The Eighth Asia-Pacific Conference on

Wind Engineering, December 10-14, 2013,

Chennai, India
 

A MODIFIED VON KARMAN MODEL FOR OFFSHORE WIND FIELD

GENERATION
Shou OH1, Takeshi Ishihara2
1
Graduate Student, Department of Civil Engineering, The University of Tokyo, 7-3-1 Hongo,
Bunkyo, Tokyo, 113-8656, Japan, s_oh@bridge.t.u-tokyo.ac.jp
2
Professor, Department of Civil Engineering, The University of Tokyo, 7-3-1 Hongo,
Bunkyo, Tokyo, 113-8656, Japan, ishihara@bridge.t.u-tokyo.ac.jp

ABSTRACT

Accurate modeling of three-dimensional turbulent flow field is necessary for dynamic analysis of wind
turbines. In this study, a modified von Karman model is proposed for auto-correlations and auto-spectra as well
as cross-correlations and cross-spectra for offshore wind generation. Measurement is carried out on an offshore
meteorological mast for validation. Proposed model for auto-correlation and auto-spectra agree well with
measurement data, while the original von Karman model underestimates auto-correlation and peak frequency in
auto-spectra for lateral and vertical components. Proposed cross-correlation and cross-spectra also agree with
measurements. The values of model parameters are proposed for ratios of standard deviation and normalized
Reynolds stress based on measurement. Proposed values showed better agreement with measurement compared
to the recommended values in IEC61400-1.

Keywords: spectrum, correlation, von Karman model, the IEC requirement

Introduction
The characteristics of input wind field has a considerable effect in dynamic response
analysis of structures, and accurate modeling of three-dimensional turbulent flow field is
necessary. A turbulence model can be written in both spectral and correlation form. Since the
former is commonly used to describe turbulent characteristics and the latter can be directly
used in generation of time series of wind field using AR model, it is meaningful for a
turbulence model to be described in theoretically corresponding correlation and spectral form.
Also the model parameters such as standard deviations and length scales are important factors
of accuracy of the modeling. Therefore, for those structures like wind turbines and bridges
that is built both onshore and offshore, it is necessary for the parameters to be newly proposed
for offshore condition since the values provided in the design codes are usually based on
onshore measurement.
Numerous turbulence models are proposed based on both theory and experiment in
previous studies. Two representative models, Mann model (1988) and Kaimal model (1972),
are commonly used for structure designing against wind. Mann model is a theoretical model
based on von Karman model which is discussed below. However, due to its complex form of
correlation, it is difficult to apply to generation of wind field. Kaimal model provides auto-
spectrums for longitudinal, lateral, and vertical components based on measurements. However,
since the model is empirical, no theoretical equations for auto-correlations are available.
Similarly, one empirical model for uw component is provided by Kaimal for cross-correlation

1/8 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

between different components at one point, although equations for cross-correlation are not
available due to the same reason.
A well-known theoretical turbulence model is provided by von Karman (1948) where
spectrums and corresponding correlation functions for each component are obtained
theoretically assuming homogeneous and isotropic turbulence. It is commonly agreed that this
model is compatible with measurement for longitudinal component, however Maeda and
Makino (1992) pointed out that the spectrum of lateral and vertical components do not agree
well with measurements of wind field which is normally non-isotropic.
In this study, a modified von Karman model with theoretically consistent correlations
and spectra is proposed and the accuracy is validated by measurement from an offshore
meteorological mast. The model parameters such as standard deviation ratio and length scale
ratio for are proposed based on the measurement.
Proposed Turbulence Model
In the original von Karman model, auto-correlation of longitudinal component is
described with a correlation function f (r ) . Then the correlation function for lateral and
vertical component g (r ) is obtained from f (r ) function using equation of continuity under
isotropic turbulence field. However several measurements show that the shape of the auto-
spectrum is quite similar for all the three turbulence components. Therefore, in this study, it is
proposed to use the same shape for the auto-correlation and auto-spectrum for all the three
components based on von Karman’s longitudinal model. In other words, the use of same
f (r ) function for all the components is proposed. Based on this idea, auto-correlation Ri (U )
and auto-spectra Si ( f ) for i th component are written as Eq.(1) and Eq.(2).
Ri (U )  1 a1U
1/ 3
K1/ 3 ( a1U ) (1)
fSi ( f ) 4  fLi U 
 (2)
 i2
1  70.8( fL 
56
i U )2
where U is velocity of longitudinal component,  i is the standard deviation, Li is the length
scale,  is the time lag and K  is modified Bessel function of second kind of the order of  .
The coefficient 1 is 22 / 3 / (1/ 3)  0.593 and a1 is  (5/ 6) / (1/ 3) Li  0.7468Li respectively.
As auto-correlation is defined and described as R ij (U )  E[ui (t )ui (t  U )] /  i2  f (r ) ,
cross-correlation between different component defined as R ij (U )  E[ui (t )u j (t  U )] /  i j is
assumed to be able to be described with the same f (r ) function with a correcting factor  .
This  can be determined as   ui u j /  i j considering that R ij (0)  ui u j /  i j . Once the
Reynolds stress ui u j is obtained, the cross-correlation and cross-spectrum can be expressed
with Eq.(3) and Eq.(4).
ui u j
Rij (U ) 
1/ 3
1 a1U K1/ 3 ( a1U ) (3)
 i j
fCij ( f )

ui u j 
4 fLij U              (4)
 i j

 i j 1  70.8( fL U )2

56
ij

where Rij (U ) is the cross-correlation, Cij ( f ) is the cross-spectrum, a1  0.7468Lij and length
scale is assumed to be Lij  ( Li  L j ) / 2 . When i  1  and j  2 , the cross-correlation and the
cross-spectrum are modeled as 0 following the assumption in atmospheric boundary layer.
In the following chapters, first these proposed models are validated with measurement
data, and then the model parameters are proposed for offshore condition.

2/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

Outline of Measurement
Measurement data used for the validation of proposed model are from a
meteorological mast located 3.1km offshore Choshi, Japan. The location and the outline of the
met mast are shown in Figure 1. The mast is equipped with 3 sonic anemometers, 22 cup
anemometers, 23 wind vanes, and sensors for other meteorological data such as pressure and
temperature. In this study, data from the sonic anemometer located at 80m height is used,
since proposed turbulence model focuses on wind speeds at one point. Data from other
anemometers will be used in future discussion on spatial cross-correlation. Measurement has
started from November 2012, and data from 16:00 to 16:40 on March 15th 2013 is used for the
validation since high wind speed were observed for whole the day. Table 1 shows the mean
wind speed, the mean wind direction, the standard deviation, and Reynolds stresses for each
10 minutes and the averaged data. Among the averaged values, only the length scales are not
2
simple average of each 10 minutes data but are calculated with Li  S i (0) / 4 i where Li is
average length scale, S i (0) is average power spectrum at 0Hz,  i is averaged standard
deviation. All the calculation is performed after removing linear trend in original time series,
which will be discussed in the following section.
(a) (b)

Figure.1 (a) Location and (b) detail of observation tower

Table.1 Statistical characteristics of measurement data

No. Data U  u v w uv uw vw Lu Lv Lw
time (m/s) (deg) (m/s) (m/s) (m/s) 2
(m /s )
2 2
(m /s )
2
( m2 / s 2 ) (m ) (m ) (m )
1 16:00 17.7 206.5 1.15 0.89 0.73 -0.039 -0.16 -0.14 42.8 12.7 7.93
-16:10
2 16:10 19.2 206.3 1.09 0.94 0.75 0.043 -0.23 -0.18 51.7 12.3 8.19
-16:20
3 16:20 20.7 204.3 1.08 0.76 0.63 -0.042 -0.19 -0.16 56.4 20.4 12.4
-16:30
4 16:30 19.6 202.3 1.11 0.80 0.66 -0.11 -0.18 -0.24 51.1 26.3 16.8
-16:40
Mean - 19.3 204.9 1.11 0.84 0.69 -0.037 -0.20 -0.18 66.2 26.6 17.0

Data Analysis
As mentioned in the previous section, low frequency trend in 10 minutes time series
are removed prior to any evaluations. The trend is assumed to be linear and is calculated using
least-square method. Figure 2 (a) shows the length scales obtained from the original time
3/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

series and it is seen that the values of longitudinal component and lateral component are not
stable and some data are several times larger than other cases. Figure 2 (b) shows the length
scales calculated after removing linear trend. The effect of detrending can be seen in that
values for longitudinal and lateral components became almost same for every case while those
for vertical component did not change significantly. For calculation of power spectra, 10
minutes data are divided into blocks of 2048 samples and are adapted to discrete fast Fourier
transformation. The spectra are then applied to Hamming window before averaged. Finally
calculated correlations and spectra for data No.1 to No.4 are averaged and bin average is
adapted to the spectra at the size of 0.025Hz.
(a)  300
(b) 300

250
longitudinal 250 longitudinal
lateral

LengthScale (m)
LengthScale (m)

lateral
200 vertical 200
vertical
150 150

100 100

50 50

0 0
1 2 3 4 1 2 3 4
Data Number
Data Number  
Figure 2. Length scales for three components (a) before and (b) after removing trend
Validation of Proposed Model
Figure 3 shows the comparison between proposed and measured auto-correlation and
normalized spectrum for three components. For longitudinal component, where proposed
model is the same as von Karman model, the measurement shows good agreement with the
models. For the lateral and vertical components, where the proposed model differs from von
Karman’s original model, there is a small difference between proposed and original von
Karman model for the lateral component and a clear difference for the vertical component. In
both frequency and time domain, the proposed model gives satisfying results when compared
to the measurements.
Comparison of proposed and measured cross-correlation and cross-spectrum between
different components are shown in Figure 4. Figure 4 (a) shows that there is no clear
correlation between longitudinal and lateral components, which agrees with the assumption.
On the other hand, clear correlation exists between longitudinal and vertical (b) and lateral
and vertical (c) components. The proposed cross-correlation models agree well with the
measurement data.
Model Parameters of Proposed Model
In the proposed model, the mean wind speed U , the standard deviation  i , the length
scale Li and normalized Reynolds stress ui u j /  i j are the parameters that have to be
determined. Since U , 1 and L1 are usually determined by design requirement and the ratio
of length scales are calculated from the ratio of standard deviation, the ratios of standard
deviations and normalized Reynolds stresses are the parameters that have to be specified
when no measurement data is available. For calculation of the ratio of length scale, auto-
spectrum in asymptotic inertia subrange Si is derived from Eq.(2) as Eq.(5). Then, taking the
ratio of Eq.(5) between two components, Eq.(6) is obtained where the ratio of spectra is
S1 / S 2  S1 / S3  0.75 considering the theory of locally isotropic turbulence in inertia subrange,
In order to compare the values of parameters with the design code, the international
design code of wind turbines, IEC61400-1 is discussed below as an example. In IEC61400-1,
4/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

the ratios of standard deviations are determined as  2  0.81 and  3  0.51 for Kaimal model
whereas the measured result is and  2  0.761 and  3  0.621 which agree with the value
 2  0.82 1 and  3  0.631 proposed by Moraes (1988) based on the Kansas observation.
Therefore the parameters  2  0.81 and  3  0.61 are proposed in this study for offshore
condition. The predicted ratio of length scale for lateral component L2 / L1 is 0.33, which is
same as that in Kaimal model and close to the measurement of 0.40. The predicted ratio of
length scale for vertical component L3 / L1 is 0.081, which shows some difference compared
with measured ratio of length scale L3 / L1  0.14 and is better than Kaimal model with
L3 / L1  0.081 . Proposed values for Reynolds stress are determined from observation as
u1u2 /  1 2  0 , u1u3 / 1 3  0.13 , and u2u3 /  2 3  0.14 .  Comparison of model parameters of
observation, Kaimal model and proposed modified von Karman model are shown in Table 2.

(a) 1
(d) 1

Observation

normalized auto-spectrum
0.8
Proposed
auto-correlation

0.6 0.1
von Karman

0.4

0.01
0.2 observation
Proposed
0 Karman

0 10 20 30 40 50 0.001
0.001 0.01 0.1 1
time lag (s)
frequency (Hz)
(b) 1
(e) 1

Observation
normalized suto-spectrum

0.8
Proposed
auto-correlation

0.1
0.6 von Karman

0.4

0.01
Observation
0.2
Proposed
Karman
0
0.001
0.001 0.01 0.1 1
0 10 20 30 40 50
time lag (s) frequency (Hz)

(c) 1
(f) 1

Observation
normalized auto-spectrum

0.8
Proposed
von Karman
auto-correlation

0.1
0.6

0.4

0.01
Observation
0.2
Proposed
Karman
0
0.001
0 10 20 30 40 50 0.001 0.01 0.1 1

time lag (s) frequency (Hz)

Figure 3 Comparison between calculated and measured auto-correlation and normalized auto-
spectrum for (a) (d) longitudinal, (b) (e) lateral and (c) (f) vertical component

5/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

(a) 0.2
(d) 2.5

2
Observation
0.1
proposed

cross-spectrum
cross-correlation

1.5

0
1

0.5
-0.1

0
-0.2 observation
-0.5
proposed
-0.3 -1
0 10 20 30 40 50 0.001 0.01 0.1 1
time lag (s) frequency (Hz)
(b) 0.1
(e) 0.5

0 observation
0
proposed

cross-spectrum
cross-correlation

-0.5

-0.1
-1

-0.2 -1.5

observation -2
-0.3
proposed -2.5

-0.4 -3
0 10 20 30 40 50
0.001 0.01 0.1 1
time lag (s) frequency (Hz)
(c) 0.2
(f) 1

0.5
0.1
cross-correlation

0
cross-spectrum

0
-0.5

-0.1 -1

observation
-1.5
-0.2
proposed Observation
-2
proposed
-0.3
0 10 20 30 40 50 -2.5
0.001 0.01 0.1 1
time lag (s)
frequency (Hz)
Figure 4 Comparison between calculated and measured cross-correlation and cross-spectrum
for (a) (d) longitudinal and lateral, (b) (e) longitudinal and vertical and (c) (f) lateral and
vertical component
 
 
fSi  f 

 4 fLi
U  fL 
 0.115  i 
 23

(5)
 i2 70.8   U 
5
5 fLi 3
6
U

 U
fSi ( f ) 2
fLi 3
3
0.115  i 
3
i 2
Li  S j  2
      (6)
 j
fS j ( f ) 2
 fL  3 L j  Si  
0.115  j 
 j2  U

6/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

Table 2. Parameters in Kaimal and proposed model

Observation IEC(Kaimal) Proposed
 2 / 1 0.76 0.8 0.8
 3 / 1 0.62 0.5 0.6
L2 / L1 0.40 0.33 0.33
L3 / L1 0.26 0.081 0.14
u1u2 /  1 2 -0.028 - 0
u1u3 / 1 3 -0.124 - -0.13
u2u3 /  2 3 -0.148 - -0.15

Conclusions
In this study, a modified von Karman model is proposed to describe the non-isotropic
wind field in the forms of theoretically consistent correlation and spectrum. Following
conclusions are obtained:
1) Proposed auto-correlation and auto-spectrum model agree well with measurement data for
all components, while the original von Karman model under-estimates the value of auto-
correlation and the peak frequency in auto-spectrum for lateral and vertical component.
2) Proposed cross-correlation and cross-spectrum also agree well with measurements for
longitudinal and vertical component, and lateral and vertical component. The longitudinal
and lateral component can be assumed as 0 for both correlation and spectrum .
3) Ratios of standard deviation and normalized Reynolds stress in the modified von Karman
model are proposed for offshore condition. Proposed values show better agreement with
measurement than those in Kaimal model of IEC61400-1.
Appendix
The original von Karman model for auto-correlations and spectra shown in Figure 3
are calculated with Eq.(a1) and E.(a2) for longitudinal component, and Eq.(a3) and Eq.(a4)
for lateral and vertical component.
Ri (U )  1 a1U
1/ 3
K1/ 3 ( a1U ) (a1)
fSi ( f ) 4  fLi U 
 (a2)
 i2
 
56
1  70.8( fLi U ) 2
a1r
Ri (U )  1 a1U ( K1/ 3 ( a1U ) 
1/ 3
K 2 / 3 ( a1r )) (a3)
2
2( Li / U )(1  188.8  fLi U  )
2
fSi ( f )
 (a4)
 i2
1  70.8( fL 
11 6
i U )2
The auto-spectrum for Kaimal model is shown in Eq.(a5) for the three components, and the
empirical spectrum model for longitudinal and vertical components is shown in Eq.(a6).
fSi ( f ) 4  fLi U 
 (a5)
 i2 1  6( fLi U )
53

fCuw ( f ) 1.066  fLi U 

 (a6)
 u w 1  1.745( fLi U )
7/3

7/8 
 
 
The Eighth Asia-Pacific Conference on Wind Engineering, December 10-14, 2013, Chennai, India

Acknowledgment
This research is carried out as a part of a project funded by The New Energy and
Industrial Technology Development Organization (NEDO), Japan. The authors wish to
express their deepest gratitude to the concerned parties for their assistance during this study.
References
International Electrotechnical Commission. Wind turbines - Part 1:Design requirements. Number 918025243.
International Electrotechnical Commission, 3rd edition, 2005.
J. C. Kaimal, J. C. Wyngaard, Y. Izumi, and O. R. Coté. Spectral Characteristics of Surface-layer Turbulence.
Quarterly Journal of the Royal Meteorological Society, 98(417):563–589, 1972.
J. Maeda and M. Makino. Characteristics of Gusty Winds Simulated by an ARMA Model. Journal of Wind
Engineering and Industrial Aerodynamics, 44:427–436, 1992.
J. Mann, The spatial structure of neutral atmospheric surface-layer turbulence, Journal of Fluid Mechanism,
volume.273, pp.141-168, 1994
Moraes, O.L.L., The velocity spectra in the stalbe surface layer. Bound.-Layer Meteorol., 43: 591-600, 1988
T. von Karman. Progress in the Statistical Theory of Turbulence. Proceedings of the National Academy of
Sciences of the United States of America, 34(11):530–539, 1948

8/8