 (4) (2007) regions based on turbulent LES inflow fields, Wind Energy 20 (6) (2017)
405e432. 1003e1015.
[56] S. Cao, T. Wang, Y. Ge, Y. Tamura, Numerical study on turbulent boundary [87] B. Nebenführ, L. Davidson, Prediction of wind turbine fatigue loads in forest
layers over two-dimensional hills-effects of surface roughness and slope, regions based on turbulent LES inflow fields, J Wind Energy 20 (2017)
J. Wind Eng. Ind. Aerod. 104e106 (2012) 342e349. 1003e1015.
[57] M.J. Churchfield, S. Lee, J. Michalakes, P.J. Moriarty, A numerical study of the [88] E. Patton, Large-Eddy Simulation of Turbulent Flow above and within a Plant
effects of atmospheric and wake turbulence on wind turbine dynamics, Canopy, PhD thesis, University of California Davis, 1997.
J. Turbul. 13 (2012) N14. [89] M. Mohr, J. Arnqvist, H. Abedi, H. Alfredsson, M. Baltscheffsky, H. Bergstro € m,
[58] S.E. Belcher, I.N. Harman, J.J. Finnigan, The wind in the willows: flows in I. Carlen, L. Davidson, A. Segalini, S. So € derberg, Wind power in forests ii,
forest canopies in complex terrain, Annu. Rev. Fluid Mech. 44 (1) (2012) Report 499 (2018). Energiforsk; 2018.
479e504. [90] E.G. Patton, P.P. Sullivan, K.J. Davis, The influence of a forest canopy on top-
[59] X. Yang, F. Sotiropoulos, R.J. Conzemius, J.N. Wachtler, M.B. Strong, Large- down and bottom-up diffusion in the planetary boundary layer, Q. J. R.
eddy simulation of turbulent flow past wind turbines/farms: the virtual wind Meteorol. Soc. 129 (590) (2003) 1415e1434.
simulator (vwis), Wind Energy 18 (12) (2015) 2025e2045. [91] J. Arnqvist, J. Freier, E. Dellwik, Robust processing of airborne laser scans to
[60] S.P. Breton, J. Sumner, J.N. Sørensen, K.S. Hansen, S. Sarmast, S. Ivanell, plant area density profiles, Biogeosci. Discuss. 2020 (2020) 1e19.
A survey of modelling methods for high-fidelity wind farm simulations using [92] M. Weiss, F. Baret, G. Smith, I. Jonckheere, P. Coppin, Review of methods for
large eddy simulation, Phil. Trans. Math. Phys. Eng. Sci. 375 (2091) (2017) in situ leaf area index (LAI) determination: Part II. Estimation of LAI, errors
20160097. and sampling, Agric. For. Meteorol. 121 (1) (2004) 37e53.
[61] A. Silva Lopes, J.M.L.M. Palma, F.A. Castro, Simulation of the Askervein flow.  Boudreault, A. Bechmann, N.N. Sørensen, A. Sogachev, E. Dellwik, Canopy
[93] L.E.
part 2: large-eddy simulations, Boundary-Layer Meteorol. 125 (2007) structure effects on the wind at a complex forested site, J. Phys. Conf. 524
85e108. (2014), 012112.
[62] F.K. Chow, R.L. Street, Evaluation of turbulence closure models for large-eddy  Boudreault, Reynolds-averaged Navier-Stokes and Large-Eddy Simula-
[94] L.E.
simulation over complex terrain: flow over Askervein hill, J. Appl. Meteorol. tion over and inside Inhomogeneous Forests, PhD thesis, Denmark Technical
Climatol. 48 (5) (2009) 1050e1065. University, 2015.
[63] J.C. Golaz, J.D. Doyle, S. Wang, One-way nested large-eddy simulation over [95] M. Monsi, T. Saeki, On the factor light in plant communities and its impor-
the Askervein hill, J. Adv. Model. Earth Syst. 1 (3) (2009). tance for matter production, Ann. Bot. 95 (3) (2005) 549e567.
[64] A. Bechmann, Large-eddy Simulation of Atmospheric Flow over Complex 
[96] L. Etienne Boudreault, A. Bechmann, L. Tarvainen, L. Klemedtsson,
Terrain, PhD thesis, Denmark Technical University, 2006. I. Shendryk, E. Dellwik, A lidar method of canopy structure retrieval for wind
[65] A. Bechmann, N.N. Sørensen, J. Berg, J. Mann, P.E. Re thore, The Bolund modeling of heterogeneous forests, Agric. For. Meteorol. 201 (2015) 86e97.
experiment, part ii: blind comparison of microscale flow models, Boundary- [97] A.N. Ross, Boundary-layer flow within and above a forest canopy of variable

827
H. Abedi, S. Sarkar and H. Johansson Renewable Energy 180 (2021) 806e828

density, Q. J. R. Meteorol. Soc. 138 (666) (2012) 1259e1272. 1063(EN) in Denmark, Forskningscenter Risø. Risø-R, 1998.
[98] J.M. Jonkman, M.L. Buhl Jr., et al., Fast User's Guide vol. 365, National [112] A. Crespo, J. Hernandez, S. Frandsen, Survey of modelling methods for wind
Renewable Energy Laboratory, Golden, CO, 2005, p. 366. turbine wakes and wind farms, Wind Energy 2 (1) (1999) 1e24.
[99] J. Arnqvist, Software to convert airborne laser scans to plant area density [113] G. Hayman, Mlife Theory Manual for Version 1.00, National Renewable En-
(als2pad). https://github.com/johanarnqvist/ALS2PAD.git, 2020. ergy Laboratory (NREL), 2012.
[100] IEC 61400-1:2019, Wind Energy Generation Systems - Part 1: Design Re- [114] E. Artigao, S. Martín-Martínez, A. Honrubia-Escribano, E. Go mez-Lazaro,
quirements. Tech. Rep. 61400-1:2019, International Electrotechnical Com- Wind turbine reliability: a comprehensive review towards effective condi-
mission, 2019. tion monitoring development, Appl. Energy 228 (2018) 1569e1583.
[101] P. Richards, R. Hoxey, Appropriate boundary conditions for computational [115] S. Sheng, Wind Turbine Gearbox Reliability Database, Condition Monitoring,
wind engineering models using the k-ε turbulence model, J. Wind Eng. Ind. and Operation and Maintenance Research Update. Tech. Rep, National
Aerod. 46e47 (1993) 145e153. Renewable Energy Lab.(NREL), Golden, CO (United States), 2016.
[102] L. Breuer, K. Eckhardt, H.G. Frede, Plant parameter values for models in [116] SKF, SKF Bearing Maintenance Handbook, 2011. https://www.skf.com/
temperate climates, Ecol. Model. 169 (2) (2003) 237e293. binaries/pub12/Images/0901d1968013be94-SKF-bearing-maintenance-
[103] R.H. Shaw, G. Den Hartog, H.H. Neumann, Influence of foliar density and handbook—10001_1-EN(1)_tcm_12-463040.pdf. (Accessed 23 April 2021).
thermal stability on profiles of Reynolds stress and turbulence intensity in a [117] IEC 61400-12-1:2017, Wind Energy Generation Systems - Part 12-1: Power
deciduous forest, Boundary-Layer Meteorol. 45 (4) (1988) 391e409. Performance Measurements of Electricity Producing Wind Turbines. Tech.
[104] J. Smagorinsky, General circulation experiments with the primitive equa- Rep. 61400-12-1:2017, International Electrotechnical Commission, 2017.
tions. part I, the basic experiment, Mon. Weather Rev. 91 (3) (1963) 99e164, [118] N. Kelley, B. Smith, K. Smith, G. Randall, D. Malcolm, Evaluation of Wind
https://doi.org/10.2514/8.3713. Shear Patterns at Midwest Wind Energy Facilities. Technical Report, NREL/
[105] E.R. Van Driest, On turbulent flow near a wall, J. Aeronaut. Sci. 23 (11) (1956) CP-500-32492, National Renewable Energy Laboratory, USA, 2002.
1007e1011, https://doi.org/10.2514/8.3713. [119] J.S. Irwin, A theoretical variation of the wind profile power-law exponent as
[106] E. Balaras, C. Benocci, U. Piomelli, Two-layer approximate boundary condi- a function of surface roughness and stability, Atmos. Environ. 13 (1) (1967)
tions for large-eddy simulations, AIAA J. 34 (6) (1996) 1111e1119, https:// 191e194, 1979.
doi.org/10.2514/3.13200. [120] M. Kelly, G. Larsen, N.K. Dimitrov, A. Natarajan, Probabilistic meteorological
[107] N. Jarrin, S. Benhamadouche, D. Laurence, R. Prosser, A synthetic-eddy- characterization for turbine loads, J. Phys. Conf. 524 (2014), 012076.
method for generating inflow conditions for large-eddy simulations, Int. J. [121] M. Ray, A. Rogers, J. McGowan, Analysis of Wind Shear Models and Trends in
Heat Fluid Flow 27 (4) (2006) 585e593. Different Terrains, University of Massachusetts, Department of Mechanical
[108] J. O'Sullivan, R. Archer, R. Flay, Consistent boundary conditions for flows and Industrial Engineering, Renewable Energy Research Laboratory, 2006.
within the atmospheric boundary layer, J. Wind Eng. Ind. Aerod. 99 (1) [122] G. Geernaert, Surface Layer, encyclopedia of atmospheric sciences, 2003.
(2011) 65e77. [123] S. Asadi, H. Johansson, Multibody dynamic modelling of a direct wind tur-
[109] Vestas Wind Systems A/S, General specification V90e1.8/2.0 MW 50 Hz VCS. bine drive train, Wind Eng. 44 (5) (2020) 519e547.
http://ventderaison.eu/gembloux/eie_ABO-WIND/Annexes/Annexe_N_1_ [124] B. Yang, A. Morse, R. Shaw, K. Paw U, Large-eddy simulation of turbulent flow
Courbe_acoustique_V90.pdf, 2010. (Accessed 23 April 2021). across a forest edge. part ii: momentum and turbulent kinetic energy bud-
[110] J. Jonkman, S. Butterfield, W. Musial, G. Scott, Definition of a 5-mw Reference gets, Boundary-Layer Meteorol. 121 (3) (2006) 433e457.
Wind Turbine for Offshore System Development, Tech. Rep., National [125] F. Kanani, K. Tr€aumner, B. Ruck, S. Raasch, What determines the differences
Renewable Energy Lab.(NREL), Golden, CO (United States), 2009. found in forest edge flow between physical models and atmospheric mea-
[111] K. Thomsen, The Statistical Variation of Wind Turbine Fatigue Loads. No. surements? an LES study, Meteorol. Z. 23 (1) (2014) 33e49.

828