J. Wind Eng. Ind. Aerodyn.

125 (2014) 22–29

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

Wind resource assessment on complex terrain based on observations

of a single anemometer
M.X. Song, K. Chen, Z.Y. He, X. Zhang n
Key Laboratory for Thermal Science and Power Engineering of Ministry of Education, Department of Engineering Mechanics, Tsinghua University, Beijing, PR China

art ic l e i nf o a b s t r a c t

Article history: A method for numerical wind energy assessment of wind farm based on observations of a single
Received 7 April 2013 anemometer is presented. Utilizing computational fluid dynamics, the present method establishes a
Received in revised form rough relation between the boundary wind velocity and the wind velocity at the anemometer, guided by
27 November 2013
which a feedback process is conducted to search for the boundary velocity matching the measurement of
Accepted 28 November 2013
Available online 27 December 2013
the anemometer. The present method is able to provide reliable wind resources distribution for wind
farm on complex terrain without applying any mesoscale models. The present method is validated
Keywords: through measurements of anemometers installed within a certain wind farm in China.
Wind power & 2013 Elsevier Ltd. All rights reserved.
Anemometer
Wind resource assessment
Computational fluid dynamics

1. Introduction nonlinear models are becoming more feasibly benefited to the fast
development of computer hardware and technologies. The non-
Wind power generation is one of the most mature forms of linear model uses computational fluid dynamics (CFD) to solve the
renewable energy utilization. But great difficulties still exist due to Navier–Stokes (NS) equations numerically (Palma et al., 2008).
the spatial nonuniformity and temporal instability of wind energy Both the Reynolds Averaged Numerical Simulation (RANS) model
caused by terrain and climate. In order to reduce wake flow (Maurizi et al., 1998; Castro et al., 2003) and the Large Eddy
interferences of wind turbines and to increase the efficiency of Simulation (LES) model (Uchida and Ohya, 2008) have been
power generation of the wind farm, it is essential to accurately studied as a turbulence model coupled with NS equations in the
evaluate the wind resource distribution. Moreover, the highly effi- wind farm simulation scenarios.
cient wind-lens wind turbine is developed in recent years (Ohya and Boundary conditions are essential for either model of airflow
Karasudani, 2010), which is a great advantage for the distributed roof simulation. Especially, the distribution of wind speeds and direc-
energy system, but requires accurate wind resource evaluation for tions within the border surfaces must be known for the closure of
urban area where more complicated airflow motions exist. the governing equations. The WAsP model can be coupled with the
Existing methods for local wind resource assessment include Karlsruhe Atmospheric Mesoscale Model (KAMM) (Frank et al.,
linear and non-linear models (Palma et al., 2008). The linear model 2001), which provides the calculated wind distribution of a larger
used in the WAsP software is one of the most commonly used area with lower resolution as the boundary conditions for small
ways to evaluate the wind resources of wind farms (Frank et al., scale assessment. Other improved mesoscale models have con-
2001). The basic idea of the WAsP model is to linearize the tinuously been developed in recent decades, such as the KAMM2
momentum equations that govern the airflow motion (Jackson model with improvements on the applied physical and numerical
and Hunt, 1975), which significantly reduces the difficulty in approximations (Adrian, 1999) and the WRF model with moisture,
solving the nonlinear partial differential equations and the time to map projection, nesting and other features (Skamarock et al.,
cost by calculation. However, the experimental observations 2008; Carvalho et al., 2012). However, the mesoscale simulation
demonstrate that the linear model overestimates the local accel- runs based on the result of global numerical weather prediction
eration of airflow around the top of the hills, and is unable to (NWP) which has much lower resolution comparing to a local
resolve flow separation at the lee of the hills (Ayotte and Hughes, wind farm. The errors during the NWP and mesoscale simulation
2004). Although improved linear models have been developed, the may be accumulated, causing systematic error in small-scale wind
resource assessment of wind farm. Observations of local anem-
ometers are necessary for correction.
n
Corresponding author. Tel.: þ 86 10 62772668. The Measure-Correlation-Prediction (MCP) method is a widely
E-mail address: x-zhang@tsinghua.edu.cn (X. Zhang). used approach to assess wind resources. For example, the WindPRO

0167-6105/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jweia.2013.11.011
 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29 23



software includes a regression MCP feature that runs with NCEP/ ∂k ∂ μt ∂k μ ∂ui ∂uj 2
NCAR data (Thøgersen et al., 2007). The NCEP/NCAR data is a ρuj ¼ þ t þ  ρɛ ð3Þ
∂xj ∂xj Prk ∂xj 2 ∂xj ∂xi
continually updating data set representing the global atmosphere
state. The MCP method uses linear or nonlinear regression to


∂ɛ ∂ μt ∂ɛ μ ∂ui ∂uj 2 ɛ ɛ2
determine the correlation of global status and local observations. ρuj ¼ þC 1 t þ  C2ρ ð4Þ
∂xj ∂xj Prɛ ∂xj 2 ∂xj ∂xi k k
It then predicts the future of local status according to the global
forecast and the obtain correlation. However, the MCP method has where ui and uj are the components of Reynolds-averaged velocity,
two prerequisites: (1) It works only when the global data and the p is the Reynolds-averaged pressure, k is the turbulent energy, ɛ is
local data have good correlation. (2) The history data with a certain the dissipation rate of turbulent energy, μt is turbulent viscosity
time span of both the global and local observations is necessary. calculated by
Above complex terrain, due to the nonlinearity of fluid mechanics, it 2
could be difficult to find a fixed form of formula that expresses the k
μt ¼ ρC μ ð5Þ
correlation. ɛ
In this paper, the wind resource assessment method based only The constants in the equations are: C 1 ¼ 1:44, C 2 ¼ 1:92,
on the observations of a single anemometer is developed. Accord- Prk ¼ 1:0, Prɛ ¼ 1:85, C μ ¼ 0:033 (Palma et al., 2008).
ing to the measured velocities of an anemometer install inside the
wind farm area, the present method searches for boundary wind
velocities that would produce flow fields matching the observa- 2.2. Boundary conditions
tions of the anemometer. For multiple observed records of the
anemometer, during the searching process, the present method is In a typical wind farm or urban scenario, the dimensions of
constructing a numerical data set which stores all the calculated obstacles (terrain topography or buildings) are the order of 10 m,
flow fields. The data set is called Anemometer Phase Graph (APG), wind speed is the order of 10 m/s, air viscosity is 1:51  10  5 m2 =s,
which is explained in detail in Section 3. The APG provides which leads to the Reynolds number to be the order of 107.
information to accelerate the searching process. The present Therefore, approximate boundary conditions for high Reynolds
method has the following advantages comparing to the traditional number are specified in the present study (Patankar et al., 1978).
MCP method: (1) The present method is not based on statistical Boundary condition for turbulent energy near the terrain surface is
regression, therefore there is no requirement on correlation given by


between observations. The present method calculates the distri- ∂k
bution of wind resources throughout the domain based on the ! ¼0 ð6Þ
∂ n surface
digital elevation data and the observations of a single anem- !
ometer. It needs no history data of the target positions. The wind where n is the normal vector of the surface. In the grids closest to
energy distribution throughout the domain can be obtained just the terrain surface, dissipation rate of turbulent energy is calculated
from the data of a single anemometer. (2) Instead of using fixed by
formula for correlation, the present method employs the APG 3=2
C 3=4
μ k
technique, which contains a large data set for searching matching ɛ¼ ð7Þ
boundary velocity. The numerical data set of APG provides a more
κx
accurate description of relations between the anemometer and all where x is the distance between the grid node and the nearest
the other points throughout the domain, while the MCP method surface, κ ¼ 0:41 is the von Karman constant.
only obtains correlation between two points where history data is The surrounding surfaces are the most difficult part when
available. The remainder of this paper is organized as follows: specifying the boundary conditions for numerical calculation.
Section 2 describes the numerical process of flow field simulation. In real scenarios, it is not likely to have sufficient measured
Section 3 presents an Anemometer Phase Graph technique for velocities within all surrounding surfaces. In the present study,
searching boundary velocities and flow fields that match the the calculated terrain is smoothly expanded to flat horizons, and
observations of the anemometer. Section 4 validates the present the logarithmic velocity profile for the atmosphere boundary layer
method by actual measurements of a certain wind farm in China. is specified at the surrounding surfaces with flat horizons (Burton
Section 5 presents the conclusions. et al., 2005):
lnðz=z0 Þ lnðz=z0 Þ
uðzÞ ¼ uref cos θ; vðzÞ ¼ uref sin θ; wðzÞ ¼ 0
2. Flow field simulation lnðzref =z0 Þ lnðzref =z0 Þ
ð8Þ
2.1. Governing equations
where θ is the direction of incoming wind.
Since the terrain topography or buildings inside the domain
The motion of airflow over complex terrain or in the urban area
produce strong turbulence, the values of k and ɛ at the surround-
is governed by the incompressible Navier–Stokes equations. Since
ing surfaces do not have a significant effect on the inner flow field.
the intention of the present study is to develop a method to assess
At the surrounding surfaces, turbulent energy is set to be
the average wind energy resources, the steady governing equa- 2
k ¼ 0:01u2 , dissipation rate is calculated by ɛ ¼ 500C μ k =uL, where
tions are used, and the Reynolds Averaged Numerical Simulation
L is the characteristic length (Patankar et al., 1978).
(RANS) method is employed for turbulence simulation. The gov-
erning equations of the present problem are the continuity
equation, momentum equations and k  ɛ equations (Launder 2.3. Non-dimensionalization
and Spalding, 1974):
∂ui Choosing the characteristic variables as in Table 1, the govern-
¼0 ð1Þ ing equations are non-dimensionalized into:
∂xi
!


∂ k~ ∂u~ i
2
∂ui ∂p ∂ ∂u ∂u~ ∂p~
ρuj ¼ þ μt i ð2Þ u~ j i ¼  þ ð9Þ
∂xj ∂xi ∂xj ∂xj ∂x~ j ∂x~ i ∂x~ j ɛ~ ∂x~ j
24 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29

Table 1 triangle as the optional position for anemometer installation. The

Characteristic variables. 3D model of the area is shown in Fig. 2.
The observed velocity at a single point inside the domain does
Characteristic variable Value
not provide sufficient boundary conditions that close the govern-
Characteristic length L zref in Eq. (8) ing equations. But with the logarithmic profile, the flow field is
Characteristic velocity U uref in Eq. (8) determined only by boundary wind speed and boundary wind
Characteristic turbulent energy U2 direction. Therefore, to deduce the boundary velocity backward
K
Characteristic dissipation rate E
from inside point is possible. This section presents the algorithm
Cμ K 2 Cμ U3
UL
¼
L
to search for boundary velocity that produces flow field matching
Characteristic pressure P ρU 2 the measured velocity of the anemometer.
The first stage is to establish a rough relation between the
anemometer velocity, the boundary velocity and the flow field,
which is named as Anemometer Phase Graph.

3.1. Anemometer phase graph

100
Define the Anemometer Phase Space (APS) to be the phase space
representing the horizontal velocity measured by the anem-
ometer. The axes of APS are velocity components u and v
0 measured by the anemometer. Every point in the APS stands for
y (m)

a possible velocity measured by the anemometer. The Anemometer

Phase Graph (APG) is a set of points in APS, each of which has the
anemometer measured velocity components (u,v) as its coordi-
nate, and a boundary velocity and a flow field matching the
-100
anemometer's measurement as its properties.
To generate an APG, first generate 16 boundary velocity direc-
tions that uniformly cover 0–3601, e.g., 01; 22:51; 451; …; 337:51.
For each angle denoted as θ, process as follows:
-200

-100 0 100 200 1. Calculate the dimensionless boundary velocity components

x (m) at the height zref :

Fig. 1. Plan view of the urban area. (For interpretation of the references to color in
u~ B ¼ cos θ; v~ B ¼ sin θ ð15Þ
this figure caption, the reader is referred to the web version of this paper.) where the subscript B stands for boundary.
! 2. Numerically calculate the dimensionless flow field, denote

2 2
∂k~ 1 ∂ k~ ∂k~ 1 ∂u~ i ∂u~ j k~
2
the dimensionless velocity vector at each grid node as
u~ j ¼ þ þ  C μ ɛ~ ð10Þ ~ j; kÞ, where i; j; k are the grid indices.
∂x~ j Prk ∂x~ j ɛ~ ∂x~ j 2 ∂x~ j ∂x~ i ɛ~ uði;
3. Interpolate the dimensionless flow field for dimensionless
!
2 velocity components (u~ A ; v~ A ) at the position of the anem-
1 ∂ k~ ∂ɛ~
2
∂ɛ~ 1 ∂u~ i ∂u~ j ~ ɛ~ 2
u~ j ¼ þ C1 þ k C 2 C μ ð11Þ ometer, where the subscript A stands for anemometer.
∂x~ j Prɛ ∂x~ j ɛ~ ∂x~ j 2 ∂x~ j ∂x~ i k~ 4. Calculate the norm of (u~ A ; v~ A ), denoted as J u~ A J . Due to the
The boundary conditions at the grids closest to the terrain surface similarity of flow field as discussed in Section 2.3, the whole
are non-dimensionalized into: velocity field can be linearly scaled to normalize (u~ A ; v~ A Þ:
! U A ¼ u~ A = J u~ A J ; V A ¼ v~ A = J u~ A J
∂k~ k~
3=2

¼ 0; ~
ɛ ¼ ð12Þ U B ¼ u~ B =J u~ A J ; V B ¼ v~ B =J u~ A J ; ~ j; kÞ= J u~ A J
Uði; j; kÞ ¼ uði;
!
∂n μ κ x~
C 1=4
surface ð16Þ
The boundary conditions of the surrounding surfaces are non- These matching velocities represent a point in the APG,
dimensionalized into: where ðU A ; V A Þ is the coordinate in APS, corresponding to
lnðz~ zref =z0 Þ lnðz~ zref =z0 Þ ðU B ; V B Þ and Uði; j; kÞ.
~ z~ Þ ¼
uð cos θ; ~ z~ Þ ¼
vð sin θ; ~ z~ Þ ¼ 0
wð
lnðzref =z0 Þ lnðzref =z0 Þ
ð13Þ Take the position marked by red triangle in Fig. 1 as the example,
the height of the anemometer is set to be 5 m. The initial APG is
calculated containing 16 points. The points are connected in the
500k~
2

k~ ¼ 0:01u~ 2 ; ɛ~ ¼ ð14Þ
u~
It can be noticed that the dimensionless velocity components u~ i
do not depend on uref . Flow fields with the same θ but different
uref are linearly similar to each other.

3. Matching boundary velocity

A typical urban area of Tsinghua University in Beijing is used as

the example during the elaboration in this section. The plan view
of the urban area is as Fig. 1, where a point is marked by red Fig. 2. 3D model of the urban area.
 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29 25

order of boundary velocity direction, as plotted in Fig. 3. Although the The iteration above finishes with an APG point T found to be
APG curve shows only the normalized anemometer velocity ðU A ; V A Þ, sufficiently close to the actual measurement of the anemometer.
note that each of the points has its corresponding matching Its corresponding boundary velocity ðU B;T ; V B;T Þ and corresponding
boundary velocity and matching flow field calculated. velocity field UT ði; j; kÞ approximately match the normalized actual
measurement ðU n ; V n Þ. Then the dimensional boundary velocity
and dimensional velocity field that match ðun ; vn Þ are calculated
3.2. Search for matching boundary velocity
through the similarity of flow field:
After the initial APG is generated, the searching process can be uB ¼ U B;T J un J ; vB ¼ V B;T J un J ; uði; j; kÞ ¼ UT ði; j; kÞ J un J ð20Þ
conducted. Assume the measured velocity at the anemometer is
When processing a time series with multiple records of velocity
ðun ; vn Þ, the searching process is performed following the steps:
measurements of the anemometer, the iteration steps described in
this section are applied successively for each record. Since a
1. Normalize ðun ; vn Þ by its norm J un J to be ðU n ; V n Þ. Denote
certain number of APG points are inserted into the APG during
the point ðU n ; V n Þ in APS as Pn .
each iteration cycle, the APG becomes finer as the process goes.
2. If Pn is sufficiently close to a certain point T on the APG
The larger amount of nodes is included in the APG, the more
curve, use point T as the approximation of Pn . Iteration
information it provides to describe the correlation between the
finishes.
anemometer and the boundary velocity. The iterations for the
3. Project Pn to the nearest segment of the APG curve.
subsequent records are converged faster than the beginning ones.
The projection point is denoted as P. Denote the two endpoints
of the segment as T1 and T2 . Calculate:
ðT1  Pn Þ  ðT2  T1 Þ 4. Results and discussion
r¼ ð17Þ
J T2  T 1 J
r is in the range of ½0; 1, representing the relative position of P The present method is applied to real wind farm on complex
on the segment. terrain. The scale of the wind farm area is approximately 9200 m
4. Calculate the estimation of dimensionless boundary velocity from west to east and 5600 m from south to north. The terrain
components: elevation with resolution of 1 arc second, as plotted in Fig. 4, is
provided by ASTER GDEM (Global Digital Elevation Model), a
u~ ′B ¼ ð1  rÞU B1 þ rU B2 ; v~ ′B ¼ ð1  rÞV B1 þ rV B2 ð18Þ product of The Ministry of Economy, Trade and Industry (METI)
where ðU B1 ; V B1 Þ is the corresponding boundary velocity of of Japan and The National Aeronautics and Space Administration
APG point T1 , ðU B2 ; V B2 Þ is the corresponding boundary (NASA) of The U.S. There are two anemometers installed at the
velocity of APG point T2 . positions marked by circles and numbers. Anemometer 1 is
5. Calculate the dimensionless flow field u′ði; ~ j; kÞ using the located on the east side, with height at 40 m, while Anemometer
boundary velocity ðu~ ′B ; v~ ′B Þ. 2 is located on the west side, with height at 70 m.
6. Interpolate the calculated dimensionless flow field for the The terrain is smoothly expanded to flat horizons as discussed in
dimensionless velocity at the anemometer ðu~ ′A ; v~ ′A Þ. Section 2.2, based on which terrain following grids are generated
′
7. Denote the norm of ðu~ ′A ; v~ ′A Þ as J u~ A J , scale ðu~ ′A ; v~ ′A Þ, ðu~ ′B ; v~ ′B Þ, for flow field calculation, plotted in Fig. 5. The x-length is extended
and the dimensionless flow fields to normalize ðu~ ′A ; v~ ′A Þ:
Z
′ ′
U ′A ¼ u~ ′A = J u~ A J ; V ′A ¼ v~ ′A = J u~ A J 2000 1660
′ ′ ′ 1640
′ ′ ′ ′
U B ¼ u~ B = J u~ A J ; V B ¼ v~ B = J u~ A J ; ~
U′ði; j; kÞ ¼ u′ði; j; kÞ= J u~ A J 1620
1000 1600
ð19Þ 1580
1560
y (m)

0 2 1 1540
1520
8. Insert a point into APG, using ðU ′A ; V ′A Þ as its coordinate, 1500
-1000
corresponding to ðU ′B ; V ′B Þ and U′ði; j; kÞ. Go to Step 2. 1480
1460
-2000 1440
1420
1400
1 -4000 -2000 0 2000 4000 1380

x (m)

Fig. 4. Terrain elevation of the wind farm, x-axis points east, y-axis points North.
0.5

0
V

-0.5
5000
4000
-1 2000 0
Z
0 X
-1 -0.5 0 0.5 1 Y Y X
-2000 -5000
U -4000

Fig. 3. The initial Anemometer Phase Graph for the marked point. Fig. 5. Expanded terrain and grids for calculation.
26 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29

to 12,880 m, y-length is extended to 7840 m. The velocity on the 1. Acquire the data of wind speed and direction recorded by
surrounding borders is set according to the logarithmic profile, with Anemometer 2 from March 1st to March 31th of a certain year,
roughness length at 0.03 m, the typical value for broad farms with with time interval at 1 h. Totally 744 records.
few trees and buildings (Burton et al., 2005). 2. Apply the present method to deduce the boundary velocity and
Two grid resolutions are respectively used to test the grid the flow field that match each of the 744 observations of
dependency of CFD method: 111  69  20 and 140  98  25. Anemometer 2 successively.
In the test calculations, the boundary reference wind speed is set 3. Interpolate each of the matching flow fields, calculating the
to be 6 m/s at 40 m height from the ground. The solved velocity velocity at the position of Anemometer 1.
fields with the two grid systems are interpolated respectively, 4. Compare the calculated result with actual measurements of
obtaining the profiles of velocity magnitude at 40 m height from Anemometer 1.
the ground within the vertical cross section at X ¼0, plotted in
Fig. 6. It is noticed from the figure that the velocity profile does not The time series of actually measured velocity components u
change much when increasing grid resolution. The average error of and v of Anemometer 2 is plotted in Fig. 7. The calculation totally
the two velocity profiles is 0.27 m/s. As the terrain contains took 86,694 s on an ordinary 8-core PC. Fig. 8 is the comparison of
fluctuations of elevation in all spatial scales, boundary shape velocity component u of Anemometer 1, while Fig. 9 is the
changes when the grid number increases. According to the tested comparison of velocity component v of Anemometer 1, where
results, the solutions can be approximately accepted as grid purple solid curves are actually measured data, and blue dashed
independent. Hence, grid resolution at 111  69  20 is used in curves are the calculated results from the present method.
the following computations in order to save time. It is noticed from the figures that most of the calculated results
and the measured data are in good agreement. The average error
4.1. Hourly calculation for velocity component u of Anemometer 1 is 1.22 m/s, while the
average error for velocity component v is 1.11 m/s. The correlation
Since the measured point of Anemometer 2 is further away from of velocity component v between the two anemometers is better
the ground than that of Anemometer 1, airflow around Anem- than that of u. The possible reasons for the error are:
ometer 2 will be less influenced by the terrain topography and more
similar to the boundary status. Therefore, observed data recorded 1. The inaccuracy of boundary conditions during the flow field
by Anemometer 2 is used to search for matching boundary velocity calculation. Due to the lack of boundary information, the
and matching flow field, while data of Anemometer 1 is used for terrain was smoothly expanded to flat horizons, and the
validation. The time interval of the captured raw data is 10 min. velocity profiles on the boundaries are identical. The studied
Since the employed CFD method solves the steady equations, the wind farm is of the length scale of 10 km, the velocity profiles
time span of each simulation must be sufficiently long to approx- along the borders are more likely to be nonuniform. To have
imate a steady averaged velocity field. The length of the domain is assumed that the boundaries have identical velocity profiles
in the order of 104 m, the typical wind speed is in the order of leads to errors in some time periods.
10 m/s, so the characteristic time for the air to blow across the 2. The assumption of steady flow. Since the incoming wind speed
domain is in the order of 1000 s. Therefore, the records are averaged and direction are always varying in reality, the steady state of
into time series with interval at 1 h, which is also in the order of airflow motions required by the governing equations is never
1000 s. The validation process of the present method is performed going to be exactly met.
as follows: 3. The obstacle information of the calculated area is lacking.
Correlation of the two anemometers may be influenced by
8 buildings nearby, which is unable to be considered in the
7.5 present study,
7
Wind speed (m/s)

6.5
4.2. Annual assessment
6

5.5 To reduce the error caused by the instability of boundary velocity,

5 the same calculation is performed for the daily averaged data. In this
4.5
section, the velocity components measured by Anemometer 2 of the
111x69x20
140x98x25
entire year are averaged to daily records, based on which the daily
4
average velocities at Anemometer 1 are calculated and compared to
3.5 actually measure data of Anemometer 1, as plotted in Figs. 10 and 11.
-4000 -2000 0 2000 4000
The calculation of 366 records took 43,257 s on the same 8-core PC as
X (m)
for the hourly records. The errors of the velocity components for each
Fig. 6. Wind speed profiles for grid test. month are listed in Table 2.

30
Velocity component (m/s)

20
10
0
-10
-20 u
v
-30
0 100 200 300 400 500 600 700
Time (hour)

Fig. 7. Measured hourly velocity components of Anemometer 2.

 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29 27

20

Velocity component (m/s)

Measured u
15 Calculated u

10
5
0
-5
-10
0 100 200 300 400 500 600 700
Time (hour)

Fig. 8. Measured and calculated hourly velocity components u of Anemometer 1.

20
Velocity component (m/s)

10

-10
Measured v
-20 Calculated v

0 100 200 300 400 500 600 700

Time (hour)

Fig. 9. Measured and calculated hourly velocity components v of Anemometer 1. (For interpretation of the references to color in this figure caption, the reader is referred to
the web version of this paper.)

15
Velocity component (m/s)

10

-5 Measured u
Calculated u
-10
0 50 100 150 200 250 300 350
Time (day)

Fig. 10. Measured and calculated daily velocity components u of Anemometer 1.

15
Velocity component (m/s)

10
5
0
-5
-10 Measured v
Calculated v
-15
0 50 100 150 200 250 300 350
Time (day)

Fig. 11. Measured and calculated daily velocity components v of Anemometer 1.

It is noticed that by using daily averaged records, the errors of Table 2

velocity components u and v for March have been reduced to Monthly velocity error of the calculated year (m/s).
0.77 m/s and 0.70 m/s respectively, comparing with the ones in
Month January February March April May June
the previous section (1.22 m/s and 1.11 m/s respectively) when u-error 0.75 0.67 0.77 0.87 1.24 1.13
using hourly records. The annual average errors of velocity v-error 0.94 0.83 0.70 1.02 0.76 0.65
components u and v are 1.02 m/s and 0.83 m/s respectively. Month July August September October November December
The calculated annually averaged wind speed is 5.57 m/s, u-error 1.14 1.04 0.82 0.94 1.75 1.13
which is 2.5% smaller than the measured one (5.72 m/s). The error v-error 0.54 0.54 0.56 0.81 1.31 1.31
of velocity magnitude is evaluated through fitting the data into
Weibull distributions. The Weibull distribution is used to model
the probability distribution of wind speed in most of the wind The observed range of wind speed is divided into subranges
energy scenarios (Burton et al., 2005). The probability density with interval at 1 m/s. Measured and calculated records within
function of a Weibull distributed wind speed U (U Z 0) is: each subrange are counted respectively. The normalized probabil-

 "
k # ities are plotted as bars in Fig. 12. The Weibull parameters of
k U k1 U both the measured and the calculated records are fitted through
f ðUÞ ¼ exp  ð21Þ
λ λ λ the least squares method respectively, as listed in Table 3.
The probability density functions are plotted as solid curves in
where λ is the scale parameter, and k is the shape parameter. Fig. 12, and their comparison is plotted in Fig. 13.
28 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29

0.15 0.15

Probability

Probability
0.1 0.1

0.05 0.05

0 0
0 5 10 15 0 5 10 15
Wind speed (m/s) Wind speed (m/s)

Fig. 12. Fitted Weibull distributions of the measured and calculated wind speeds.

Table 3
Fitted parameters of Weibull distributions.
0.15 Measured
Parameter λ k Calculated

Measured 6.374 2.183

Calculated 6.174 2.314

Probability 0.1
The scale parameter λ of the calculated data is 3.1% smaller
than the one of the measured data, which leads to the same
conclusion of underestimation of annually averaged wind speed.
The shape parameter k of the calculated data is 6.0% larger than 0.05
the one of the measured data, indicating that the calculated speed
variation is weaker than the measured one. The reason for
obtaining a larger k is mainly attributed to the tower heights of
the anemometers. The 70 m high Anemometer 2 should be less
interfered by ground obstacles than the 40 m high Anemometer 1. 0
0 5 10 15
Airflow at higher place is steadier than the lower place. Therefore,
the calculated results according to Anemometer 2 have lower Wind speed (m/s)
variation than the actual measurements of Anemometer 1.
Fig. 13. Comparison of the Weibull distributions of the measured and calculated
results.

5. Conclusion
validated through comparing with the actual measured data of
A method to assess wind resources of wind farm on complex another 40 m high anemometer. Through the data and discussion
terrain based on observations of a single anemometer is presented. in the previous sections, it is concluded that, based on the measured
For each observed record of wind velocity at the anemometer, time series of velocity at one point, the present method is able to
the method searches for a boundary velocity that produces a calculate an approximate time series of velocity of another point
flow field matching the measured velocity at the anemometer, within the area above complex terrain. The calculated wind resource
using the Anemometer Phase Graph technique developed in this characteristics have good agreement with the measured ones,
paper. As the flow fields matching all the observed records of indicating that the present method can be used for wind resource
the anemometer are numerically calculated respectively, wind assessment with the observations of only a single anemometer.
resources of the corresponding time period can be assessed. The
present method has the following advantages: (1) Instead of
statistical regression, the numerical calculation is based on physical Acknowledgments
models of fluid dynamics, which has a broader scope of application
on complex terrains comparing with the linear model of WAsP. The recorded measurements of the anemometers in the real
(2) The present method does not need history data of the target wind farm are provided by China HuaDian Corporation.
locations. Only with the elevation data and the observations of a This research is supported by the Program of International S&T
single anemometer, it calculates the distribution of wind resources Cooperation of China (No. 2011DFG13020), China Postdoctoral
throughout the domain. (3) As the assessment process goes in the Science Foundation (2013M530043), and the National High-Tech
present method, the APG keeps growing into a large data set of R&D Program (863 Program) of China (No. 2007AA05Z426).
calculated flow fields. It provides more accurate information for the
search of matching boundary velocity than a fixed form of correla-
References
tion formula as in the traditional MCP method.
The present method is applied to a real wind farm in China. The
Adrian, G., 1999. Parallel processing in regional climatology: the parallel version of
boundary velocities and flow fields are calculated to match the the “Karlsruhe Atmospheric Mesoscale Model” (KAMM). Parallel Comput. 25,
measured data of a 70 m high anemometer, the results of which are 777–787.
 M.X. Song et al. / J. Wind Eng. Ind. Aerodyn. 125 (2014) 22–29 29

Ayotte, K.W., Hughes, D.E., 2004. Observations of boundary-layer wind-tunnel flow Ohya, Y., Karasudani, T., 2010. A shrouded wind turbine generating high output
over isolated ridges of varying steepness and roughness. Bound. Layer power with wind-lens technology. Energies 3, 634–649.
Meteorol. 112, 525–556. Palma, J.M.L.M., Castro, F.A., Ribeiro, L.F., Rodrigues, A.H., Pinto, A.P., 2008. Linear
Burton, T., Sharpe, D., Jenkins, N., Bossanyi, E., 2005. Wind Energy Handbook. John and nonlinear models in wind resource assessment and wind turbine micro-
Wiley & Sons Ltd. siting in complex terrain. J. Wind Eng. Ind. Aerodyn. 96, 2308–2326.
Carvalho, D., Rocha, A., Gómez-Gesteira, M., Santos, C., 2012. A sensitivity study of Patankar, S.V., Sparrow, E.M., Ivanovic, M., 1978. Thermal interaction among the
the WRF model in wind simulation for an area of high wind energy. Environ. confining walls of a turbulent recirculating flow. Int. J. Heat Mass Transf. 21,
Modell. Softw. 33, 23–34. 269–274.
Castro, F.A., Palma, J.M.L.M., Lopes, A.S., 2003. Simulation of the Askervein flow. Part Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M., Duda, M.G., Huang,
1: Reynolds Averaged Navier–Stokes equations (k–ɛ turbulence model). Bound. X.Y., Wang, W., Powers, J.G., 2008. A description of the advanced research WRF
Layer Meteorol. 107, 501–530.
version 3. Technical Report, National Center for Atmospheric Research, Boulder,
Frank, H.P., Rathmann, O., Mortensen, N.G., Landberg, L., 2001. The numerical wind
Colorado, USA.
atlas—the KAMM/WAsP method. Technical Report. Risø National Laboratory,
Thøgersen, M.L., Motta, M., Sørensen, T., Nielsen, P., 2007. Measure-correlate-
Roskilde, Denmark.
predict methods: case studies and software implementation. In: EWEC 2007
Jackson, P.S., Hunt, J.C.R., 1975. Turbulent wind flow over a low hill. Q. J. R. Meteorol.
Soc. 101, 929–955. Conference Proceedings.
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent flows. Uchida, T., Ohya, Y., 2008. Micro-siting technique for wind turbine generators by
Comput. Methods Appl. Mech. Eng. 3, 269–289. using large-eddy simulation. J. Wind Eng. Ind. Aerodyn. 96, 2121–2138.
Maurizi, A., Palma, J.M.L.M., Castro, F.A., 1998. Numerical simulation of the atmo-
spheric flow in a mountainous region of the North of Portugal. J. Wind Eng. Ind.
Aerodyn. 74–76, 219–228.