WAUDIT Book of Proceedings - Wind resource assessment, audit

Transcription

Book of Proceedings
WAUDIT Book of Proceedings
Edited by
Affiliation
Address
Telephone
E-mail
Document Type
Version
Date
Dissemination Level
Status
Abstract
WAUDIT
Contract#
238576
Javier Sanz Rodrigo
CENER
C/ Ciudad de la Innovación 7, 31621-Sarriguren, Spain
0034 948 25 28 00
[email protected]
Deliverable
v1
5-07-2013
Public
Draft
This book of proceedings compiles a selection of publications
from the WAUDIT network.
CENER National Renewable Energy Centre
C/Ciudad de la Innovación 7
Tel. (+34) 948 25 28 00
31621-Sarriguren - Spain
Fax. (+34) 948 27 07 74
Pag. 1/167
Book of Proceedings
Preface
The WAUDIT network (2009-2013) has gathered 23 Marie Curie fellows from 20 different nationalities
working in 13 European institutions to develop a scientific and training programme under the wind
resource assessment topic.
A training programme, under the umbrella of the European Wind Energy Academy and with the
contribution of 30 organizations, resulted in an excellent environment for early stage researchers to
initiate a career in wind energy research with a common European collaborative working philosophy.
Industry has also participated providing lectures at seminars and offering internship periods. On the
scientific and technical level the network has contributed to the development of 15 PhD thesis and has
published more than 100 contributions to journals, conferences and technical reports. A review of a wide
range of state-of-the-art techniques for wind resource assessment has been carried out, from remote
sensing with lidar systems to microscale modeling with CFD and wind tunnel and mesoscale modeling
using numerical weather prediction. Focus research areas include offshore, forests, atmospheric stability,
wakes, complex terrain, ensemble prediction and high resolution downscaling. Common to all model
developments, the network has discussed on the standardization of model evaluation procedures as an
effective way of quality assurance.
This book of proceedings presents a selection of some publications of the Marie Curie fellows of the
WAUDIT network.
Javier Sanz Rodrigo
Coordinator of the WAUDIT network
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 2/167
Book of Proceedings
Contents
[1] Barranger N (2010) Assessing wind energy potential using the high resolution meso-scale model
RAMS. 6th EAWE PhD Seminar, NTNU, Trondheim, Norway, September 2011.
[2] Borbón F, et al. (2011) Investigation of sources for lidar uncertainty in flat and complex terrain.
EWEA-2011, Brussels (Belgium), March 2011.
[3] Chávez R, Lozano S, Correia P, Sanz Rodrigo J, Probst O (2013) On the application of the Principal
Component Analysis for an efficient climate downscaling of surface wind fields. Energy Procedia 40:
67-76.
[4] Conan B, van Beeck J, Aubrun S (2012) Sand Erosion Technique Applied to Wind Resource
Assessment. Journal of Wind Engineering and Industrial Aerodynamics 104: 322-329.
[5] Cuzzola F, Aubrun S, Leitl B (2012) Characterization of a wind turbine model. EWEA Special Topic
Conference ‘The Science of Making Torque from Wind’. Oldenburg, Germany, October 2012.
[6] Desmond C, Watson S (2012) A study of stability effects in forested terrain. Journal of Physics:
Conference Series and Proceedings of The Science of Making Torque from Wind 2012, Oldenburg,
9-11 October 2012. In press for Journal of Physics
[7] Duraisamy JV, Dupont E, Carissimo B (2013) Downscaling wind energy resource in complex terrain
from mesoscale to microscale model and data assimilating field measurements at few locations.
European Wind Energy Association, Vienna, Austria, February 2013.
[8] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2011b) The Anisotropic Multifractal Model and
Wind Turbine Wakes. 7th PhD Seminar on Wind Energy in Europe, 115–118.
[9] Holmes H, Sanz Rodrigo J, Cabezón D, Schatzmann M (2013) Model evaluation methodology for
wind resource assessment
[10] Koblitz T, Bechmann A, Sogachev A, Sørensen N, Réthoré P-E (2013) CFD model of stratified
atmospheric boundary-layer flow. Wind Energy, accepted for publication.
[11] Muñoz-Esparza D, Canadillas B, Neumann T, van Beeck J (2012) Turbulent fluxes, stability and
shear in the offshore environment: Mesoscale modelling and field observations at FINO1. J.
Renewable Sustainable Energy 4, 063136, http://dx.doi.org/10.1063/1.4769201
[12] Stathopoulos C, Sanz J (2013) Evaluation of a meso-micro scale application over complex terrain for
wind farm sitting. 9th Phd seminar of European Wind Energy Association, Conference proceedings,
Gotland, Sweden, September 2013.
[13] Volker PJH, Badger J, Hahmann AH, Hansen KS (2013) Wind-Farm Parametrisations in Mesoscale
Models. ICOWES, 2013.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 3/167
Book of Proceedings
Assessing wind energy potential using the high resolution meso-scale model
RAMS
Nicolas Barrangera,1
a
University of Athens, Athens, Greece
Presented at the 6th EAWE PhD Seminar on Wind Energy in Europe, NTNU, Trondheim, Norway, September 2011.
Abstract
The regional atmospheric model RAMS has been developed for meteorological purpose. It solves the
unsteady dynamical equations governing the atmospherics physics from the synoptic scale down to
mesoscale and microscales. Its last evolution include particle and pollutant dispersion as far as an
elaborate parametrisation of cloud physics.
The aim of this study is to use it as a tool for wind farm sitting and focus on the microscale physics of the
so called atmospheric boundary layer. A special attention will be devoted to the interaction between the
topography and the induced turbulence. A clear assessment of the turbulent scheme will be detailed and
its limitations and improvement will be discussed.
Keywords: Mesoscale model, turbulence, wind sitting.
1. Introduction
Generally, advanced CFD is used as a model to study the flow structure over complex terrain [1]. The
alternative proposed here is to use a meteorological model to catch the microscale effect such as
turbulent flux and wind speeds near the surface.
RAMS model is a regional meteorological model that can provide a complete description of the
mesoscale effects of turbulence. The usual low resolution turbulence models(Mellor-Yamada) is often
applied blindly at higher resolutions that does not fit to microscale turbulence characteristics [2]. Two
equation turbulence model E-ε and one equation E-l has been implemented this last decade [3]. This last
development enables the use of very high grid resolution (order of meters) and models very precisely
particle dispersion in complex area such as urban canopy [4]. The effort that have been deployed to
“tune” this mesocale model in order to solve turbulence on smaller scale is an asset in modelling wind
speed and turbulent kinetic energy for wind sitting purpose.
The advantages of using this model is a complete description of the thermodynamic equations and the
study of interaction between mechanically forced wind over stable unstable or neutral conditions.
A complete parametrisation of radiation and moisture processes allows a better description of the
microscale physics. Several other capabilities are implemented such as data assimilation and nudging [5].
Models prediction can be significantly improved using stochastic methods such as Model output statistics
[6] or recently developed Kalman filters [7,8].
1
Corresponding author: Nicolas Barranger
University of Athens, Department of Physics, Atmospheric Modeling and Weather Forecasting Group, University Campus, Bldg
Phys-5, 15784 Athens, Greece, Tel. +30-210-7276835, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 4/167
Book of Proceedings
2. Turbulent scheme and their implementation in RAMS
RAMS allows the use of 4 different turbulent schemes which are or Yamada 2.5 Level [9,10], Deadorff
[11], E-l and E-ε [12]. It solves the Navier Stokes equation using finite difference method and use a first
order approximation to compute turbulent flux using K-theory.
Mellor Yamada Level 2.5 scheme is aimed to calculate vertical diffusion coefficients and therefore
horizontal diffusion terms are neglected. The diffusion coefficient is computed as follow:
K m= S m l 2 E 1/ 2
where l is the master turbulent length scale, Sm is a function that depends on empirical
constants related to shear stress. The master length scale is computed using the Blackadar [13]
formulation. A prognostic equation is set up in order to calculate turbulent kinetic energy E in the
boundary layer approximation. The diffusion coefficient for turbulent kinetic energy are computed in the
same way as the momentum one, the two coefficients differs from a constant.
The Deadorff scheme is a large eddy simulation model. It means that it solves the subgird scale model
using subgrid filter. Here the filter is the grid itself. There we consider the turbulent length scale l to be the
equal to the grid box length. In other words l= ∆x .
The turbulent momentum coefficient is computed using the formula:
K m= 0.1l 2 E
1/ 2
The prognostic equation for the TKE is the same as mentioned for the Mellor Yamada scheme. The shear
production is now computed in all direction.
The E-l scheme is also based on a complete TKE equation and the turbulence is also defined as
anisotropic. It means that turbulent diffusion coefficients are the same in each directions. Practically, it
has to be pointed out that this scheme represent well small scale turbulence and that the grid domain size
must be equal in each direction.
The E-ε scheme is a two equation turbulence scheme that use prognostic equation from turbulent kinetic
energy and its dissipation rate. It is well adequate for small scale anisotropic turbulence for the same
reasons mentioned in the two previous schemes.
3. Advantages of the model
On smooth surfaces, the four schemes represent well the turbulent kinetic energy and wind speed and is
coherent with data observations. However, on complex terrain each model gives significant deviation from
data experiments [14]. The Mellor yamada Level 2.5 is generally used for low resolution simulation. It
uses the boundary layer approximation that is not well suited for three dimension topography. However it
has been pointed out that It catches well recirculation immediately down a hill and gives correct values of
wind speed. The three other schemes better predict turbulent kinetic energy profiles in complex
topography. The Deadorff scheme is very sensitive to the grid generation since the filter used (large eddy
simulation theory) is the grid itself and assumes that over the grid scale the turbulence is developed by
the bulk motion. For E-l model a correct boundary layer parametrisation is essential since it uses one
equation for turbulent prognostic but its implementation into RAMS gives the best values of turbulent
kinetic energy profiles. E-ϵ provide a complete equation system for turbulence closure and provide a good
transport of turbulent kinetic energy towards higher level.
4. Conclusions
This paper covers the state-of-the-art in using the regional model RAMS on high resolution. Details about
turbulent parametrisation have been presented. It clearly shows that Mellor Yamada model has some
limitation concerning high resolution simulation and is not appropriate for isotropic small scale turbulence.
The E-ε scheme is the most advanced high resolution turbulent scheme available and still needs further
development in its implementation into RAMS. E-l and Deadorff scheme have been demonstrated to be
the most reliable schemes and are a good basis for validation purpose.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 5/167
Book of Proceedings
The future works will be to improve the turbulent schemes and to assure that they are working in a large
range of situation such as a multiple choice of topography in agreement with complex terrain diversity. In
facts, E-ε model is highly influenced by topography and the closure hypothesis needs to be double
checked for each case.
Implementation of other turbulent schemes will be discussed as it could be better for comparison of high
resolution turbulence. K-ω model will be an advantage. Even if this model is not common in atmospheric
physics, it is often used in CFD models and seems to be more adequate than E-ε in modelling the
atmospheric boundary layer. A step further will be to take advantage of RAMS possibilities such as
running simulations in non-neutral conditions, use grid nesting and Lagrangian particle tracer for wind
turbines wakes.
References
[1] Stangroom P., CFD modelling of wind flow over complex terrain, 2004, PhD thesis University of
Nothingham.
[2] Hara T., Trini-Castelli S., Obha R., Tremback C., 2009: Validation of turbulence closure schemes for
high resolution in mesoscale meteorological models- A case of gas dispersion at the local scale,
Atmosheric environements 43 3745-3753.
[3] Trini Castelli S., Ferrero E. and Anfossi D., 2001: Turbulence closure in neutral boundary layer over
complex terrain. Boundary layer Meteorology, n. 100, 405-419.
[4] Trini Castelli S. and Reisin T. G., 2008, Application of a modified version of RAMS model to simulate
the flow and turbulence in presence of buildings: the must COST732 exercise, International Journals
of Environment and Pollution, in press
[5] Trini Castelli S., Reisin T. G. and Tinarelli G., 2008, Development and Application of MicroRMS
Modelling System to Simulate the Flow, Turbulence and dispersion in the Presence of buildings. Air
Pollution Modelling and its Applications XIX, Ed., 81-89
[6] Alessandrini S., Decimi G., Palmieri L., Ferrero E.,2009: A wind power forecast system in complex
topographic conditions, EWEC 2009 Marseille
[7] Galanis G., Louka P., Katsafados P., Kallos G., Pytharoulis I., 2006: Application of Kalman filters
based on non-linear functions to numerical weather predictions, Annales Geophysicae.
[8] Von Bremen L.,2007: Combination of Deterministic and Probabilistic Metorological Models to enhance
Wind Farm Power Forecasts: 2007, Journal of Physics: Conference Series 75.
[9] Mellor G., Yamada T.,1982: Devellopement of a turbulent closure model for geophysical fluid
problems, Rev. Geophysics. Space Physics. 20 , 851-875.
[10]
Mellor G., Yamada T.,1974: A hierarchy of turbulence closure models for planetary boundary
layers, Journal of atmospheric Science 31, 1791-1806.
[11]
Deadorff J., 1980: Stratocumulus-capped mixed layers derived from a three-dimentional model,
Boundary layer meteorology, 18, 495-527.
[12]
Trini Castelli S., Ferrero E., Anfossi D. and Ohba R., 2005: Turbulence closure models and their
applications in RAMS. Environmental Fluid Mechanics, 5, 169-192.
[13]
Blackadar A., 1962: The vertical distribution of wind and turbulence exchange in a neutral
atmosphere, Journal of Geophyscics 67, 3095-3102
[14]
Ferrero E., Tritelli S., Anfossi D., 2003: Turbulence fields for atmospheric dispersion models in
horizontally non-homogeneous conditions. Atmospheric environments.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 6/167
Book of Proceedings
Investigation of sources for lidar uncertainty in flat and complex terrain
Fernando Borbón Guillénaa,c,2, Paula Gómeza, Javier Sanz Rodrigoa,
Michael S. Courtneyb, Alvaro Cuervac
a
b
National Renewable Energy Centre of Spain (CENER), Sarriguren, Spain
Risø DTU National Laboratory for Sustainable Energy (DTU), Roskilde, Denmark
c
Technical University of Madrid (UPM), Madrid, Spain
Presented at the European Wind Energy Association Annual Event EWEA-2011, Brussels, Belgium, March 2011
Abstract
The main sources of lidar uncertainty have been studied for two consecutive measurement campaigns in
flat and complex terrain conditions. The same two lidar equipment have been used and compared to
standard cup anemometry. It has been verified that non-uniform wind flow plays a very important role in
lidar uncertainty. In this study, the non-uniform wind flows were caused by multi-MW wind turbine wakes
at the flat terrain site and by orography properties at the complex terrain site. Under these conditions,
lidar bias from standard anemometry showed values of up to 30% and 5% respectively.
The presence of low clouds or foggy conditions has been found to affect the lidar availability and the
measurements’ quality as well. Continuous wave lidars are in special more sensitive to these effects. The
bias introduce by these conditions can be in an order of magnitude higher than those caused by the
terrain conditions, which are the main interest of this article. For this reason, the effects of foggy
conditions have been filtered out as much as possible.
Keywords: Remote sensing, lidar, complex terrain, measurement uncertainty, wind resource.
1. Introduction
The present work shows the results observed from two subsequent wind speed measurement campaigns
using the same lidar equipment and compared to standard cup anemometry. The first campaign was
performed at flat terrain conditions and the second one in complex terrain conditions. The main sources
of lidar uncertainty are studied and special attention is paid to the sources of non-uniform wind flow,
which has been accredited to play a major role on lidar uncertainty.
Literature indicates that lidar technologies have close correlation to standard cup anemometer
measurements in flat terrain conditions. However, in complex terrain, the performance is degraded.
Bingöl et al. [1] have defined the vertical wind speed gradient as the main source of deviation in lidar
measurements with bias in the mean wind speed in the order of 5% to 10%. Still, some considerable work
has to be done in order to reduce the lidar uncertainty [2].
2
Corresponding author: Fernando Borbón Guillén
National Renewable Energy Centre (CENER), Ciudad de la Innovación, nº 7; 31621 Sarriguren (Navarra), Spain. Tel.: + 34 948 25
28 00, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 7/167
Book of Proceedings
2. Measurement campaigns
2.1 Risø Høvsøre test site: Flat terrain conditions.
Risø’s test site at Høvsøre is located in a flat terrain region, close to the coast line. It is surrounded by
crop fields and two big water masses, as represented in the map of Figure 1 (left). There are few farm
buildings in the surroundings and basically the biggest structures around are those of the six multi MW
wind turbines and their corresponding met masts as seen in Figure 1 (center). The mast and turbine
locations are indicated by two rows of colored points in Figure 1 (right). At the west side of the turbine and
mast rows, the lidars and the reference mast locations are indicated. The lidar scanning discs at several
heights are also represented in the latter figure.
The reference met mast is instrumented with a top mounted cup anemometer at 91 meters height.
Additional anemometers are installed on southern booms at 89, 71, 51 and 31 m. Moreover, two cup
anemometers are installed on northern booms at 71 and 51 m. There is a wind vane at 89 m on the
northern boom as the only wind direction reference. Furthermore, pressure and absolute temperature
sensors are installed at 89 m. Finally, at 3 m there are installed relative humidity, rain and temperature
sensors to complete the mast instrumentation.
Figure 1. Høvsøre test site. Location map, aerial view and lidar sitting.
Figure 2. CENER’s Alaiz test site. Lidars and met mast location indicated.
2.2 CENER Alaiz: Complex terrain conditions.
The second measurement campaign is currently been performed at CENER’s Alaiz test site which is
located at the top of a mountain of approximately 700 m height above surrounding plateaus. The
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 8/167
Book of Proceedings
mountain has a uniform slope facing north which extends nearly parallel to the west-east direction. The
remote sensing devices are installed on the ridge top, as seen in Figure 2. Towards the south-east end of
the ridge, the orography becomes more complex. The met mast has been equipped with cup
anemometers to sense the horizontal wind speed, propeller anemometers to sense the vertical wind
speed and wind vanes to sense the horizontal wind direction at 78, 90, 102 and 118 m height. Also
temperature, rain and atmospheric pressure sensors are present at several heights.
Correspondingly, the lidars were programmed to sense the wind velocity at the four different heights
coinciding with those from the met mast instrumentation. Measurements have been recorded during a
period of approximately five months.
3. Theory
Lidar devices measure wind velocity component parallel to the laser beam direction (so called radial or
line of sight velocities). These measurements are spatially located in the perimeter of a horizontal circle
situated at the desired height, as indicated by the points A and B in Figure 3 where the actual wind
vectors are inserted. Based on these measurements, an internal algorithm estimates the wind velocity. In
flat terrain conditions where the wind flow is supposed to be highly spatially homogeneous (see wind flow
streamlines in Figure 3); the calculated wind vector (indicated by C in the same figure) is not so different
from the measured wind vectors.
Figure 3. Lidar and met mast located in flat terrain and complex terrain conditions. Wind flow streamlines are shown. The
wind velocity vector is drawn in locations A and B at spatial points where the lidar scans the radial velocity component.
The velocity vector calculated by the lidar is drawn at the circle centre marked as C.
Conversely, the presence of big obstacles as a multi MW wind turbine can alter considerably the wind
flow sensed by the lidar. Therefore, this can introduce more uncertainty in the measurements. For the
case of complex terrain conditions, the uniform wind flow is disturbed by the irregular terrain orography or
for instance by the presence of forest regions. Therefore, changes in direction, vertical tilt, turbulence,
flow acceleration and wind profile changes are present. Subsequently, the measured vectors at points A
and B can be considerably different both in direction and magnitude and do not necessarily represent
properly the wind vector at the circle centre. As expected, the calculated wind velocity (based on the
assumption of a homogeneous wind field) results in more discrepancy when compared to point
measurements at the desired height; obtained with a cup anemometer installed in a met mast.
Furthermore, lidars work emitting laser beams that find it difficult to propagate in air with strong humidity
condensation. Rain and especially low clouds or ground fog can affect negatively the lidar performance
and reduce its availability. The use of rain and cloud detectors like ceilometers can be of great benefit to
filter out time periods with adverse atmospheric conditions.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 9/167
Book of Proceedings
3. Results
3.1 Correlation lidar cup and data filtering
The present work summarizes the partial results from two measurement campaigns using the same two
lidar equipments. As indicated above, the first campaign was performed in flat terrain and the second one
is being performed in complex terrain conditions. In both cases, the reference wind speed is taken as the
one given by the cup anemometer at the corresponding measurement height. From the two campaigns,
the only coincident measurement height is at 90 m. Therefore, the inter-comparison has been done using
measurements at this height, unless indicated otherwise.
Firstly, the correlation of the horizontal wind speed (U) between cup anemometer and the lidars is
presented in Table 1. As can be seen, the ZephIR and Windcube measurements for the horizontal wind
speed are compared with those from cup anemometers for flat and complex terrain conditions. Notice that
in the four figures, the correlation is generally good once some data quality filters have been applied. The
sources of bias in the lidar measurements will be explained in section 3.2.
Table 1. Correlation of the horizontal wind speed (U) between lidars and cup anemometer at 90 m height for flat and
complex terrain conditions. The lidar data filters applied are indicated in the “Comments” column.
Regarding the data filtering criteria, there are some variables in common and some others very specific
for each kind of device. For the cup anemometer data, the first filtering criterion is that the ambient
temperature must be higher than 2 ºC to avoid frozen or braked anemometers. Also, only wind speeds
higher than 4 m/s are considered due to the calibration range of the cup anemometers. The higher limit of
this calibration range is normally 16 m/s, however data above this limit was still considered in order to
assess the lidar performance at these velocity values.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 10/167
Book of Proceedings
For the case of lidar data, rainy periods are filtered out since rain affects lidar measurements (vertical
wind speed component and data availability) [3]. Additionally, each lidar registers several signals apart
from the measured wind speed. These signals offer the chance to filter out the data based on the device
individual properties. For instance, at each 10-min period, the ZephIR records the number of circular
scans at each height that are accomplished during each 10-min period. A minimum amount of scans is
selected to assure that enough data was recorded and the 10-min period average is representative of the
wind conditions during that period. This amount depends whether the ZephIR is scanning once (as in
Alaiz) or three times per height (as in Høvsøre). Respectively, 28 or 50 scans-in-average are chosen as a
quality filter. There is not a 3 factor difference since the scanning time increases but the lens focusing
time between heights remains constant.
Moreover, the average number of radial velocities retrieved at each circular scan is used as a filter since
this allows determining whether there was enough backscatter from the atmosphere from all directions or
whether there was any obstacle for the laser beam. This is important since the ZephIR’s processing
algorithm performs a fit to a rectified cosine function based on the number of points-in-fit. If there are few
points, the uncertainty of the fitted function is higher and that is why a minimum requirement of 35 (one
scan per height) or 105 (three scans per height) is selected as a data filter.
For the ZephIR, there is an additional parameter that helps to identify the goodness of the fitting function.
The turb parameter indicates how much deviation was there in general from the measured points to the
estimated function to be fitted. The turb parameter gives information with resemblance to what in statistics
offers the sum of squared residuals. Consequently, higher values of the turb parameter can be an
indication of considerable un-uniformity of the wind velocity field where the ZephIR was scanning. This
can help to identify very turbulent wind flows. Additionally, the turb parameter can be an indicator of noise
in the retrieved signal. For this study, only datasets where the turb value was 0.1 or less have been taken
into account.
Furthermore, the ZephIR’s scaling factor is a parameter related to the strength of the backscattered
signal. It can be understood as the gain that has to be added to the incoming signal in order to detect it
properly. A weaker signal would need a higher scaling factor. Normally, weak signals indicate very clear
air where low aerosol concentration backscatters just a small part of the energy. On the other hand, a low
scaling factor value indicates that the returning signal was strong enough to be easily detected. This can
reveal the presence of higher aerosol concentration and most importantly for our interest, the presence of
low clouds or fog at ground level. In general, a scaling factor higher that 25 or 50 from the lowest
measuring height (38 m as default) can help to filter out periods with high presence of foggy conditions.
For the case of Alaiz test site, since it is a considerably high mountain surrounded by flatter regions,
ambient humidity frequently condensates around its top conducing to highly foggy conditions during
winter months.
In respect to the specific Windcube data filtering criteria, this is basically reduced to the device availability.
This parameter indicates how much of the time during each 10-min period, the device was able to
properly measure the signal backscatter and estimate an equivalent wind vector. In this study, only
periods with 100% of lidar availability where chosen. The reasons of lower availability can be due to
signal obstruction due to foggy conditions or low clouds.
It is important to notice that the Windcube availability has been used as a filtering criterion for the ZephIR
as well. The hope is that most of low availability periods are caused by the presence of fog or very low
clouds and consequently this help us to filter out data that can negatively affect the ZephIR performance3.
3
Differently to continuous wave lidars as the ZephIR, the Windcube is a pulsed wave lidar. The main difference in the working
principle is that continuous wave lidars sense the signal backscatter constantly. The main assumption is that the returning signal
comes from the desired measuring height which is achieved by using lens to focus the laser energy there. If the signal is actually
backscattered from a different height, there is no a direct way to detect it. Contrary, the pulsed lidar emit a signal pulse of a specific
length and calculates the time necessary to reach and return from the desired measuring height. Then it opens a temporary sensing
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 11/167
Book of Proceedings
As a final filtering criterion that in general affects the performance of all the devices is the presence of
obstacles that obstruct the free wind flow and can produce wind acceleration or form wakes with
considerable wind speed reduction and turbulence intensity. The first example is the presence of the met
mast itself that directly affects the flow around with distortions that reach the cup anemometer location. It
is necessary to filter out data based on the wind direction in order to reduce as much as possible the met
mast effects. It has been seen that the mast wake effect differs depending on the mast structure.
Additionally, for those wind directions where there is not direct wake incidence in any of the cup
anemometers, the mast effects are still noticeable. For instance, if plotting the ratio of two cup
anemometers (at the same height but at opposite sides of the met mast) as function of the wind direction,
a kind of sinus wave shape reflects how the flow is disturbed by the mast presence. This information
could be used to “correct” the cup anemometer measurements in order to reduce the mast effects [4].
Nevertheless, this approach has not been performed for the present study.
Other aspect taken into account when selecting the wind direction sectors to be used for the lidar-cup
correlation is the presence of any big structure nearby, or of especial terrain conditions. These structures
can produce strong wind turbulence or flow obstruction at several scales affecting differently the cup
anemometers and the lidars. This is explained with more detail in section 3.2.
Despite applying the mentioned filters, still there exists some dispersion in the correlation graphs,
especially at complex terrain conditions. A first attempt to identify the main lidar bias sources affecting the
lidars in flat and complex terrain is introduced in section 3.2. As a final remark for this section, the present
study is based on two early lidar units and therefore the results might differ to some extent if a different
device would be used. Furthermore, second generation lidar devices of each brand are already available
in the market and performance improvements can be expected.
3.2 Lidar bias sources
When facing the task of comparing the lidar performance in flat and complex terrain conditions, it is
necessary to identify what are the main sources of bias between lidar and cup anemometer
measurements. It can result in a big challenge since diverse variables can affect cups and lidars
differently and some others can affect them simultaneously. Ideally, it should be possible to identify the
individual effect of each variable over each device, however some bias sources can appear
simultaneously and therefore separating their influence becomes a difficult assignment.
For the case of flat terrain as shown in figures from Table 2, there is a wind direction sector from which
the lidar bias is more obvious (saw shape). It was identified that from this direction region, the presence of
turbine wakes affected both the cup and lidar measurements. Depending on the direction angle, the wake
could be impacting only the cup and not the lidar and vice versa, therefore the big positive or negative
differences in the wind speed values sensed by them. Other issue is that the lidars are scanning a
perimeter whose diameter length scale is comparable to the one of the turbine wake, while the cup
anemometer can be considered as a point in space.
As mentioned before, some variables are difficult to separate when studying the bias sources. During the
measurement campaign at Høvsøre, it was found that most of the low cloud presence was precisely
when wind was blowing from the sector with turbine wakes. Therefore, the influence of clouds in the lidar
measurements (especially in the continuous wave lidar) was difficult to separate from the turbulence and
speed deficit due to the turbine wakes.
window to measure the backscatter. If the laser signal is blocked before reaching the desired height or if there is almost not
backscattered at all due to very clear air, during the sensing window there is be nearly no signal at all to be measured and a “null”
registry is recorded. These two sensing approaches have both advantages and disadvantages as have been previously discussed
by other authors [5], [6].
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 12/167
Book of Proceedings
Table 2. Lidar horizontal velocity bias vs. wind wane direction.
The complex terrain graphs show that the lidars tend to underestimate the wind velocity from northern
and southern sectors, precisely where the there is more tilt in the flow due to the alignment with the
mountain slope. As the wind direction changes and the tilt angle is reduced as wind flows parallel to the
mountain flatten top (eastern and western sectors), the lidar bias is reduced. At 270º there is a group of
data revealing the mast effect. Here the cup anemometer might be sensing a reduced wind speed due to
the wind flow obstruction caused by the mast.
Additionally to the horizontal component of the wind vector, the lidars are also able to determine the
vertical component (W). Unfortunately, at the flat terrain campaign there were no sensors at the met mast
to measure the vertical wind velocity. For this reason, just the absolute value of the vertical component
sensed by the lidars is presented in left figures from Table 3. Notice here the difference between the
ZephIR and the Windcube measurements in terms of magnitude. The sinuous wave curve seen in the
Windcube data is suspected to be caused by a mis-aligned internal mirror that was removed and replaced
during a previous experiment. It is probably not typical of a standard Windcube. Though, this figure helps
to highlight the importance of correct leveling (use an accurate level sensor) when installing any lidar.
Notice again the saw shape in the curve due to the turbulence from the turbine wakes.
At the complex terrain site in Alaiz, the mast has installed vertical propeller anemometers at several
heights. The two graphs at the right of Table 3 show the lidar bias of the vertical wind velocity at 118 m
height. The reason of using this height instead of 90 m as previously is that data availability from the
propeller at 90 m is much reduced. In this set of graphs is difficult to identify a clear behavior due to the
data dispersion in the plots. However, observed that when wind blows from south (180º) the lidars tend to
overestimate the vertical component (with negative magnitude in this case). Contrary, when wind blows
from north, the lidars seem to underestimate the vertical component, and therefore the bias is negative
too. The gap in the data from around 25º to 145º is simply because there is almost no wind blowing from
that sector at this site. The wind rose is very directional north/south for Alaiz.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 13/167
Book of Proceedings
Table 3. Lidar vertical velocity bias vs. wind wane direction.
Remember that the lidars and the mast are installed precisely at the mountain edge where the northern
uniform slope gives place to a flatten hill top. So the lidar beams are sensing radial velocities above and
incline surface at north and at a flat surface at south. This result in measuring different wind vectors
assuming the flow follows the ground contour as in left figure from Table 4. Besides, it is important to
mention that the Windcube vertical component is recorded with the opposite sign as the ones registered
by the ZephIR and the propeller anemometers [3]. For this reason the Windcube value of –W has been
used instead.
Several authors have pointed out that the main lidar source of uncertainty in complex terrain conditions is
the vertical wind speed gradient [1], [7]. To verify this hypothesis, the lidar horizontal velocity bias is
plotted as function of the vertical velocity (not the gradient since it is not possible yet to estimate it from
the available sensors at Alaiz) in figures from Table 4. As explained above, since the lidars are installed
at the mountain edge, this is precisely the location where we expect the highest vertical speed gradient.
The two graphs show very concise information about the effect of the vertical component at this siting for
the lidar measurements. Very similar results are obtained if the tilt angle is used instead of W. It is clear
that when the vertical component increases in magnitude, the lidars tend to underestimate the horizontal
wind speed. Different slopes are observed whether the wind is blowing uphill or downhill. The reason of
this behavior is not totally understood yet and further analysis is needed.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 14/167
Book of Proceedings
Table 4. Lidar horizontal velocity bias vs. propeller vertical velocity.
In previous paragraphs, the influence of wind turbulence was indicated to play an important role in lidar
uncertainty. With that purpose the lidar bias is plotted as function of the turbulence intensity measured by
the cups as shown in Table 5. Figures from flat terrain conditions show there is not a perfect correlation,
but at least it is clear that the highest lidar bias occurs in periods when there is high turbulence intensity.
The graphs include all the wind directions. Here it is important to mention that the definition of turbulence
intensity is the wind speed variance divided by the mean wind speed during each 10-min period. Using
this concept and based on cup anemometer measurements, it is not possible to know if the wind flow was
for example very uniform and even uni-directional and the only parameter changing was the wind speed
magnitude or if the flow was very spatially chaotic like that one inside a turbine wake. In the case of point
measurements like the cup anemometers there is no a big impact but for volume measuring lidars this
distinction should be more relevant to take into account.
In the case of complex terrain conditions, there were no big structures in the surroundings and the wind
turbulence can be considered as ambient and terrain contributions only. Here the correlation between
lidar bias and turbulence intensity measured with the cups is once again not very clear and using another
variable related to the wind flow uniformity might be more useful.
A final variable studied for this work is the influence of the wind shear over the lidar bias. Since probe
length can be of the order of several meters ate the studied heights, the wind shear might have an impact
in what the lidar is measuring since the volume average of the wind speeds is at this region might differ
from the actual speed at precisely the desired height. In Table 6, the lidar bias distribution at the lowest
measuring height, for the two locations, is presented at the left column. Notice that in flat terrain the bias
distribution is centred at zero while in flat terrain at a negative value. This means in the complex terrain
site, lidars mostly underestimate the wind speed. The second column shows the lidar capability to sense
the wind shear. In the flat terrain site the wind shear is in general very vertically straight but in the
complex terrain site is possible to find higher (even negative) differences between the speed at higher
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 15/167
Book of Proceedings
and lower heights. The two lidars in general show similar results to cup anemometers but still seem to be
slightly less sensible to detect the wind speed differences at different height when they are in the range
from 0 to 1.2 m/s.
Table 5. Lidar horizontal velocity bias vs. turbulence intensity.
Table 6. Lidar horizontal velocity bias at 40m as function of the wind shear properties.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 16/167
Book of Proceedings
Fitting the measured wind profiles to the power law function U(z) = Ur (z/zr)α, we obtain the parameter
that helps to characterize the wind profiles. Plotting the lidar bias as function of this α parameter can
reveal how the wind shear affects the lidar bias. This is shown in the right column figures of Table . There
seems to be a not a clear correlation for the flat terrain case, but for the complex terrain site, the lidar
underestimation of the wind speed seems to be reduced as the α parameter increases, this means, when
the wind speed is faster at higher heights than at lower heights. This insight gives an important motivation
to further study the influence of the atmospheric stability over the results obtained so far.
4. Conclusions
The performance of two different lidar systems has been compared to cup anemometer measurements at
two different locations. The first measurement campaign was realized at flat terrain conditions while the
second one at complex terrain conditions.
Identifying the main sources of lidar uncertainty (assuming lidars and cups are properly calibrated) is a
difficult task since the studied bias sources can affect the measurements simultaneously and identifying
their individual contribution is sometimes not possible.
Lidar measurements have shown to be sensitive to the wind flow uniformity (i.e. spatial wind velocity field
variation). The origin of non uniform flows can be due to the presence of big structures like multi MW wind
turbines or due to orography conditions like the presence of mountains, hills, forests, etc. It was shown
that the variation of the vertical component of the wind velocity plays an important role in the lidar bias
occurrence for the Alaiz campaign. It is still necessary to verify if this relationship is causal or if at the lidar
siting (mountain edge) the vertical velocity gradient dW/dx scales with W.
Since lidars work emitting a laser beam, the presence of low clouds or foggy conditions can be an issue
that affects the availability of the device and the quality of the measurements. For the case of the
continuous wave lidar used during the measurement campaign, the bias due to very low clouds of foggy
conditions can reach an order of magnitude greater that the other bias sources. Therefore a proper
methodology to identify the occurrence of these conditions is very important to assure the quality of the
collected data. The continuous wave lidar has been recently upgraded with a new firmware version that is
expected to reduce these effects. Yet, more data is necessary to be collected in coming months to asses
its effectiveness.
It is important to remember that comparing the lidar data to cup anemometer data, there is always
uncertainty in both sensing devices and certain factors can affect their performance either separately or
simultaneously. For the cup anemometer, it is of great significance to asses the influence of the met mast
where it is installed, the response to the wind flow tilt angle and the effects of icing that can not only stop
completely, but also slow down the normal anemometer rotation.
Further lidar bias analysis is under development and new variables will be integrated, mentioning with
special importance the influence of the atmosphere stability conditions. Furthermore, the use of fast data
will be implemented in order to compare instantaneous lidar and cup anemometer data rather than 10min averages. The purpose of this study is the development of a lidar bias correction methodology for
complex terrain conditions.
Acknowledgements
Thanks to the technician personnel at both CENER and Risø-DTU for their support setting up and
keeping running all the referred equipment. The measurement campaigns are financially supported by the
FP7 SAFEWIND project. The author is financially supported by the FP7 WAUDIT project.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 17/167
Book of Proceedings
References
[1] Bingöl, F. et al. Lidar performance in complex terrain modeled by WAsP Engineering. EWEC 2009
Proceedings. Marseille, March 2009
[2] Lindelöw, P et al. Wind shear proportional errors in the horizontal wind speed sensed by focused,
range gated lidars. IOP Conf. Series: Earth and Environmental Science 1 (2008) 012023
[3] Antoniou,I. et al. Remote sensing (UPWIND WP6) six-month progress report. Risø National
Laboratory for Sustainable Energy. Roskilde, November 2006
[4] Lindelöw, P et al. Flow distortion on boom mounted cup anemometers. Report number: Risø-R-1738
(E). Risø National Laboratory for Sustainable Energy. Roskilde, July 2010
[5] Jaynes, D. et al. Massachusetts Technology Collaborative: Final Progress Report: LIDAR. University
of Massachusetts, July, 2007
[6] Wagner, R. Accounting for the speed shear in wind turbine power performance measurement. RisøPhD-58(EN) - Short version. Risø National Laboratory for Sustainable Energy. Roskilde, April 2010
[7] Boquet, M. Combination of Wind Lidar with CFD tools for improving measurements in complex terrain.
LEOSPHERE SAS. Orsay, 2010
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 18/167
Book of Proceedings
On the application of Principal Component Analysis for accurate statisticaldynamical downscaling of wind fields
Roberto Chávez-Arroyoa,b, Sergio Lozano-Galianab, Oliver Probsta,1
a
b
Tecnológico de Monterrey, Monterrey, Mexico
National Renewable Energy Centre of Spain (CENER), Sarriguren, Spain
From Energy Procedia 40: 67-76; European Geosciences Union General Assembly 2013 (EGU) Division Energy, Resources & the
Environment (ERE); http://dx.doi.org/10.1016/j.egypro.2013.08.009
Abstract
A new methodology for the accurate statistical-dynamical downscaling of surface wind fields for long-term
periods is described in this work. This new method is based on stratified sampling of long-term mean Sea
Level Pressure fields combined with Principal Component Analysis for the determination of the most
representative synthetic year. Validation is performed with 9 years of dynamically downscaled wind fields
for the Iberian Peninsula obtained with the mesoscale model SKIRON. The results show that compared to
the traditional method of dynamically downscaling random annual periods, the error in the predicted
average wind speed is reduced by almost 30%.
Keywords: Statistical-dynamical downscaling; long-term wind resource assessment; principal component analysis; mean sea level
pressure
1. Introduction
Methods providing an accurate estimation of the long-term wind resource are important for a range of
applications including wind resource assessment, where an accurate knowledge of the long-term wind
resource at mesoscale levels is required both in the early prospection and project development phases,
as well as for financing and insurance. Since extended high-quality wind measurement records as
suitable references are not always available in many parts of the world, mesoscale modelling or
downscaling tools emerge as a natural alternative. The basic idea is the construction of regional wind
maps derived from long-term large-scale data sets such as global reanalysis data sets. Three main
approaches can be distinguished: (1) Dynamical , (2) statistical, and (3) statistical-dynamical.
In the dynamical method (e.g. [1-4]) a large-scale data set is downscaled to the regional level by using a
dynamical computational model of the atmosphere such as WRF-ARW [5] or SKIRON [6] based on
temporal and spatial initialization by the global model. Available long-term global data sets are thereby
translated into corresponding regional maps from which time series for the assessment of long-term
averages and fluctuations can be extracted. Although the dynamical downscaling of either global or
otherwise coarse numerical weather data is the most physically consistent and in principle most accurate
method to simulate the wind characteristics at a regional level, the computational cost of downscaling a
long-term global data set to the regional level is considerable, and the computing time achievable at
standard facilities is often not acceptable. An alternative approach are statistical procedures (e.g. [7-10])
which attempt to establish statistical relationships between large-scale and mesoscale phenomena and
use these relationships for the long-term prospection at the site of interest. A hybrid class of methods
combining the strengths of both preceding methods at a reduced computational cost are statistical-
1
Corresponding author: Oliver Probst
Physics Department, Tecnológico de Monterrey, Eugenio Garza Sada 2501, Monterrey 64849, MexicoTel.: +52-81-8358-2000,
e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 19/167
Book of Proceedings
dynamical methods [11-25]. The underlying idea in most statistical-dynamical downscaling (SDD)
methods is the assumption that the climate of a region can be classified into several typical weather
situations which should be sufficiently characteristic of the long-term behaviour of the phenomena under
study. Afterwards, each group is integrated with a high-resolution numerical weather model to predict the
local climate features [15].
Although several authors have described SDD methods, based on either stationary classification
schemes [11-20] or on the classification of weather episodes of different duration using linear and nonlinear methods [21-26], it is worth noting that comparisons against purely dynamical procedures are
scarce, in particular for the wind statistics important for wind power engineering such as the seasonal and
daily wind profiles, wind frequency roses and average wind speed.
SDD methodologies based on the stationary classification of synoptic-scale weather regimes are
generally physically meaningful and often lead to a reasonably small number of classes. There are
however several important downsides limiting their usefulness for wind prospecting purposes. First of all,
the statistical treatment of the input data leads to a loss of time-dependent phenomena. Secondly, the
simulation of daily or seasonal phenomena, required for a meaningful validation, is difficult because of the
large number of classes required for such a distinction. An alternative are SDD methods based on
algorithmic approaches that look for dynamical or quasi-dynamical weather episodes; such approaches
allow for the simulation of entire weather cycles, and their predictions are easier to compare with
measurements. Fuentes and Heimann [20] selected particular multi-day mesoscale episodes by
hierarchical cluster analysis in combination with a spatio-temporal distance measure to aggregate
consecutive dates with similar synoptic patters described by the first Principal Components of the
geopotential at 500hPa. They then dynamically downscaled those representative episodes selected by
the k-means cluster algorithm applied to the episodes with the same temporal length. Hagemann [21]
proposed the use of Self-Organizing Maps (SOM) to select a representative continuous set of 365 days
which were afterwards downscaled dynamically. The selection was based on the evaluation of those 365days sets that best matched the frequency of occurrence of each SOM node (clusters) compared to the
long-term clusters’ frequencies. Similarly, Hahmann et al. [22] also proposed the application of SOM for
the generation of wind speed atlases by using SOM as a synoptic classification tool to select
representative samples of large scale wind forcing. Rife et al. [23] also identified representative periods
and based their selection on testing the similarity between the long-term period and a random sample
drawn from a large number of reduced daily sets. The similarity metric was based on wind speed and
direction distributions obtained from the first native vertical level of the MERRA global reanalysis data set
for the point that best correlated with the measurements. In a similar approach Tammelin et al. [24]
dynamically downscaled 4 years of different monthly periods requiring the geostrophic wind components
(winds at 850hPa) of each 48-month period to be as representative as possible of the wind speed and
direction distributions of the reference 19-year long-term period. In a recent work, Martínez et al. [25] use
Principal Component Analysis to determine the Empirical Orthogonal Functions (EOF) corresponding to
the long-term large-scale climate and downscale them to the regional scale. By applying the Principal
Component time series of the large-scale data set to the downscaled version of the corresponding EOF
they were able to extract time series which could be compared to measured wind speed records.
In the present work, a new methodology is proposed which is based on the determination of an optimal
set of 365 daily episodes which best represents the long-term large-scale climate in a region of interest.
Maps of the mean sea level pressure (SLP) field from the NCEP-DOE global reanalysis II project [27] are
used as a proxy for the large-scale long-term climate. Once this optimal or representative 365-day largescale climate set has been determined, it can be used for dynamical downscaling for the creation of
regional surface wind field data sets. The selection of this period is based on the following steps: (1)
Creation of a large number of monthly data sets by randomly sampling the long-term data base using a
technique similar to the one by Rife et al. [23]. This approach is also known as stratified sampling. (2) The
Empirical Orthogonal Functions (EOF) both for the long-term fields and the random 365-day sample are
calculated and compared to each other by calculating the Euclidean distance between the two EOF sets.
The optimal or representative sample is chosen by the requirement that the distance between the EOF
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 20/167
Book of Proceedings
sets should be minimal. (3) The results are validated by comparing the wind fields obtained from
dynamically downscaling 9 years of NCEP/GFS (National Centers for Environmental Protection / Global
Forecast System) global climate data for the Iberian peninsula using the SKIRON model [6, 27] against
the wind fields extracted from the SKIRON results for the representative period determined in step (2).
Since SKIRON runs were conducted on a daily basis (36-hour runs, discarding the initial 12 hours of each
period to account for spin-up) this extraction procedure is equivalent to downscaling the 365-day
representative period. Wind statistics relevant for wind resource assessment defined in section 2.3 are
used for discussion.
2. Methods and Data
2.1 Selection of the most representative period
As mentioned above, the new method presented in this work belongs to the field of statistical-dynamical
downscaling of large-scale climate data. As a suitable variable representing the synoptic meteorology we
use the mean sea level pressure (SLP) field from the R-2 which has been successfully related to the wind
speed variability in literature [28, 29]. Following the approach proposed by Rife et al. [23] a large number
N of samples, typically N=100,000, composed of 365-day daily time series are created by random
sampling the long-term SLP data set, ensuring that each sample contains 28|30|31-days per month; this
is referred to as a stratified sampling technique. Then, each month of the random sample is compared to
the corresponding average month of the long-term period by using the procedure described below.
j
ij
The monthly (j-th month) anomalies for the long-term period and i-th sample set, X and Y , respectively,
are calculated by subtracting both the time average and spatial average from the sea level pressure
(SLP) field (Lui et al. [31])
 x1,1 ⋯ x1,S 


X ( t, s ) =  ⋮ ⋱
⋮  = SLP (t , s) − SLP (s) − SLP (t )
x

 T ,1 … xT ,S 
(1)
Where T is the time and S the spatial dimension of the field X (t, s); X stands for either X j or Y ij. Principal
Component Analysis (PCA) is applied to both X j and Y ij, yielding an orthonormal base, commonly known
as the set of Empirical Orthogonal Functions (EOFs), for each field. We will refer to these bases as {e i, m
X
} and {e ij, m Y}, respectively. If {e m} is referred to as either set of EOFs, then the field X can be expressed
as
X ( t, s ) =
P
∑ αm (t )em (s)
(2)
m =1
Where the coefficients or amplitudes α m(s) are usually referred to as the Principal Components (PC) of
the field X. The basis {e m} is formally obtained from the eigenvalue problem of the covariance matrix Σ
of X, and the corresponding eigenvalues λ m represent the variance associated with each EOF em.
The idea behind the new method proposed in this work is to use the similarity between the sets {e i, mX}
and {e ij, mY} as a criterion of the representativeness of the sample Y ij with respect to the long-term field Xj
for a given month j. While in principle the full sets could be used for comparison, most of the variation of
the field is contained in the first few eigenvalues of the covariance matrix Σ. This allows to significantly
increase the computational efficiency of the procedure by limiting the comparison to the first M EOFs
which capture a certain fraction of the total variance. In our case, we found 70% to be a convenient
number, i.e.
M
S
m =1
m =1
∑ λm ≅ 0.7 ∑ λm = 0.7Tr ( Σ )
WAUDIT
Contract#
238576
(3)
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 21/167
Book of Proceedings
Where S, as before, is the spatial dimension of X and therefore the dimension of the covariance matrix. In
the second equality we have expressed the fact that the trace of a (square) matrix is equal to the sum of
its eigenvalues and invariant under changes of basis. Consequently, the total variance can be calculated
without having to calculate the full set of EOFs.
Once the reduced sets of EOFs have been determined, the sample Rj which is most representative of the
long-term base is determined as the monthly sample that minimizes the Euclidean distance
between
the bases {ej, m X}’ and {e ij, m Y}’, where the apostrophe (’) indicates that only the first M EOFs of each
original base have been considered:
δ ij =
M
∑
m =1
e jX,m − eijY,m
2
{ }
R j = X ij | δ ij = inf δ ij
i
(4)
(5)
An example of the results from the PCA decomposition method applied to the SLP fields is depicted in
Figure 1 where the first two eigenvectors from the 33-years SLP anomalies for December are shown.
These EOFs represent two of the most important SLP archetypes over Europe since their variance
accounts for more than 40% of the total variance and from which we can find the well-known pattern
associated with the North Atlantic Oscillation [32] in the first EOF and the Euro-Asian configuration for the
second EOF [33].
Figure 1: First two EOFs from the 33-years SLP December anomalies with their corresponding eigenvalues indicated at the
top.
In order to compare the new method to standard wind industry practices for the estimation of the longterm wind resource (as described by Rife et al. [23]) a reference method was also implemented, where
the representative period is obtained as a random sample drawn from the long-term period. In this
approach each calendar day of the representative year is determined from randomly selecting among the
corresponding set of repetitions of that calendar day within the long-term period, in this case for a 33-year
period. In the following we will refer to this method as the traditional one (TRAD).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 22/167
Book of Proceedings
2.2 Validation procedure –Iberian Peninsula case study.
The performance of the proposed statistical-dynamical downscaling method is assessed by the
comparison of the wind fields obtained with the regional weather model SKIRON for a domain covering
the Iberian Peninsula for the selected (representative) 365-days sample and the full 9-year period
modeled; the model domain is shown in Figure 2 (dotted area). The SKIRON model is currently
developed and supported by the Atmospheric Modeling and Weather Forecasting Group (AM&WFG) at
the University of Athens (NKUA) [28]. This regional model is based on the Eta/NCEP model which in turn
is built over the eta coordinates system in the vertical and over the semi-staggered Arakawa E grid in the
horizontal [6].
The SKIRON wind fields are obtained by dynamically downscaling the 12UTC cycle analysis from the
NCEP/GFS data down to a horizontal resolution of 0.05° x 0.05° with 45 eta levels in the vertical.
Additionally, daily SST (Sea Surface Temperature) information is used for model initialization. The
temporal horizon for the SKIRON runs is 36 hours; the first 12 hours are discarded to avoid the spin-up of
the model (see Gaston et al. [3] for more details). The results from the downscaling runs are postprocessed in order to obtain the wind values at 80m height by performing a power law interpolation from
the eta levels.
For the PCA methodology a wide enough domain from the R-2 SLP data that includes Europe, northern
Africa and part of Middle East is used in order to account for the synoptic patterns affecting the
Peninsular region (Figure 2). The SKIRON domain is limited to an area slightly larger than the Iberian
Peninsula with the purpose of leaving a buffer zone than can be used for the transition needed to
accommodate the dynamical and physical differences between the GFS data used for initialization and
SKIRON.
Figure 2: European window used for the PCA decomposition of the SLP fields from the R-2 data set, where grid points are
represented by black dots. The dotted black rectangle represents the domain covered by the downscaling performed with
the SKIRON model.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 23/167
Book of Proceedings
2.3 Validation metrics.
Several metrics are used to compare the long-term (9-year) time series and the time series generated
from the 365 selected days (representative sample R) for each point of the domain simulated with
SKIRON. These metrics are inspired by wind statistics that are meaningful for wind resource assessment
studies. The following metrics have been used in this work are described below:
The first metric is the average wind speed difference:
ε s = S LT − S RP
(6)
where SLT(x, y) and SRP(x, y) are the wind speed maps (80m) for the long-term and the representative
period, respectively. The second, third, and four metrics are the mean absolute error (MAE) and the
correlation coefficients (CC) for hourly and seasonal profiles, respectively:
εT =
ρT =
1 T
∑ SpLT − Sp RP
T p =1
cov ST LT , ST RP
(
)
σ (ST
)
LT
)σ (ST
RP
(7)
(8)
Where T = 12 (months) or T = 24 (hours) for the seasonal and hourly wind speed averages and
correlation coefficients, respectively. The last metric considered in this study is the weighted compound
frequency histogram defined as:
D=
nbins
∑
b =1
wb
fbLT − fbRP
fbLT
(9)
Where fbLT and fbRP are the relative frequencies for the long-term and representative period computed in a
total number of bins given by nbins for the wind speed and the wb is a weighting factor which was defined
as the long-term frequency i.e. wb = fbLT. The aforementioned validation procedure was carried out for the
PCA method and then compared with the results obtained with the traditional method described above.
3. Results
An example of the results of the PCA methodology in terms of the SLP patterns can be observed in
Figure 3 where the leading EOF for May is depicted for a) the period 2003-2012 as well as for b) the final
selection made by the PCA methodology and c) for a randomly selected period (industry standard
method). Despite the expected differences, it can be noticed that the di-pole system between Greenland
and the Azores Islands present in the long-term fields is well reproduced with the novel method proposed
in this work, but degenerated by the traditional method. This difference is illustrated by a comparison of
the Euclidean distance (equation 4) between the long-term and the sample set; this distance is 1.42 for
the traditional compared to 0.5 for the new PCA-based method.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 24/167
Book of Proceedings
Figure 3: First EOF for the SLP May anomalies for a) the long-term period, b) PCA the representative period selected by the
new PCA-based method, c) the period selected by the industry-standard method (TRAD) . The associated Euclidean
distance to the long-term EOF is shown in the lower right corner of figs b) and c).
εS
ρ12
|ε24|
|ε12|
D
0.8
1.2
ρ24
1
0.8
1
0.6
0.15
1
0.8
0.8
0.7
0.4
0.6
0.6
0.2
0.6
0.8
0.4
0.5
0
0.4
0.1
0.6
0
0.2
0.3
0.4
−0.4
−0.2
0.2
0.05
−0.6
0
−0.4
0.2
0.1
−0.2
−0.8
−1
0.2
0.4
−0.2
↑
↑
↑
PCA
TRAD
PCA
↑
TRAD
0
↑
↑
PCA
TRAD
0
↑
↑
PCA
TRAD
−0.6
↑
↑
PCA
TRAD
−0.8
↑
PCA
↑
TRAD
Figure 4: Resulting metrics (equations 6 - 9) from the comparison between periods selected by the PCA method and the
traditional method (TRAD) for the different wind-based statistics described in section 2.3.
The influence of this synoptic system on the surface winds is shown in the boxplots of Figure 4 which
illustrate the overall results obtained for the 99051 grid points in the SKIRON domain for the metrics
defined in section 2.3. In all cases, except for the correlation coefficients, the lower the values of the
statistics the better the prediction of the long-term wind resource. The more significant improvements can
be observed for the difference in average wind speed ε s and the monthly and daily wind cycle MAE (|ε 12|
and |ε 24|) in which not only the dispersion is reduced but also the median of all points is drastically
reduced by ~80% , ~25% and ~20% for ε s, |ε12| and |ε 24|) respectively.
Figure 5: Wind statistics for the virtual time series generated for one SKIRON point with a) the wind speed frequency plot
together with their Weibull distribution curves (dotted lines) where the corresponding scale (“c”) and shape (“k”) factors
are shown in the legends. The right- hand figure b) shows the monthly (dotted lines) and the hourly (solid lines) profiles for
the three time series.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 25/167
Book of Proceedings
These differences are illustrated in Figure 5 for a single point time series within the SKIRON domain
(coordinates [42.65 , -7.75]) in which the wind speed frequency is shown together with their
corresponding Weibull probability density function curves (Figure 5a) for the long-term, the new PCAbased method and the traditional approach (TRAD); the new method clearly provides an improved
prediction of the long-term Weibull parameters. Figure 5 (b) shows the monthly and hourly wind profiles
for these three data sets; in both cases the PCA-based method provides an improved prediction with a
noteworthy improvement in the daily profile which is very relevant for time-of-day generation of wind
farms.
Regarding the spatial distribution of the errors, it can be observed from Figure 7 that the new PCA-based
method produces a far more homogeneous error map than the industry-standard approach. Higher wind
error statistics are related to areas with higher wind variability as illustrated in Figure 6. This coincidence
was expected since extra variability is highly related to both windy areas and sites with strong influence of
local phenomena.
Figure 6: Standard deviation (m/s) for the SKIRON 9-years wind map.
4. Conclusions
A new approach for the statistical-dynamical downscaling for the generation of accurate long-term wind
maps is presented in this work. The new method is based on the construction of a representative period
of 365 daily episodes which best matches the long-term climate in a region of interest. The selection of
this period is achieved through a method based on Empirical Orthogonal Function from a large number of
sample candidates generated through a random process. As any statistical-dynamical downscaling
method the new approach offers a very significant reduction of computing time and power over the direct
downscaling of the long-term wind climate, while providing an accurate prediction not only of the average
wind speed but also of wind velocity parameters relevant to wind resource assessment, including wind
speed distributions and daily and monthly wind speed profiles. Compared to the industry-standard
approach for the generation of long-term wind maps based on the downscaling of a randomly selected
annual period very significant improvements in prediction accuracy have been achieved. A particularly
noteworthy improvement of this methodology is a better representation of the temporal variability with
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 26/167
Book of Proceedings
typical MAE values of 0.4m/s for the monthly and 0.15 m/s for the hourly profiles, which is around 30%
better than the industry standard method and which can be considered small when compared to other
important sources of uncertainties such as those related to the NWP physics.
Figure 7: Spatial distribution of the Mean Absolute Errors in the monthly profiles of the proposed Principal Component
Analysis method (left) and the traditional method (right)
Acknowledgements
The financial support of Tecnológico de Monterrey through funding of the research chair for wind energy
(contract number CAT158) as well as the support from the Mexican Council for Science and Technology
(CONACyT) is gratefully acknowledged. A late stage of this project also benefitted from financial aid from
the Marie Curie training program through the WAUDIT project at CENER.
References
[1] Larsén XG, Mann J, Berg J, Göttel H, Jacob D. Wind climate from the regional climate model REMO.
Wind Energy. 2010; 13:279–96.
[2] Al-Yahyai S, Charabi Y, Gastli A. Review of the use of Numerical Weather Prediction (NWP) Models
for wind energy assessment. Renew. Sust. Energ. Rev. 2010;14(9):3192–8.
[3] Gastón M, Pascal E, Frías L, Martí I, Irigoyen U, Cantero E, et al. Wind resources map of Spain at
mesoscale. Methodology and validation on. EWEC conference proceedings. Brussels, Belgium.
2008.
[4] Hahmann AN, Rostkier-Edelstein D, Warner TT, Vandenberghe F, Liu Y, Babarsky R, et al. A
Reanalysis System for the Generation of Mesoscale Climatographies. J. Appl. Meteorol. 2010;
49(5):954–72.
[5] Skamarock WC, Klemp JB, Gill DO, Barker DM, Duda MG, Wang W, et al. A Description of the
Advanced Research WRF Version 3. Atmospheric Research. Boulder, CO; 2008.
[6] Kallos G, Nickovic S, Papadopoulos A, Jovic D, Kakaliagou O, Misirlis N, et al. The regional weather
forecasting system SKIRON: an overview. Proceedings of the International Symposium on Regional
Weather Prediction on Parallel Computer Environments 1997. pp. 109–122. Athens,
[7] Zorita E, von Storch H. The Analog Method as a Simple Statistical Downscaling Technique:
Comparison with More Complicated Methods. J. Climate. 1999; 12(8):2474–89.
[8] Wilby RL, Wigley TML. Downscaling general circulation model output: a review of methods and
limitations. Prog. Phys. Geog. 1997; 21(4):530–48
[9] Wilby RL, Charles SP, Zorita E, Timbal B, Whetton P, Mearns LO. Guidelines for Use of Climate
Scenarios Developed from Statistical Downscaling Methods. Supporting material of the IPCC. 2004; p
1–27.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 27/167
Book of Proceedings
[10] Pryor SC. Empirical downscaling of wind speed probability distributions. J. Geophys. Res. 2005; 110
p. D19109.1-D19109.12
[11] Wippermann F, Gross G. On the construction of orographically influenced wind roses for given
distributions of the large-scale wind. Beitr. Phys. Atmos. 1981; 54, 492-501.
[12] Heimann D. Estimation of regional surface layer wind field characteristics using a three-layer
mesoscale model. Beitr. Phys. Atmosph. 1986;59(4):518–37.
[13] Frey-Buness F, Heimann D, Sausen R. A Statistical-Dynamical Downscaling Procedure for Global
Climate Simulations. Theor. Appl. Climatol. 1995;50:117–31.
[14] Fuentes U, Heimann D. Verification of statistical-dynamical downscaling in the Alpine region. Climate
Res. 1996;7:151–68.
[15] Mengelkamp H. Statistical-dynamical downscaling of wind climatologies. J. Wind Eng. Ind. Aerodyn.
1997;67-68:449–57.
[16] Bergström H. Wind Statistics Mapping Using a Higher-order Closure Meso-scale Model. EWEC
conference proceedings 2007.
[17] Mengelkamp H-T. Wind Climate Simulation over Complex Terrain and Wind Turbine Energy Output
Estimation. Theor. Appl. Climatol. 1999 Aug 24;63(3-4):129–39.
[18] Frank HP, Landberg L. Modeling the wind climate of Ireland. Bound. Lay. Meteorol. 1997;359–78.
[19] Frank HP, Rathmann O, Mortensen NG, Landberg L. The Numerical Wind Atlas - the KAMM / W AsP
Method. Riso report ris-r-1252, 2001
[20] Fuentes U, Heimann D. An Improved Statistical-Dynamical Downscaling Scheme and its Application
to the Alpine Precipitation Climatology. Theor. Appl. Climatol. 2000; 19;65(3-4):119–35.
[21] Hagemann K. Mesoscale Wind Atlas of South Africa. PhD thesis. University of Cape Town; 2008.
[22] Hahmann AN, Vincent CL, Division WE. Applications of Self Organizing Maps in Wind Energy
Meteorology. EMS Annual Meeting. Zurich, Switzerland; 2010.
[23] Rife DL, Vanvyve E, Pinto JO, Monaghan AJ, Davis CA, Poulos GS. Selecting Representative Days
for More Efficient Dynamical Climate Downscaling: Application to Wind Energy. J. Appl. Meteor. &
Climatol. 2013;52:47–63.
[24] Tammelin B, Vihma T, Atlaskin E, Badger J, Fortelius C, Gregow H, et al. Production of the Finnish
Wind Atlas. Wind Energy. 2013;16(2011):19–35.
[25] Martinez Y, Yu W, Lin H. A New Statistical–Dynamical Downscaling Procedure Based on EOF
Analysis for Regional Time Series Generation. J. Appl. Meteor. & Climatol. 2013 Apr;52(4):935–52.
[26] Otero-Casal C, Miguez-Macho G, López P, Canoura F. Calculating a high-resolution Wind Resource
Map with WRF using SOM clustering techniques. 12th WRF Users’ Workshop. 2011.
[27] Kanamitsu M, Ebisuzaki W, Woollen J, Yang S-K, Hnilo JJ, Fiorino M, et al. NCEP–DOE AMIP-II
Reanalysis (R-2). B. Am. Meteorol. Soc. 2002 ;83(11):1631–43.
[28] Pytharoulis I, Katsafados P, Kallos G. The Weather Forecasting System SKIRON - Model
configuration and setup. Athens, Greece; 2005.
[29] Conway D, Jones PD. The use of weather types and air flow indices for GCM downscaling. J. Hydrol.
1998 Dec;212-213:348–61.
[30] Davy RJ, Woods MJ, Russell CJ, Coppin P a. Statistical Downscaling of Wind Variability from
Meteorological Fields. Bound. Lay. Meteorol. 2010;135(1):161–75.
[31] Liu Y, Weisberg R., He R. Sea Surface Temperature Patterns on the West Florida Shelf Using
Growing Hierarchical Self-Organizing Maps. J. Atmos. Oceanic Technol. 2006;23:325–38.
[32] Hurrell JW, Van Loon H. Decadal variations in climate associated with the north atlantic oscillation.
Climatic Change. 1997;36:301–26.
[33] Barnston AG and Livezey RE. Classification, Seasonality and Persistence of Low- Frequency
Atmospheric
Circulation
Patterns.
Mon.
Wea.
Rev.
1987;
115,
1083–1126
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 28/167
Book of Proceedings
Sand erosion technique applied to wind resource assessment
Boris Conana,b,5, Jeroen van Beecka, Sandrine Aubrunb
a
von Karman Institute for Fluid Dynamics, 1640 Rhode-Saint-Gènese, Belgium
b
PRISME Laboratory, Orléans University, France
From J. Wind Eng. Ind. Aerodyn. 104–106: 322–329 (2012); doi.org/10.1016/j.jweia.2012.03.017
Abstract
One of the major challenges of the wind energy sector is to accurately predict the wind potential. This
task is especially difficult in mountainous terrains where the topography can imply complex relief-induced
flows. Wind tunnel testing is one of the possibilities to simulate and predict the wind for wind turbine
micro-siting. Most advanced quantitative measurement techniques can be used in the wind tunnel,
however, measuring the whole terrain to find the highest wind potential zones is very time-consuming.
This paper proposes to use a very simple, quick and cheap technique to detect and evaluate the high
wind speed areas over an entire model. Commonly used for pedestrian wind comfort assessment, the
sand erosion technique is here applied to wind resource assessment. The technique can provide valuable
qualitative informations but can also give an order of magnitude of the local speed-up. It is first applied to
a backward facing step flow and then on a mountainous terrain. An amplification factor and the fractional
speed-up ratio (FSR) can be calculated over the entire mountain. For high speed positions results
extracted from sand erosion appears to be comparable the one calculated by particle image velocimetry.
The technique is repeatable, able to perform a detection of the high speed area, and capable of giving an
estimation of the amplitude of the wind. The technique allows to restrict the use of quantitative
measurements to the most interesting areas.
Keywords: Sand erosion technique; PIV; Wind energy; Wind resource assessment; Wind tunnel test
1. Introduction
In the fast development of wind energy, wind-farms tend to be more and more located in complex
terrains. On a cliff, a hill or a mountain, the wind speed-up created at the top of the topography is an
advantage for the wind farm productivity. However, the complexity of the terrain increases the difficulty of
determining the wind characteristics (direction, mean speed, turbulence); therefore the prediction of the
wind resource and the profitability of a wind farm becomes more challenging.
The wind resource assessment in complex terrain and the determination of the local effect of a
topography on wind characteristics can be performed by physical modelling in the wind tunnel. To
measure the flow in the wind tunnel, measurement techniques like hot-wire anemometry (HWA) and
Particle Image Velocimetry (PIV) are used. The state-of-art PIV technique allows the determination of the
mean wind speed and the turbulence level of the three velocity components in a volume (usually reduced
to a few centimetres). However, for cost effective reasons, two velocity components in a two dimensional
plane is the most common technique. Despite the very good spatial resolution, the frequency resolution of
PIV is often a limitation for measuring the turbulence spectra (>10 kHz needed) that is an order of
magnitude above the classical PIV possibilities. The hot-wire technique can complement the PIV
measurements by a punctual measurement of the three velocity components with a very high frequency
5
Corresponding author: Boris Conan
Environmental & Applied Fluid Dynamics Department, von Karman Institute for Fluid Dynamics, 1640 Rhode-Saint-Gènese,
Belgium, Tel.: +32 2 359 97 60, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 29/167
Book of Proceedings
resolution (>10 kHz). This technique is used for computing the spectral density distribution and the
turbulence length scales. Those two very accurate techniques can fully characterize the wind profile
(Conan et al., 2011a).
However, the installation and the use of these techniques require time, precision and a lot of precautions.
A first estimation of the location of the speed-up areas is a very valuable information to save time by
reducing the measurement zone. This paper presents a simple tool for a global approach of the wind over
complex terrains: the sand erosion technique.
For the assessment of pedestrian-level wind in urban areas, where computational techniques remain very
difficult to use, erosion tests are ordinary carried out in a wind tunnel to predict the wind comfort: see
Stathopoulos (2006), Plate (1999), Dezsö (2006) and van Beeck et al. (2009). Simple, quick and cheap,
erosion techniques are commonly performed for studying urban flows. Based on this experience, the
sand erosion technique is here tested in another application: wind resource assessment in complex
terrain.
The objective of this work is to evaluate the possibility of using the sand erosion technique as an initial
qualitative vision of the potential wind park siting areas on a large domain. The technique allows to focus
further investigations with more expensive quantitative measurement techniques. The study presents the
technique, proposes a methodology to use it, assesses the reliability of the results, discusses its
limitations and presents visualizations and quantitative measurements compared with proven techniques.
Tests are performed first on a backward-facing step (BFS) and then on a mountainous terrain, the Alaiz
mountain (Spain).
2. The sand erosion technique
2.1 Principle
The sand erosion technique is commonly used for pedestrian-level wind assessment in urban area. It is
based on the erosion of sand placed on a model. Different authors developed methodologies to use the
technique for wind comfort assessment: Viegas and Borges (1986), Williams (1986), Livesey et al.
(1992), Blocken and Carmeliet (2004); and Dezsö (2006).
In practice, the model is placed in a wind tunnel and covered with a thin layer of 1 mm to 1.5 mm of
sieved sand. The surface of the model is usually painted black and the sand used is either white or
coloured to contrast with the background. The sand placed on a surface has the property to erode at a
given friction-velocity called here threshold friction-velocity, U*th. The velocity of the wind tunnel is then
increased step by step and a picture is taken after 1 min of exposition to a given free-stream velocity. At
each velocity step, areas on the model are more and more eroded and contrast with the rest of the model
still covered by the sand. Revealed sand contours are iso-friction-velocity contours and the frictionvelocity is close to the threshold friction-velocity of the sand (U*th). The relationship between the sand
erosion patterns and the friction-velocity is still not completely understood, especially in detached zones.
Dezsö-Weidinger et al. (2003) showed that the eroded area contours are linked to the mean frictionvelocity and the RMS. In regions with high turbulence level, the sand erodes for a lower mean frictionvelocity due to large fluctuations around the mean that are higher than the threshold friction-velocity of the
sand (U*th). Another limitation of the technique is the easier entrainment of particles due to up-wind
particle impacts, this is called "down-wind erosion" and discussed in Section 4.2.4.
Sand contours are evolving at each velocity step and give a visualization of the locations of the high
velocity zones. Different techniques (Section 2.3) allow to compute an amplification factor map or to
retrieve the velocity at a higher altitude. Figure 1 presents an example of sand erosion patterns obtained
on a large model of a mountain (details are given in Section 4).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 30/167
Book of Proceedings
Figure 1: Example of a model at the beginning of the test (left) after 1 min at 6 m/s (middle) and after 1 min at 7 m/s (right).
2.2 Sand characteristics
The sand used is sieved and measured to be statistically mono-dispersed around a mean diameter of 400
µm. A friction velocity calibration is performed on a flat-smooth plate. To do so, a layer of sand is placed
on a flat plate and the wind tunnel speed is raised step by step until the sand erodes, at that moment, the
friction-velocity exceeds the threshold friction-velocity of the sand. For that speed, the velocity profile is
measured by hot-wire anemometry (HWA), Particle Image Velocimetry (PIV) or Laser Doppler
Velocimetry (LDV). The friction-velocity is deduced from Bradshaw's method (Bradshaw and Huang,
1995). For the sand employed in this study, U*th = 0.27 m/s. Because the threshold friction-velocity is a
property of the sand, it has to be calibrated only once and then it can be used in different set-up with
different configurations.
2.3 Methodology
To extract useful information from the sand erosion test, a specific methodology is followed. The model is
first covered partially or completely with 1 mm sand layer as described in Section 2.1, the free-stream
velocity of the wind tunnel, Ui, is then increased by steps of 0.5 m/s and an image is taken after 1 min of
exposure to each velocity step. One minute is long enough so that sand contours are stable and do not
depend much on the initial sand thickness non-uniformities, and short enough so that extreme gusts do
not play an important role (see van Beeck et al., 2009; Dezs+o, 2006). At each step, at the sand contour,
the friction-velocity is U*th. The free-stream velocity of every step (Ui) can be compared to the free-stream
velocity (at the same height) for which the sand flies on a flat surface with an empty test section: Uref. This
allows to define an amplification factor, A, giving the speed-up or the speed-down due to the model (see
Blocken and Carmeliet, 2004):
A=
U ref
Ui
(1)
Where the sand erodes for a free-stream velocity lower than the reference velocity (Ui<Uref), the model
creates a local speed-up (A>1). At the contrary, if locations are not eroded for Ui>Uref, it means that those
locations are speed-down zones (A<1). Thanks to the different velocity steps realized and an automatic
detection of contours, a map of amplification factor can be drawn.
The well known Fractional Speed-up Ratio (FSR) can also be computed. In this case, (Ui) is varying at
each step and Uref is a constant:
FSR =
U ref − Ui
= A −1
Ui
(2)
With this method, there is no need to know the threshold friction-velocity of the sand, however Eqs. (1)
and (2) are valid under some major hypothesis (Blocken and Carmeliet, 2004): the flow is Reynolds
number independent in the range of velocity used, the flow is fully developed, the flow near the wall
follows the logarithmic law profile and the sand erosion occurs always at the same ground-level wind
speed.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 31/167
Book of Proceedings
Knowing the threshold friction-velocity allows to compute the velocity at a height, z, thanks to the
universal log-law of the wall for turbulent flow over smooth wall (Bradshaw and Huang, 1995)

 z ⋅ U* th  
U (z ) = U* th ⋅  5 + 2.5ln 

 ν 

(3)
with U(z): velocity (m/s) at altitude z (m) and ν: kinetic viscosity of the air (m2/s). The applicability of Eq.
(3) is presented in Section 3. The methods presented allow to extract quantitative data like amplification
factors or FSR.
2.4 Contour detection method
All images taken at the different velocity steps are processed with an in-house MatLab code detecting the
sand contours. In the wind tunnel, the contrast is increased by using white sand on black-painted model.
The code transforms the image in black and white and then in binary. The contours are then smoothed
and finally detected (Fig. 2). A full amplification factor or FSR map can be extracted.
Figure 2: Post-processing steps: raw image (left), binary image (middle) and contours detection (right).
3. Validation test: the backward facing step
For validation purposes, the technique is applied to a wellknown case study, the Backward Facing Step
(BFS). The aim is to perform a sand erosion test as described in Section 2.3 and to compare the results
with quantitative measurements.
3.1. Experimental set-up and quantitative measurements
The experiment is conducted in a blowing type wind tunnel able to provide 20m/s with 0.3 % free-stream
turbulence intensity. The test section (0.2 x 0.2 m2) is equipped with a 1000mm long wooden flat plate
with a H = 20 mm height backward-facing step. A sand paper strip is placed at the start of the plate to
trigger the development of a turbulent boundary layer (Fig. 3). The Reynolds number based on the step
height and the wind tunnel free-stream velocity (U∞ = 17.1m/s) is: ReH = (U∞·H)/ν = 21,800.
Figure 3: Wind tunnel set-up for the backward-facing step test case (Dezsö, 2006).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 32/167
Book of Proceedings
The flow downstream of the step is measured using particle image velocimetry (PIV) and a set of 500
images is used for computing the time-averaged flow field. The statistical error on velocity in the freestream is assessed to 2% at 98% confidence level. The single velocity measurement error is
approximately 70.25 m/s.
Presented in numerous literature papers like (Scarano et al., 1999) or (Le et al., 1997), the time-averaged
flow field, shown in Fig. 4, is characterised by a clockwise recirculation area extending up to 5.5 H
downstream of the step and a counter-clockwise corner vortex at the foot of the step.
Figure 4: Time averaged velocity magnitude and velocity streamlines on the BFS at Re = 21,800 (Dezsö, 2006).
Figure 5: Comparison of the FSR computed from PIV measurements and from sand erosion tests (top). Top view of the
sand erosion pattern after 1 min at 17 m/s (bottom). (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 33/167
Book of Proceedings
3.2 Sand erosion tests and comparison of the results
The downstream part of the step is covered with a thin layer of sand (Fig. 3). The amplification factor (A)
and the FSR (Eqs. (1) and (2)) are computed at five free-stream velocities: 15.3 m/s, 16 m/s, 16.5 m/s, 17
m/s, and 17.6 m/s.
To be comparable with sand erosion, the mean velocity of the PIV data is calculated as U = ( u ) . Fig. 5
presents the comparison of the FSR extracted fromsand erosion (dots) with the PIV results (full lines).
Two curves are plotted from the PIV results: the mean velocity, U (in red), and the mean velocity plus the
RMS, U + Urms (in blue).
The sand erosion result has the same trend and falls between the two curves extracted from PIV. For low
turbulence (x/H < 3), the sand gives rather good agreement (2%) with quantitative measurement.
However, at the re-attachment zone (4 < x/H <6), the sand erosion gives values closer to U + Urms.
Sand erosion contours are thus overestimating the mean velocity in regions with high turbulence intensity.
This is conservative for wind comfort studies because uncomfortable zones are never missed. However,
for wind energy assessment this is less favorable because a soon erosion can be due to a high mean
velocity or a high turbulence level (see Section 2.1). Consequently, for application of the sand erosion
technique to the wind energy sector, the high speed locations are never missed but the level of
turbulence has to be assessed by other means in order to establish confidence in the mean velocity
prediction from sand contours.
4. Aplication to complex terrain
In this section, experiments are performed on a model of a mountainous terrain situated next to
Pamplona in the North of Spain, the Alaiz mountain. Wind farms already exist on site and field
measurements are currently being performed. CFD computations by Cabezón et al. (2007) and MunozEsparza et al. (2011) as well as wind tunnel experiments (Conan et al., 2011a,b) have already been
carried out concerning this site.
4.1 Description of the experiment
4.1.1 The Alaiz mountain
The mountain is 1130 m high and is stretching over 10 km in the W–E direction and over 8 km in the N–S
direction (Fig. 6). The configuration tested in the wind tunnel is the dominant wind direction: North.
Upstream the mountain, to the North, a 200 m high ridge is facing the incoming wind (X = 0.75 m in Fig.
6) and the wind tunnel mock-up is designed to include it because it is expected to affect the incoming
flow. The area modelled is 16 km x 15 km. Giving the test section constraints, the scaling factor is 5300.
The mock-up is the one used by Conan et al. (2011a), which was directly drilled in Necurons from the 3D
topographic file with 1/10 mm precision. The finishing is of the order of 10 mm. As a first approach, the
roughness of the terrain is not modelled.
A line following the wind direction is defined to perform quantitative measurements (Fig. 6). Two main
crests can be defined on the mountain: a first one, the main crest, is at the position of P4 and a second
one just before the position P6.
4.1.2 Atmospheric boundary layer modelling
Tests are performed in the VKI-L1 boundary layer wind tunnel. The test section is 15 m long, 3 m wide
and 2 m high. This length allows the development of a neutral atmospheric boundary layer generated
thanks to a grid and a step at the entrance of the test section, and roughness elements spread over 12 m
on the floor. The boundary layer modelled represents both the velocity and the turbulence profile of a
moderately rough to rough terrain with z0 = 2.2 m at real scale (see Conan et al., 2011a; VDI-guidelines,
2000; Eurocode, 2004).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 34/167
Book of Proceedings
Figure 6: Alaiz model: top view and profile of the terrain following the measurement line. Red area: sand erosion zone.
Points R1 and R2 are reference positions and points P1 to P7 are measurement position. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
4.1.3. Quantitative measurements: particle image velocimetry and hot-wire anemometry
A quantitative assessment of the wind over the mountain is performed with particle image velocimetry
(PIV) (Raffel et al., 1998) and hot-wire anemometry (HWA) (Bruun, 2002). Measurements are performed
on the vertical plane presented in Fig. 6, between measurement points R1 and P7.
PIV is performed between R2 and P7 with a free-stream velocity at 1 m above the floor of 14 m/s. The
mountain height based Reynolds number is ReM = 105,000. The average velocity vector field is computed
in the vertical plane (U, W) over 500 images taken at a sampling frequency of 3 Hz. The statistical
uncertainty associated is assessed to 1.5% at 95% confidence level for the mean speed and 8% for the
RMS at the same confidence level.
Additionally, hot-wire measurements are performed at each of the positions at a sampling frequency of 5
kHz. The uncertainty is assessed to 1% for mean values and 5.7% from RMS at 95% confidence level.
PIV and HWA profiles are in very good agreement, within less than 2%.
For the assessment of the wind characteristics, the combination of PIV and hot-wire anemometry is very
powerful: the PIV gives the mean velocity profile and the turbulence intensity on a 2D field with a very
high spatial resolution and the hot-wire provides a punctual time series leading to turbulence length scale
and turbulent spectra.
The fractional speed-up ratio (FSR) is computed at 90 m real scale (common hub height) as
FSR(90m) =
U (90m) − Uref (90m)
Uref (90m)
(4)
With Uref the reference velocity at 90 m for the position R1.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 35/167
Book of Proceedings
This parameter gives information on the ratio of change of the wind at a given location with reference to
the inlet condition. Results in Fig. 11 (red curve for PIV and red dots for HWA) show a speed-down before
the hill and two major areas of high wind speed situated next to positions P4 and P6.
4.2 Sand erosion on a mountainous terrain
4.2.1 Sand erosion tests
The sand erosion technique is tested on the Alaiz mountain in the area of interest indicated with a red
square on Fig. 6. The reference velocity used for the FSR (Uref(90 m) in Eq. (4) is calculated at 90 m real
scale in the wind tunnel from the threshold friction-velocity and the law of the wall (Eq. (3)): Uref(90 m) =
5.21 m/s. To verify it, the velocity at 90 m (real scale) is measured in the wind tunnel when the sand
erodes without the model. This measurement gives Uref(90 m) = 5.28 m/s. Both estimations of the
reference velocity agree. The technique is verified for a log-law profile.
In this study, 13 velocity steps (Uref(90 m)) are performed in the range 3–7 m/s. Table 1 lists the velocities
used.
Table 1: Velocities, U(90 m), tested in the wind tunnel.
Velocity at 90 m (m/s)
3.35
3.55
3.8
4.04
4.29
4.79
5.04
5.28
5.53
5.78
6.03
6.28
4.2.2. Visualisation
Before trying to extract any quantitative values, the sand erosion is a very valuable technique for the
visualisation of the high wind speed areas, the observation of the erosion contours illustrates the
repartition of the high and low speed areas. Fig. 7 presents a set of pictures taken at different increasing
velocities.
The sand is cleared step by step from the mock-up surface. The technique allows a very quick
visualisation of the highspeed areas that are eroded first. The two crests are quickly appearing to be the
highest wind speed positions. They are clearly good candidates for deeper analysis to state their
suitability for wind turbine siting.
Figure 7: Example of an evolution of the erosion pattern with the free-stream velocity (red area on Fig. 6). (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
4.2.3. Repeatability study
To assess the repeatability of the method, a set of five runs is performed in the same conditions and with
the same sand spread over the same area of the model. Tests are conducted at different days and the
sand is spread on the model by different people. Fig. 8 presents the superposition of the sand contours
for the five independent runs at two velocities together with the elevation map. Generally speaking,
contours are very similar, at low speed, the up-wind contour of the sand area is very well reproducible but
more scatter is observed at the down-wind contour. In general, low friction-velocity gradients lead to
higher variability of the sand contours and introduce more uncertainty on the position of the sand
contours, but after a certain velocity, sand contours match. A quantification of the repeatability is
performed by comparing the eroded surfaces for the different runs. At 3.8 m/s, there is around 7% of
scatter between independent tests, this value decreases rapidly with increasing speed, it goes lower than
3%, at 5 m/s. The repeatability study corroborates the observations of (Dezsö-Weidinger et al., 2003) that
the repeatability of the technique is good.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 36/167
Book of Proceedings
Figure 8: Superposition of the sand contours of five independent runs performed at two velocity steps 4.29 m/s (left) with
6.5% scatter and 4.79 m/s (right) with 5.5% scatter.
4.2.4. ‘‘Down-wind erosion’’ study
This phenomenon, described by Dezs+o-Weidinger et al. (2003) and Williams (1986), is observed when
an important area is covered with sand, in this case, the downstream sand is more likely to erode. The
two main explanations brought forward are: an easier erosion due to upwind sand impacts and the
increase of the wall turbulence due to the surface roughness of the sand.
To assess its importance in this particular case, tests are performed by covering partially the area of
interest, first only the first ridge (area between the two red lines in Fig. 9) and in another test only the
second ridge (area between the two green lines in the same figure) are covered. Results are compared
with the fully covered test (black contours in Fig. 9).
Figure 9: Effect of the area covered at two velocities: 4.29 m/s (left) and 4.79 m/s (right). Only first ridge covered (red), only
second ridge covered (green) and all area covered (black). (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
From the observations, the erosion is generally weaker when the area is not completely covered. At 4.29
m/s, for the first ridge, the sand contours are at the lower limit (less erosion) of the 7% of repeatability
error. For the second ridge, the sand area before the ridge crest (down-wind contour of the sand area) is
much less eroded, however, the sand contour at the crest (up-wind sand contour) is the same as in the
reference case. At 4.79 m/s all sand contours falls in the repeatability error of 5%.
The presence of sand upstream of the area of interest implies a sooner erosion upstream the second
ridge crest, the data taken with local covering on the second ridge are the one used further in the paper to
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 37/167
Book of Proceedings
extract the velocity speed-up at that position. For the first ridge there is no difference. Generally speaking,
the upstream sand effect decreases rapidly with increasing speed.
4.3. Sand erosion compared with PIV and HWA
More than a visualisation, the methodology presented in Section 2.3 allows to go further and can give an
estimation of the over-speed on mountain tops. Fig. 10 is the FSR (Eq. (2)) map drawn on the elevation
contours of the Alaiz mountain. All velocity steps tested are here superimposed. Red areas are related to
high speeds and blue areas to low speeds. The map underlines the general speed-up created in a large
part of the top of the mountain. Low winds are limited to the recirculation area (after to the second ridge x
> 2.15 m) and to low elevation zones (x ≈ 2.08 m). At the first region cleaned, the second ridge, the FSR
is the highest with more than 55 %; the sand contour follows the crest line. The second region affected by
erosion is the crest line close to the position P4 (x ≈ 1.85 m). The FSR at this position is close to 50%.
Similarly to a recovery from a perturbation (even without separation), the FSR is the lowest right after
crest tops and gradually increases after it. Other places with low frictionvelocities are located in troughs of
the relief like next to positions x = 1.97 m and x = 2.09 m.
Figure 10: Amplification factor map on Alaiz mountain (Eq. (1)), the white dots on the image are reference positions
(masts). Axis is in meter at the wind tunnel scale. (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)
To compensate the downstream erosion effect detailed in Section 4.2.4, the processing of the images
with partial covering of the terrain is used for the calculation of the FSR at the top of the second ridge.
The FSR extracted from sand erosion is compared with the PIV data in Fig. 11 and a scatter plot is
presented. The scatter plot generally shows important discrepancies between the PIV results and the
quantitative sand erosion data (R2 = 0.04). However, the two peaks are clearly appearing at a very similar
position and with a comparable value, around 50%. A correlation coefficient of R2 = 0.81 is determined in
the region of the two ridges, the highspeed regions detected by sand erosion are confirmed by the
comparison with quantitative data. Discrepancies appear mainly on the downwind slopes of the two
ridges, speed-down calculated with the sand erosion technique is overestimated.
Unlike hot-wire and PIV, the sand erosion is a near wall evaluation of the velocity that is extrapolated
upwards with a log-law. This assumption may not be fulfilled down-wind the ridges and can explain the
discrepancies. If the near wall turbulence is high, the sand will erode earlier and that will lead to an over
estimation of the speed at a higher altitude. Additionally, the technique is omnidirectional. These
differences can explain the higher level of details given by the erosion technique in Fig. 11 and the bad
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 38/167
Book of Proceedings
correlation with PIV measurements. As PIV results are here given at 90m above the surface, near-ground
effects are smoothed.
The FSR based on the sand erosion method gives coherent results with quantitative measurements in
high speed-up areas, the high speed zones are detected very easily and the FSR is of the same order of
magnitude as a quantitative measurement. However, the way the technique is here used does not allows
a right determination of the wind speed in the downwind part of the ridge. For wind resource assessment,
only high wind speed areas are of interest.
Figure 11: (top) Comparison between the fractional speed-up ratio calculated by PIV and extracted from the erosion
technique. The FSR is plotted over the line described in Fig. 6. (bottom). Scatter plot between the PIV results and the sand
erosion results. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 39/167
Book of Proceedings
5. Conclusions
The sand erosion technique is an omnidirectional tool able to detect high speed zones on a large area. It
is used to evaluate high wind speed spots on an unknown terrain. The implementation of the technique is
very simple, fast and cheap. The image post-processing methodology is straight forward and does not
require advanced tools. The repeatability error of the technique is below 7% and decreases with speed.
This is very reasonable regarding the manner to spread the sand.
One of the main limitations comes from the fact that the sand erodes more in the downstream part. This
can be overcome by performing local sand erosion tests, i.e. putting sand only on hill tops.
Qualitative results are very interesting and high wind speed zones are detected very fast. When a terrain
is investigated for the first time, this is a great advantage that enables to select the areas where deeper
measurements have to be carried out.
For more qualitative data, the comparison with PIV and HWA measurements demonstrates that by a
simple calibration, the technique can give meaningful estimations (by 10%) of the overspeed expected on
a mountains top next to the surface as far as the log-law applies. Results obtained downstream the ridges
are however far from reality.
While evaluating an unknown complex terrain, the technique appears to be a very valuable tool to give
quickly a global approach of high wind speed locations. Included in a global methodology, the sand
erosion technique is a very interesting tool as a first approach. Detailed investigations with more delicate
and expensive measurement techniques, like PIV, can be performed at detected locations to get the wind
profiles and to assess with accuracy the suitability of placing wind turbines at these positions.
Acknowledgments
This study has been realised with the financial support of the European Commission within the WAUDIT
Initial Training Network (Wind resource assessment audit and standardisation: http://www.waudit-itn.eu),
a Marie-Curie action funded under FP7-People program.
The authors gratefully acknowledge the support of G. Dezsö for providing experimental data for Section
3.
References
[1] Blocken, B., Carmeliet, J., 2004. Pedestrian wind environment around buildings: literature review and
practical examples. Journal of Thermal Envelope and Building Science 28 (2), 107.
[2] Bradshaw, P., Huang, G., 1995. The law of the wall in turbulent flow. Proceedings of the Royal
Society of London A 451 (1941), 165.
[3] Bruun, H., 2002. Hot-Wire Anemometry. Oxford University Press.
[4] Cabezón, D., Sanz, J., van Beeck, J., 2007. Sensitivity analysis on turbulence models for the ABL in
complex terrain. In: Proceedings of the European Wind Energy Conference, May 2007, Milan.
[5] Conan, B., Buckingham, S., van Beeck, J., Aubrun, S., Sanz, J., 2011a. Feasibility of micro-siting in
mountainous terrain by wind tunnel physical modelling. In: Scientific Proceedings of the European
Wind Energy Conference, March 2011, Brussels. pp. 136–140.
[6] Conan, B., Carbajo, F., van Beeck, J., Aubrun, S., 2011b. Experimental parametric study of the
influence of an idealized upstream ridge on the flow characteristics over Alaiz mountain. in:
Proceedings of the PHYSMOD2011, August 2011, Hamburg, Germany. pp. 272–279.
[7] Dezsö, G., 2006. On assessment of wind comfort by sand erosion. Ph.D. Thesis. Technidche
Universiteit Eindhoven & von Karman Institute for Fluid Dynamics. ISBN: 2-930389-22-2.
[8] Dezsö-Weidinger, G., Massini, M., van Beeck, J., 2003. Pedestrian wind comfort—approached by
PIV and sand erosion. In: Proceedings of the PHYSMOD2003, Firenze.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 40/167
Book of Proceedings
[9] Eurocode, C., 2004. Eurocode 1, 2004. en 1991-1-42: Actions on Structures. Parts 1–4: Wind
Actions.
[10] Le, H., Moin, P., Kim, J., 1997. Direct numerical simulation of turbulent flow over a backward-facing
step. Journal of Fluid Mechanics 330 (1), 349–374.
[11] Livesey, F., Morrish, D., Mikitiuk, M., Isyumov, N., 1992. Enhanced scour tests to evaluate pedestrian
level winds. Journal of Wind Engineering and Industrial Aerodynamics 44 (1-3), 2265–2276.
[12] Muñoz-Esparza, D., Conan, B., Croonenborghs, E., Parente, A., van Beeck, J., Sanz, J., 2011.
Sensitivity to inlet conditions of wind resource assessment over complex terrain using three CFD
solvers and wind tunnel data. In: Scientific Proceedings of the European Wind Energy Conference,
March 2011, Brussels. pp. 200–204.
[13] Plate, E., 1999. Methods of investigating urban wind fields—physical models. Atmospheric
Environment 33 (24–25), 3981–3989.
[14] Raffel, M., Willert, C., Kompenhans, J., 1998. Particle Image Velocimetry: A Practical Guide.
Springer-Verlag.
[15] Scarano, F., Benocci, C., Riethmuller, M., 1999. Pattern recognition analysis of the turbulent flow past
a backward facing step. Physics of Fluids 11, 3808.
[16] Stathopoulos, T., 2006. Pedestrian level winds and outdoor human comfort. Journal of Wind
Engineering and Industrial Aerodynamics 94 (11), 769–780.
[17] van Beeck, J., Dezs+ o, G., Planquart, P., 2009. Microclimate assessment by sand erosion and Irwin
probes for atmospheric boundary layer wind tunnels. In: Proceedings of the PHYSMOD2009, RhodeSaint-Genése.
[18] VDI-Guidelines, 2000. Vdi Guidelines 3783/12—Physical Modelling of Flows and Dispersion
Processes in the Atmospheric Boundary Layer—Application of Wind Tunnels.
[19] Viegas, D., Borges, A., 1986. An erosion technique for the measurement of the shear stress field on a
flat plate. Journal of Physics E: Scientific Instruments 19, 625.
[20] Williams, J., 1986. Aeolian entrainment thresholds in a developing boundary layer. Ph.D. Thesis.
Department of Geography and Earth Science, Queen Mary College, University of London.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 41/167
Book of Proceedings
Characterization of a wind turbine model
Francesco Cuzzolaa,6, Sandrine Aubrunb, Bernd Leitla
a
b
University of Hamburg, Hamburg, Germany
PRISME Laboratory, Orléans University, France
Presented at The Science of Making Torque from Wind conference, Oldenburg, Germany, October 2012.
Abstract
A model wind turbine has been designed at the University of Hamburg within the scope of the FP7
fundend project WAUDIT. The purpose of the experiment described in this paper is to characterize the
performances of two rotors by means of measuring the thrust coefficient Ct. Ct is a similarity parameter for
the wake and is thought to be the most effective one. Its value has been directly measured using a force
balance and indirectly calculated from the velocity profiles measured three diameters downstream of the
rotor with hot wire anemometry. Results show that, in order to reproduce the wake behaviour, the
matching of the Ct, which is a quantitative achievement, has to be integrated with measurements such as
velocity profiles in the wake. In fact the velocity deficit illustrates the mechanism of transforming the axial
momentum into torque assuring qualitatively the proper reproduction of the wake. This latter information
assures that the achievement of a certain thrust force acting on the rotor is due to its performances in
transforming the axial momentum into torque and not an effect of other phenomena such as a stall at the
blades.
1. Introduction
A dimensional analysis of the wake of a wind turbine leads to the definition of four dimensionless
similarity parameters that have to be matched when an experimental simulation of the wake of a wind
turbine is intended to be carried out in conditions of similitude.
These four parameters are the Reynolds number Re, the tip speed ratio λ, the power coefficient Cp and
the thrust coefficient Ct. The latter is thought to be the most effective one [1].
At the geometric scale at which the next experiment will be carried out it is not possible to match Re and
achieve conditions of full similarity. Thus, conditions of partial similarity have to be investigated in order to
assess how to operate the wind turbine model appropriately.
The systematic variation of parameters such as λ, the tunnel wind speed U∞ and the pitch angle θ allows
the researchers to describe the dependencies of the Ct with respect to the operating conditions. Once U∞
and θ were set then λ was adjusted to the desired value and data have been collected.
In this paper we present the results for the three configurations described in Tab.1.
2. Experimental set-up
The experiment was carried out in the \Malavard" wind tunnel at the PRISME Laboratory of the University
of Orléans. In this study, the main test section (2m high, 2m wide and 5m long) of the wind tunnel is used
6
Corresponding author: Francesco Cuzzola
Meteorological Institute - University of Hamburg, Bundesstrasse 55, 20146 Hamburg, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 42/167
Book of Proceedings
to perform measurements in Homogeneous Isotropic turbulent flow by placing a turbulence grid at the
entrance of the test section. The turbulence grid is made of metallic square section bars of diameter
25mm and mesh size 100mm. The upstream mean velocity is U0 = 10 ms-1 and the turbulence intensity is
Iu0 = 3% (ratio between the standard deviation of streamwise velocity and its time average) at the rotor
location.
Table 1: Wind turbine configurations investigated
U ∞ (m s-1)
λ
Wind turbine status
10
[0.2 - 2]
active
2.5
7
active
10
2.5
passive
Aerodynamic loads are measured with a 6-axis balance located below the test section. The mast of the
model, 34 mm of diameter is used to link the balance to the model. The precision of the balance is
estimated to be 0.16 N for the drag component and 0.48 N for the lift component. The sampling frequency
is fixed to 500 Hz, and 25000 samples are acquired for each configuration. The measurement of the three
components of velocity was carried out using a Dantec triple hot-fiber probe, connected to the Dantec
StreamLine system. The probe was fixed on an automated traverse system.
The architecture of the model also allows rotors and/or blades to be exchanged. Therefore two different
blades were designed and manufactured, which are object of this investigation. The first blade, from now
on called Optimal Blade was designed and optimized for highest power coefficient Cp. The second blade,
from now on called Linear Chord blade, was designed in order to have a comparison element in
assessing the importance of the chord distribution for models of such small scale.
The blades share the same Jedelsky EJ85 airfoil, the length of 210 mm, the wet surface of 0.303 m2 and
the twist angle distribution which is a linear approximation of the optimum distribution. The only geometric
parameter which changes is indeed the chord distribution.
The geometry of the blade is de_ned by means of a self implemented BEM code which does not take into
account the tip-root losses [3]. This design procedure was previously used for designing a first blade
which was tested in an experiment carried out in the open section wind tunnel at the University of
Hamburg [2]. The choice of using the Jedelsky EJ85 airfoil was taken because of its performances at low
Reynolds number and its geometry which can be easily manufactured.
Figure 1: LC conguration and experimental set up.
Figure 2: Model wind turbine with OB rotor.
Fig.1 and Fig.2 show the experimental set-up and the model wind turbine with the two rotors.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 43/167
Book of Proceedings
The model wind turbine is equipped with a Faulhaber DC motor 3268 G024 BX4 which allows the
rotational velocity to be controlled using a power supply (active status). Disconnected from the power
supply the motor can also be used as a generator if the turbine is driven by the wind only (passive status).
Fig.3 shows the almost perfect linear dependency of the rotational speed with the increasing voltage
supplied.
Figure 3: RPM - Voltage curve for the model wind turbine.
3. Force balance measurements
3.1. Results at 10 ms-1 - active status
Fig.4 and Fig.5 show the variation of the thrust coefficient for LC and OB.
Figure 4: Effects of the pitch angle on the Ct - λ curve for
LC, U∞ = 10 ms-1, active status.
Figure 5: Effects of the pitch angle on the Ct - λ curve for OB,
U∞ = 10 ms-1, active status.
The OB keeps rotating up to θ = 70º while the LC rotor is unable to perform at pitch angles θ > 60º. This
is due to the outer part of the optimum blade which does not block the flow as much as the linear chord
blade.
The LC configuration shows a Ct which is remarkably higher than that of the OB one (30% to 50%) for a
given value of λ . Also, depending on the value of θ, the trend of the variation can be very different as can
be seen in Fig.6 and Fig.7.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 44/167
Book of Proceedings
Figure 6: C t - λ curve comparison at theta = 30º, U1 = 10 ms -1,
active status.
Figure 7: Ct - λ curve comparison at theta = 50º, U∞ = 10
ms -1, active status.
From these measurements it is clear that the chord distribution has a very important influence on the Ct
and on its dependency from the tip speed ratio. The graphs show furthermore that the maximum value of
the thrust coefficient is found at higher tip speed ratios when the pitch angle increases. At higher θ, the
trends suggest that the maximal value of Ct might be found in a range of λ higher than has been achieved
with the present DC motor-power supply system.
3.2. Results at 10 ms-1 - passive status
The disconnection of the power supply means that the rotational speed of the model depends on the wind
speed and on the electrical load applied to the DC motor/generator. In fact, while in case of a short circuit
the internal torque of the motor blocks the shaft and the turbine stops, the application of a resistance to
the output lowers the internal torque until, if an infinite resistance is applied, it tends to zero.
Fig.8 shows the Ct vs θ curves in this latter configuration. Both curves exhibit a change at θ = 40º where
the thrust coefficient starts increasing with a steeper trend.
Figure 8: Ct - θ curves, U∞ = 10 ms-1, passive status.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 45/167
Book of Proceedings
4. Velocity measurements
Horizontal velocity profiles have been measured at a distance of 708 mm downstream of the rotor plane,
which corresponds to three rotor diameters.
4.1. Results at 10 ms-1 - passive status
The graphs of Fig.9 and Fig.10 show the velocity deficit along the blades for the two rotors. The abscissa
is r = x/R where R is the radius of the rotor and x the radial distance. The velocity deficit increases with
increasing pitch angle.
Figure 9: Velocity deficit curves for LC. U∞ = 10 ms-1,
passive status.
Figure 10: Velocity deficit curves for OB. U∞ = 10 ms-1,
passive status.
Fig.11 and Fig.12 highlight the differences between the rotors. The wind speed recovers smoothly in the
OB configuration showing a trend which is consistent with previous experiment such as [5], [6] and field
experiment [7]. The LC configuration, which at θ = 50º delivers a remarkably high Ct, shows a different
wake behaviour, with the maximum of the velocity deficit occurring in the outer part of the blade.
Figure 11: Velocity deficit comparison at θ = 30º. U∞ = 10
ms -1, passive status.
WAUDIT
Contract#
238576
Figure 12: Velocity deficit comparison at θ = 50º. U∞ = 10
ms -1, passive status.
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 46/167
Book of Proceedings
Fig.12 and Fig.13 present the root mean square curves of the U component of the velocity. From this
graphs it is possible to interpret how the velocity deficit is created. The OB configuration has a fairly low
turbulence intensity, although it shows a spike corresponding to the elbow of the blade where, from
previous experiments [2], it is known that the vortex is shed.
Figure 12: Turbulence intensity for LC. U∞ = 10 ms-1, passive
status.
Figure 13: Turbulence intensity for OB. U∞ = 10 ms-1,
passive status.
The LC configuration instead shows an increase of the turbulence intensity in the outer part (r = [0:6 - 1])
of the blade. At θ = 50º this increase is much higher than in the case of OB, as can be seen in Fig.14.
Figure 14: Turbulence intensity comparison at θ = 50º. U∞ = 10 ms-1, passive status.
4.2. Results at 2.5 ms-1 - active status
In order to investigate the wake with λ = 7, a value of the tip speed ratio in use also for commercial wind
turbines, the wind speed was set to U∞ = 2:5 ms-1 and the rotational speed to Ω = 708 rpm. The result,
see Fig.15, shows that the turbine does not extract energy from the flow field but it is accelerating the
flow, the axial force acts in the opposite direction of the stream and the model is not behaving as a turbine
but as a propeller. Thus, for this model wind turbine, the matching of the similarity parameter λ is
inappropriate when conditions of similitude in the wake are intended to be achieved.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 47/167
Book of Proceedings
Figure 15: Velocity deficit curves for LC at θ = 30º and λ = 7. U∞ = 10 ms-1, passive status.
4.3. Calculation of the thrust coefficient
At the distance of three rotor diameters downstream of the wind turbine model, the V and W components
of the velocity are less than 5% of U (see Fig.16) and can be neglected in the calculation of the thrust
coefficient that, according to the blade element momentum theory [4], can be calculated using the
relation:
∞
T = 2π ∫ ρUmean (r ) (U ∞ − U mean ( r ) ) r dr
0
(1)
Figure 16: V and W dimension-less velocity components three diameters downstream of the LC rotor. U∞ = 10 ms-1, passive
status.
Tab.2 shows a comparison between the values of the thrust coefficient evaluated respectively from the
force measurements and from the velocity ones.
Summarizing, when the OB rotor is mounted the wind turbine model has a noticeably smaller thrust
coefficient but the velocity measurements show a behaviour consistent with previous literature and with
expectations. The opposite happens for the LC rotor. This suggests that the OB rotor is a better choice
when the wake of a wind turbine is intended to be reproduced in a wind tunnel.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 48/167
Book of Proceedings
Table 2: Thrust coefficient calculation - summary table
Rotor
LC
OB
Measurement type
Force
Velocity
Force
Velocity
Tip speed ratio
θ = 30º
1.6
θ = 30º
0.75
1.81
0.07
0.25
1.98
0.19
1.53
0.047
0.62
0.054
0.22
5. Conclusions
A study of the aerodynamics and of the performances of a model wind turbine model has been carried
out. In particular we tested two blade designs which differ with respect to the chord distribution only. The
similarity parameter Ct has been calculated both directly using force measurements and indirectly using
velocity profiles. This latter evaluation is carried out using the blade element momentum theory and
overestimates the thrust force, but the trends are consistent with the direct calculation. Also, the velocity
measurements allow the researchers to have a deeper insight in the understanding of how and where the
thrust force is created.
As expected, the chord distribution has a big influence in the performances and in the wake of the model.
In particular the LC blade delivers a higher thrust coefficient than the OB one whereas this latter
configuration only shows velocity profiles consistent with previous literature. Thus the main result of this
investigation is that, for a rotor model wind turbine, the matching of the thrust coefficient needs to be
integrated with velocity measurements that can assure the transformation of the axial momentum of the
flow field into torque, thus harvesting energy. Also, operating actively the turbine in order to match a
certain λ will obviously result in a change in the overall behaviour of the turbine which will add energy to
the flow.
Further development of this experiment focuses on the measurements of the power coefficient Cp
applying different resistors to the electric output of the model wind turbine.
References
[1] Neff D, Meroney R, McCarthy E and Davis E 1990 Journal of Wind Engineering and Industrial
Aerodynamics 36: 1405-1414
[2] Doerenkaemper M, Cuzzola F, Leitl B and Schatzmann M 2011 Physmod2011-International
Workshop on Physical Modeling of Flow and Dispersion Phenomena
[3] Cuzzola F, Doerenkaemper M, Leitl B and Schatzmann M 2011 Physmod2011-International
Workshop on Physical Modeling of Flow and Dispersion Phenomena
[4] Aubrun S, Devinant P and España G 2007 European Wind Energy Conference 1-8
[5] Chamorro L P and Porté-Agel F 2009 Boundary Layer Meteorology 129-149
[6] Sanderse B 2009 ECN-e-09-016
[7] Mann J 2010 Wind Energy 51-61
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 49/167
Book of Proceedings
A study of stability effects in forested terrain
Cian J. Desmonda,7, Simon Watsona
a
Loughborough University, Loughborough, United Kingdom
Presented at The Science of Making Torque from Wind conference, Oldenburg, Germany, October 2012. In press for Journal of
Physics.
Abstract
Data from four well instrumented met masts located in heavily forested European sites in different
locations and terrain types are examined. Seven stability metrics are applied to the data sets and a
novel method is used to identify the metric which most consistently identifies stability events of
importance for wind energy generation. It was found that the Obukhov length, as calculated by fast
response sonic anemometer, provides the most reliable results in these highly complex sites. It was
also found that non-neutral stabilities can be expected a significant portion of the time for wind speeds
of less than 10 m/s at the considered sites.
1. Introduction
Complex sites pose a significant challenge for the development of wind farms due to unpredictable
flow patterns and increased levels of turbulence. In this era of financial rigor it is important to be able
to understand these flows and their effect both on the energy capture and fatiguing of wind turbines.
Forestry is a particular form of complex terrain in which wind farms are now being increasingly
developed. To date, it has been common to use linearised flow models along with empirical and
experience led corrections when predicting wind characteristics in such areas. More recently, the use
of full computational fluid dynamics (CFD) solvers has become common. Whilst satisfactory results
have been achieved using these codes, simulations are predominately limited to neutral atmospheric
stabilities.
The evolution of stability effects within the atmospheric boundary layer is an intricate process which
depends on many factors such as wind shear, ambient turbulence levels, wind speeds, the Coriolis
Effect, potential temperature gradients and vertical heat fluxes. Certain surface features present in
complex terrain can severely alter some of these factors making it difficult to assess stability effects.
For example, local thermal stratification can be significantly affected by forests as they act as large
thermal masses which are slow to heat up and cool down as solar irradiance levels vary. Also,
depending on the density of the canopy, forests act as a buffer between the underlying soil and the
atmosphere which inhibits surface heat flux [1]. These factors result in characteristic thermal
stratifications within and above forests which vary significantly over the diurnal cycle. Generalised
profiles are displayed in Figure 1 which shows how mean potential temperature,
, and vertical heat
flux,
, vary throughout the canopy height. Values are normalised to values at the top of the
canopy, hc.
7
Corresponding author: Cian J. Desmond
CREST, School of Electronic and Electrical Engineering, Loughborough University, Holywell Park, Loughborough,
Leicestershire LE11 3TU, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 50/167
Book of Proceedings
Figure 1. Typical mean potential temperature and vertical heat flux profiles in and above a forest canopy for midday
and night time conditions. Reproduced from [2]
It has also been established that forests have a profound effect on both wind shear and ambient
turbulence levels above and downstream of canopies [3]. The aggregated effect of these factors is an
extremely complicated atmosphere in the vicinity of forests which makes the assessment of stability a
non trivial task. However, it is important that we arrive at a satisfactory quantifiable measure as
stability effects heavily influence how the wind interacts with the surface features present in such
complex terrain [4,5].
Data sets are examined from four forested sites in various terrains and geographical locations. Seven
stability metrics, which have been developed for various purposes, are investigated. The metric that
consistently differentiates between stable and unstable events of consequence to wind energy
generation is identified, critical values are established and the implications are briefly discussed.
2. Stability metrics
As mentioned above, the evolution of stability effects within the atmospheric boundary layer is a
complex process caused by many factors, an excellent introduction to which can be found in [6]. Here
we will limit the discussion to a brief introduction to the metrics that have been devised to quantify
stability effects.
2.1. Richardson Number
Denoted as, Ri, this is a non-dimensional parameter which relates the importance of buoyancy and
shear forces in creating turbulence. It requires measurements of both temperature and wind speed at
two heights and is given by the equation below [2].
Ri =
where,
g
g
(
dθ / dz
θ dU / dz
)
(1)
2
= 9.81 m/s ,
= Average potential temperature (K)
= rate of change of potential temperature with heights (K)
The Richardson number is thus positive for stable atmospheric stratification, negative for unstable and
zero for neutral. Near-neutral conditions can be assumed [9] for values of: -0.13 ≤ Ri ≤ 0.03.
2.2. Bulk Richardson Number
This modified version of Ri appears in a variety of forms, such as [7]:
Ri b =
WAUDIT
Contract#
238576
g
θ2
z2
dθ
(2)
2
U2
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 51/167
Book of Proceedings
The symbols have the same interpretation as above with the subscript “2” referring to the higher
measurement point. This metric again requires temperature measurements at two heights but only a
single wind speed.
2.2. Richardson Bulk Number in Forestry
This is a form of Rib which was devised specifically for investigating vertical mixing above forest due to
stability effects [8]. It will be referred to in this paper as Rib2 :
Ri b2 =
1 g dθ
2
U 2 θ 2 dz
(3)
Compared to Rib the main difference with this metric is that the higher measurements are taken above
the canopy and the lower temperature measurement within the canopy at a height of 1-2 m above the
ground.
2.3. Obukhov Length
This is another important metric for the assessment of stability in the atmospheric boundary layer and
is calculated from the equation [6]:
g
κ w 'θ '
1
=− θ
L
u*
where,
w 'θ '
u*
(4)
=
Von Kármán constant = 0.41
=
=
Average vertical heat flux (K m/s)
Friction velocity (m/s)
and,
(

u* =  u ' w '

where,
u', v', w'
u 'w '
) (
2
)
2
+ v 'w ' 

0.25
(5)
=
Wind velocity turbulent fluctuations in x , y , z (m/s)
=
Covariance of both parameters (m2/s2)
Conventionally, the Obukhov length has units of metres and can be taken as the depth of the
mechanically mixed portion of the boundary layer. 1/L will be positive in stable conditions and negative
in unstable. The following values are expected for near-neutral stability [9]: -0.07 ≤ 1/L ≤ 0.03
2.4. Kazanski-Monin parameter
This is a slight variation on the Obukhov Lengh which includes the effect of the Coriolis parameter. It
was devised for estimating the rate of plume growth for dispersion modelling and is given by the
following equation [9]:
µ=
κ u*
(6)
fL
where, f, is the Coriolis parameter: f = 2Ω sin , where is the latitude of the measurement location
and Ω is the rotational velocity of the Earth which is approximately: Ω = 7.2921·10-5 rad/s
2.5. Environmental Stability Parameter
This parameter focuses only on the potential temperature gradient [9]:
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 52/167
Book of Proceedings
S=
g dθ
θ dz
(7)
This parameter has units of s-2 and is proportional to the rate at which the generation of turbulence is
suppressed. Positive values are indicative of stable conditions; negative of unstable with values close
to zero expected for near-neutral conditions.
2.6. Standard Deviation of Direction
This measure was by proposed by the US Nuclear Regulation Commission [10] to assess atmospheric
stability for the purposes of guiding emergency responses in the event of an incident. A high value of
the standard deviation of wind direction turbulent fluctuations,
, indicates unstable conditions; low
values indicate stable conditions and near-neutral conditions are characterised by some intermediate
value. Specific values will be highly site and measurement height dependant. For the purposes of
comparison with the other metrics the inverse of this term,
, will be used
3. Site data
Data have been collected from four separate met masts located in forested terrain for the purposes of
this analysis.
3.1 Norunda
This mast is located in the middle of a heavily forested area in Sweden (60° 5’ N , 17° 28’ E) and was
established for the purposes of studying fluxes of CO2. The mast is located in flat terrain in the middle
of a dense coniferous forest with a mean canopy height of 30m. Data were available in 30 min
averages for the period 7/7/06 – 18/09/06 from a 100 m met mast recorded by a series of sonic
anemometers (Metek USA-1) and temperature sensors. Concurrent solar irradiance data were also
available from a pyranometer located at the top of the mast. For additional information on the
instrumentation please refer to [11].
It was possible to calculate values for all of the stability metrics discussed above at heights between
28 m – 36.9 m. Values of shear exponent α at 87m and turbulent kinetic energy (TKE) at 31.8 m were
calculated as the parameters of interest for the purpose of resource assessment using the equations
below:
U2
U1
α =
z
log 2
z1
log
TKE =
(
1 2
σ u + σ v2 + σ w2
2
(8)
)
(9)
refers to the standard deviation of wind speed turbulent fluctuations in the x, y, z
where
directions respectively.
3.2. Wetzstein
This mast is located in Germany (50° 27'N, 11° 27'E) and data were available for the period of
12/04/05 – 19/08/05 as 30 min averages. As it was part of the same project as the Norunda site it is
also located in the midst of a large coniferous forest. The mean canopy height is 20 m and the mast is
located on the top of a slight hill. A pair of sonic anemometers (Gill R3) was located at 24 m and 32 m
which again allowed most of the metrics to be calculated. However, data for
were not available.
No temperature data were available except for values of virtual acoustic temperature as calculated by
the sonics. It was found that these data were not of sufficient quality despite application of a correction
for the effects of humidity and pressure [12].
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 53/167
Book of Proceedings
Solar irradiance data were again available from measurements made above the canopy. Values for α
between 24 m and 32 m and TKE at 32 m were calculated as the metrics of interest for the purposes
of resource assessment.
3.3. Sirta
This site is located in the Palaiseau Ecole Polytechnique (48° 42' 50'' N , 2° 12' 39'' E). The mast is
located in a semi-urban area with a large mixed forest located to the north-east at a distance of c. 55
m which extends a distance c. 560 m along a bearing of 37° from the meteorological mast.
The instrumentation is comprised of sonic anemometers (Meter USA-1) and temperature sensors
located at 10 m and 30 m. Additional information on the instrumentation can be found in [13]. Data
were available between 04/05/07 – 18/05/09 as 10 min averages. These were filtered to examine
directions of between 25°- 47° in order that the effect of the canopy could be assessed.
All stability metrics were again calculated with the exception of
due to a lack of available data and
also Rib2 as it is not applicable outside of a canopy. For comparison purposes, α was calculated
between 10 m and 30 m and TKE was calculated at 30m. Concurrent irradiance data were not
available and so annual average data from the Meteonorm database were used [14].
3.4. Vaudeville-le-Haut
This mast is located in a wind farm in France (46° 26' 58''N , 05° 35' 02''E). There is an extensive
mixed forest with a mean canopy height of 30 m located to the east at a distance of c. 130 m. Two
operating turbines are located at a bearing of 69° and a distance of 234 m and 23° at a distance of
600 m from the mast. Data were available between 01/01/10 – 31/12/11 as 10 min averages from a
series of sonic anemometers (Metek USA-1), temperature sensors and wind vanes on a 100 m mast
[15].
, Ri and Rib have been calculated
Due to a lack of access to the full data sets, only values for S,
between heights of 40 m and 80 m. Values for α were calculated between 60 m and 80 m along with
values for turbulence intensity (TI) at 80m defined as:
TI =
σu
(10)
U
Unfortunately it was not possible to calculate values of TKE as per the other sites.
Analysis was carried out for two direction sectors. One in which the wind will have travelled through
the dense forest (“V- forest”), 240° - 260°, and the other over flatter terrain (“V-flat”) , 30° - 50°. Flow
in this second direction sector will have been perturbed by the two turbines mentioned above;
however, it will provide some comparison for the effect of the forestry. Solar irradiance data were
again provided for the same period from a pyranometer on site.
4. Methodology
The values of TKE or TI and α as outlined above were considered for each case. The spread in
observed values was considerable as can be seen from the scatter plots for Norunda in Figure 2.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 54/167
Book of Proceedings
Figure 2: Scatter plots of
and TKE/
2
against direction for Norunda.
There are many possible causes of such a spread in values particularly in forested terrain where the
drag which the canopy exerts on the flow changes seasonally. In order to identify α and TKE values
that would be indicative of neutral conditions the Pasquill stability classes were used.
In this method, classes are defined as: A- Unstable, D- Neutral, F-Unstable with C, E representing
intermediate states. Table 1 indicates when such events are expected for certain irradiance levels
and mean wind speed,
, measured at 10m above ground level [9].
Table 1: Pasquill’s stability categories
As we can see from Table 1 above, the prevailing stability conditions are sensitive to irradiance levels
at low wind speeds but this relationship is removed for higher wind speeds when neutral conditions
prevail. This trend was observed for Norunda as can be seen in Figure 3. Values for TKE have been
normalised to the square of the mean wind speed for comparison.
Figure 3: Scatter plots of TKE/
2
against irradiance for low and high wind speeds.
The reduced scatter of points at higher wind speeds, the trend of which is insensitive to irradiance
levels, suggests that a value of TKE from 0.19 to 0.25 would be expected in neutral conditions. Bars
indicating this range are included in Figure 3.
A similar collapse of data was found for both TKE and α and for all five cases. Values expected for
neutral events for each case are given in Table 2.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 55/167
Book of Proceedings
Table 2: Expected neutral values
There were some problems in identifying neutral conditions for α in the case of Sirta. This was due to
the fact that two separate mean neutral values exist depending on the season. This was taken to be a
consequence of increased foliage during the summer months in the mixed forest. Due to a lack of
measurements these points were discarded. This problem was not found in the other cases, a fact
likely due to less severe seasonal foliage variations.
A scatter plot of values for α versus TKE/
the Norunda case in Figure 4.
Figure 4: Scatter plot of TKE/
2
2
was then produced. An example of such a plot is given for
versus α for the Norunda case.
The included bars, which create nine segments, indicate the neutral ranges of α and TKE/ 2 as given
in Table 2. Segments located on the main diagonal are labelled A,D,F as per the Pasquill
classification.
It is taken that the three labelled segments represent stability effects of interest for the purposes of
resource assessment. The classification of points in the other segments is less certain and so they are
disregarded for the purposes of assessing the stability metrics. For the five cases investigated, on
average 60% of points were retained following this stage.
The next step was to indentify which stability metric was best able to differentiate between the Astable, D-neutral and F-unstable categories as identified by the method above. The scatter plots in
Figure 5 show the level of this differentiation for Norunda using two of the stability metrics.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 56/167
Book of Proceedings
Figure 5: Scatter plots of 1/L and Rib2 against wind speed for Norunda.
As can be seen from the scatter plots in Figure 5, differentiation of the stability classes is more
successfully achieved using 1/L rather than Rib2 for the Norunda case. However, in order to compare
all metrics for all cases in a non-subjective manner, it is necessary to devise a measure of the level of
differentiation. This was achieved by writing a short program which identifies the line of best fit to
segregate stable-neutral and neutral-unstable cases. The program incrementally adjusts the positions
of these dividing lines until a position which minimises categorisation error is achieved.
For example, in the Norunda case, these dividing lines were identified as 1/L = 0.00198 m-1 and
1/L= -0.00692 m-1. Thus, if a stable point’s value of 1/L is greater than 0.00198 m-1 it is correctly
categorised and is counted as a “hit”. Using this method, 89% of values were correctly categorised.
The corresponding figure for Rib2 was 70%. This indicates that 1/L is more successful at identify
stability events of interest for this particular case.
Results for all metrics and for all cases are presented in Table 3 which can be found in the appendix.
Data points for wind speeds of < 2 m/s have been excluded from this analysis as the ability to
differentiate stability class is poor at such low wind speeds and they are not relevant for wind energy
generation.
5. Discussion
The ability of each metric to differentiate between stability events of interest is discussed below.
5.1. Obukhov & Kazanski-Monin
The Obukhov length was the best performing metric for each case in which it could be calculated. The
threshold values for neutral events are relatively consistent for each case, however, quite different
from the expected values quoted in subsection 2.3.
The superior performance of this metric may be due in part to the fact all measurements are
performed by a single instrument, i.e. a sonic anemometer, which has a high level of accuracy. This
eliminates the inherent potential error in the other metrics which rely on multiple measurement
technologies to produce synchronous results.
The Kazanski-Monin parameter performs comparably which would suggest that the Coriolis Effect is
not an issue at the heights considered.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 57/167
Book of Proceedings
5.2. Richardson numbers
The Rib outperforms the Ri for each case. This is particularly interesting as the former is devised as a
crude approximation of the latter. This may be due to sensitivity of the Ri to the
term in the
denominator which produces unreliable results at low wind speeds or for low wind shear. As
mentioned above, there is also inherent error in relying on up to four instruments to provide high
quality synchronous data.
Furthermore, it may be difficult to establish a sensible gradient of mean wind speed or indeed
temperature in complex sites with turbulent flow. Comparing values for Ri and Rib in V-flat against Vforest would suggest that the metrics are more applicable in simpler terrain where the flow is less
turbulent and meaningful averages of U and θ are more likely to exist.
The results for Wetzstein highlight the need for accurate temperature measurements. Although this is
not an issue for modern instrumentation it may be a problem when converting recorded temperatures
to potential temperature values. This conversion depends on ambient humidity and pressure levels
which introduces dependence on more instrumentation and hence increases potential error. In this
paper we have used the dry adiabatic lapse rate of -0.0098 K/m for all conversions [6].
Results achieved using Rib2 are poor which is disappointing as, in [8], this metric proved successful in
identifying stability effects within five forest canopies in complex terrain. This disparity is likely due to
the fact that the cited research is in the area of CO 2 fluxes and so is focused on flow parameters and
effects of stability which may not be directly relevant to the wind energy industry.
5.3. Environmental Stability Parameter
This parameter produces results comparable to Ri and Rib. The quality of results is again reduced in
more complex sites as can be seen from comparing results from the two Vaudeville-le-Haut cases and
the flat Norunda against the hilly Wetzstein. This may be due to the lack of meaningful average values
of θ in more complex terrain as discussed above, or the increased importance of shearing effects in
such sites, which this metric ignores.
It is also important to note that depending on the stratification above the surface layer, the effect of the
local gradient in potential temperature may be insufficient to determine the full stability characteristics
of the atmospheric boundary layer. This would require measurements of temperature at greater
heights than are commonly achievable using a standard meteorological mast. This is explained by the
series of graphs on p. 170 of [6] to which the reader is referred.
5.4. Standard deviation of direction
This metric performs surprisingly well given the fact that it is by far the easiest to measure and relies
on the reliable and well established technology of the wind vane. However, it is clear that threshold
values of
will not be sufficient for the purposes of resource assessment as they are site, height
and direction sector dependant. Also, such data provide little information as to the structure of the
boundary layer for the purposes of simulation.
This metric may perhaps be best used alongside more comprehensive measurements, such as sonic
can be correlated to stability categories. This data could then be used in
anemometry, so that
the event of failure or redeployment of the other instrumentation
6. Implications
In order to gain an appreciation of the likely effect of stability on wind generation the most effective
metric for each case was applied to the complete data sets. Wind speeds were scaled up to a realistic
hub height of 100 m using the calculated α and the standard power law formula. Although this
equation relies on a number of assumptions and may not be reliable in such complex sites, the
derived values are more indicative of wind energy potential than those measured a few metres above
the canopy.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 58/167
Book of Proceedings
Data were binned into wind speeds and segmented by percentage occurrence of stability class as
shown in Figure 6. A graph showing the number of data points in each wind speed bin is overlaid for
information. Results for the Sirta are shown in Figure 6. Although it may be desirable for continuity to
display results for Norunda, this is not the best case to consider for the effects of stability on wind
energy potential due to the low wind speeds and the fact that it was found to have unusually stable
nights [16].
Figure 6: Occurrence of stability events for different wind speeds at the Sirta case. This graph was produced using
threshold values of 1/L given in Table 3.
On inspection of Figure 6, and the other graphs in the Appendix, it is clear that non neutral stability
events are common especially for wind speeds below 10 m/s.
It is also interesting to note that there is only marginal difference in the occurrence of non-neutral
events between V-flat and V-forest which may indicate that the presence of forestry has little effect on
stability. However it is important to note that although the V-flat direction sectors contain considerably
less forestry the entire site is characterised by pockets of trees. Also, flow in this direction sector is
likely be perturbed by the operational wind turbines mentioned previously. Unfortunately, unobstructed
direction sectors were not available for comparison at the other sites.
It is important to understand how these non-neutral events affect the wind resource. In [17] it was
noted that the effects of stability are specifically influential in the development of internal boundary
layers in areas of abrupt roughness, temperature and moisture changes. This would indicate that
stability effects will be particularly important when modelling the transition from forested to grassland
or other less complicated terrain.
It is desirable to quantify the effect that these non-neutral events have on the wind resource due to
altered wind shear, levels of turbulence and the persistence of wakes. However, this is difficult to
assess the without extensive modelling and so will not be considered here.
There are methods that can be used to approximate the effect of, for example, wind shear on power
generation, e.g. the equivalent wind speed for AEP method [18] and the cosine loss model for inflow
angle [19]. However, these methods are highly technology dependant as some turbines are now being
designed with the ability to adjust their operation to mitigate losses due to such effects [20]
7. Conclusions
It is clear that the Obukhov Length, L, as calculated by a fast response sonic anemometer, provides
the most consistent results in identifying non-neutral stability events in the highly complex sites
considered.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 59/167
Book of Proceedings
There is a significant level of differentiation that can be achieved between stable, neutral and unstable
events by simply analysing data that are measured as standard at potential wind farm sites, such as α
and TKE or TI. An analysis of such data was used in [21] to identify stability events in moderately
complex terrain and it would appear from the present paper that a similar approach is also applicable
in these highly complex sites.
Although simple metrics do provide information on the effects of stability events they do not provide
any information on the cause, such data will be required if realistic simulations are to be conducted. If
we are to successfully model non-neutral events in complex terrain we will require information on
thermal stratification, vertical heat flux and values for quantities such as U* in the surface layer. These
data will be required to select appropriate boundary conditions and accurately describe buoyancy
forces in computational simulations.
Given the performance of the Obukhov length it would appear that these data are best collected with
the deployment of 3D sonic anemometry.
Acknowledgments
Special thanks to Eric Dupont, Christian Feigenwinter and Oliver Texier for providing access to
comprehensive data sets which have allowed the analysis in this paper to be conducted.
This work has been carried out with funding from the EU FP7-PEOPLE program under
Marie-Curie Initial Training Network.
WAUDIT
References
[1] Belcher S et al. 2012 The Wind in the Willows. Flows in forest canopies in complex terrain Annual
review of Fluid Mechanics 44 pp 479-504
[2] Kaimal J C and Finnigan J J 1994 Atmospheric boundary layer flows (Oxford: University Press)
[3] Wylie S and Watson S 2011 A CFD study of wind flow around a model forest: Proc. EWEA
(Brussels, Belgium, 14-17 March 2011)
[4] Montavon C 2012 Modelling Of Wind Speed and Turbulence Intensity For A Forested Site In
Complex Terrain: Proc. EWEA 2012 (Copenhagen, Denmark, 16-19 April 2012)
[5] Desmond C J et al. Forest canopy flows in non-neutral stability: Proc. EWEA 2012 (Copenhagen,
Denmark, 16-19 April 2012)
[6] Stull R 1988 An introduction to boundary layer meteorology (Dordrecht: Kluwer Academic
Publishers)
[7] Irwin J and Binkowski F 1981 Estimation of the Monin Obukhov scaling length using on site
instrumentation Atmospheric Environment 15 6 pp 1091-1094
[8] Burns S et al. 2011 Atmospheric stability effects on wind fields and scalar mixing within and just
above a subalpine forest in sloping terrain Boundary Layer Meteorology 138 pp 231-262
[9] Mannan S and Lee F 2005 Lee's Loss Prevention in the Process Industries (London: Elsevier)
[10] Office of Nuclear Regulatory Research 2007 Regulatory guide 1.23: Meteorological monitoring
programs for nuclear power plants (Washington: US Nuclear regulatory Authority)
[11] Feigenwinter C et al. 2008 Comparison of horizontal and vertical advective CO2 fluxes Agricultural
and Forest Meteorology 148 pp 12-24
[12] Lanzinger E and Langmack H 2005 Measuring air temperature by using an ultrasonic
anemometer: Proc. TECO (Bucharest, Romania, 4-7 May 2005)
[13] Zaidi H et al. 2011 Evaluating the ability of two canopy models to reproduce the forested area
effects using code saturn: Proc. ICWE 13 (Amsterdam, Netherlands, 10-15 July 2011)
[14] Meteotest Meteonorm. Global Meteorological Database Version 7 Software and data for
Engineers, Planners and Education (Bern: Meteonorm)
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 60/167
Book of Proceedings
[15] Texier O et al. 2010 Integration of atmospheric stability in wind power assessment through CFD
modeling: Proc. EWEC 2010 (Warsaw, Poland, 20-23 April 2010)
[16] Feigenwinter C et al. 2010 Spatiotemporal evolution of CO2 concentration temperature and wind
field during stable nights at the Norunda forest site Agricultural and Forest Meteorology 150 pp
692-701
[17] Garratt J R 1990 The internal boundary layer. Boundary Layer Meteorology 50 pp 171-203
[18] Wagner R et al. 2011 Accounting for the wind speed shear in power performance measurement
Wind Energy 14 pp 993-1004
[19] Pedersen T et al. 2002 Wind turbine power performance verification in complex terrain and wind
farms Risø-R-1330 (Risø: DTU)
[20] Blodau T 2012 How appropriate are sales power curves on complex or forested sites?: Proc.
EWEA Technology Workshop (Lyons, France, 2-3 July 2012)
[21] Wharton S and Lundquist L 2012 Assessing atmospheric stability and its impacts on rotor disk
wind characteristics at an onshore wind Journal of Wind Energy 15 pp 525-546
Appendix
Table A1: Results the analysis of the stability metrics as presented in Section 4.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 61/167
Book of Proceedings
Figure A1: Occurrence of stability events for different wind speeds at Norunda. This graph was produced using
threshold values of 1/L given in Table 3
Figure A2: Occurrence of stability events for different wind speeds at the Wetzstein case. This graph was produced
using threshold values of 1/L given in Table 3.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 62/167
Book of Proceedings
Figure A3: Occurrence of stability events for different wind speeds for the V-forest case. This graph was produced
using threshold values of
given in Table 3.
Figure A4: Occurrence of stability events for different wind speeds for the V-flat case. This graph was produced using
threshold values of
given in Table 3.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 63/167
Book of Proceedings
Downscaling wind energy resource from mesoscale to CFD model and
data assimilating field measurements into CFD model
Venkatesh Duraisamy Jothiprakasama,1, Eric Duponta, Bertrand Carissimoa
a
EDF R&D, Chatou, France
Presented at European Wind Energy Association Annual Event EWEA-2013, Vienna, Austria, February 2013.
Abstract
The main goal of this research work is to use operational Numerical Weather Prediction (NWP) model
data as input to the Computational Fluid Dynamics (CFD) and assimilate the field measurements into
the CFD solutions. To address the problem of high spatial variation of the topography on the domain
lateral boundaries between NWP and CFD domain boundaries, 3 methods - translation, extrapolation
and Cressman interpolation are tested to impose the NWP model data on the CFD domain lateral
boundaries. Newtonian relaxation data assimilation technique is used to incorporate the field
measurement data into the CFD simulations. These techniques are studied in a complex site located
in southern France. Comparison of wind profiles between the CFD simulation, measurements and
CFD simulation with data assimilation are discussed. This combination of state-of-the-art techniques in
NWP, CFD, and field data assimilation will provide the basis of a more accurate wind resource
assessment method.
1. Introduction
The development of wind energy generation requires precise and well-established methods for wind
resource assessment, which is the initial step in every wind farm project. During the last two decades
linear flow models were widely used in the wind industry for wind resource assessment and micrositting. But the linear models inaccuracies in predicting the wind speeds in very complex terrain are
well known and led to use of Computational Fluid Dynamics (CFD), capable of modeling the complex
flow in fine details around specific geographic features. Numerical Weather Prediction (NWP) or
mesoscale models use mathematical models of the atmosphere and oceans to predict the weather by
assimilating observation of the current weather conditions to predict the flow characteristics and have
been extensively used in weather prediction and forecasting. NWP are able to predict the hourly wind
regime at resolutions of several kilometers, and cannot resolve the wind speed up and turbulence
induced by the topography features finer than 1 km. However, CFD has proven successful in capturing
flow details at smaller scales. Hence combining NWP and CFD models can result in a better modeling
approach for wind energy applications.
Linear model results showed identical flow fields for two opposite wind directions, while non-linear
CFD model results are highly asymmetrical. [4] Emphasis the use of CFD in complex terrain for the
improved accuracy of the simulation and for modeling features of the flow those are necessary in the
classification of wind turbines. The disadvantage of linear model (WAsP) and CFD over the mesoscale
modeling were highlighted in [12] for a complex terrain in western Norway. WAsP compared better
than the CFD-models to the measurements and mesoscale simulation with a complete meteorological
model indicated large wind variation with the site due to the local topographical features. The
1
Corresponding author: Venkatesh Duraisamy Jothiprakasam
CEREA, Teaching and Research Centre in Atmospheric Environment, joint laboratory EDF R&D – Ecole des Ponts/Paris-Tech,
6 quai Watier, 78401 Chatou cedex France, Tel.: 33 6 08 55 81 41, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 64/167
Book of Proceedings
disadvantage of WAsP and CFD models was that mesoscale wind variations were not taken into
account in the modeling.
In recent years many different approaches in coupling of mesoscale (NWP) and microscale (CFD)
models were carried out. The authors in [10] studied the impact of coupling a CFD (CFD-Urban) model
with the Weather Research and Forecasting (WRF) NWP model on urban scale for contaminant
transport and dispersion. CFD model prediction was significantly improved when using the wind field
produced by downscaling WRF output as initial and boundary condition. The key reason is that turning
of lower boundary layer wind and pressure gradient are well represented in the time varying 3D WRF
fields. The development of micro-to-mesoscale modeling capability for the urban regime in which WRF
model is coupled to high resolution CFD (CFD-ACE+/CFD-Urban) using the Model Coupling
Environmental Library (MCEL) was presented in [11]. [1] Proposed a methodology, using the field
measurements as a link between the NWP (ALADIN) model data and CFD (Code_Saturne) model.
This methodology previously developed at EDF R&D by coupling mesoscale and microscale model for
wind resource assessment showed significant reduction in error in comparison with WAsP simulation.
[3] presents an application of NWP model combined with CFD (VENTOS) model to improve wind
power forecasting. CFD simulations for every sector allow to establish linear relations between the
reference meteorological mast and the turbines of the park. Using the power curve and NWP outputs,
[3] evaluate the wind speed and the direction at the reference mast and are able to predict the
production up to 72 hours ahead. Methodology proposed by [2] provides high resolution wind
mapping, by combining NWP (SKIRON) and CFD (CFDWind) model with straightforward link that
relies on calibrated “geostrophic wind” and demonstrated on Granada region in Spain. The authors
[13] placed weight on the need to have a valid link between mesoscale modeling and microscale
modeling, and it is essential for application and verification of modeling results. Different route were
proposed, ranging from direct application of mesoscale data to the recommended and more
sophisticated approach involving corrections at mesoscale and microscale model.
Data assimilation is a NWP concept encompassing method for combining observations of variables
into numerical models, such an approach is not common in CFD simulation. In [8], the authors
successfully implemented a multiscale weather model, which is designed for simulation of weather
process from synoptic (WRF) scale to microscale (LES) with simultaneous nested grids along with
data assimilation. In [7], the process of assimilating mesoscale model data into CFD simulation was
done successfully and preliminary results of this study indicate that its possible to nudge the wind
profile towards the WRF while retaining the mass consistency of the CFD model. It uses the spatially
varying WRF data as inflow condition for CFD (Acusolve) model and assimilating vertical wind profiles
of the fine-scale WRF data into the CFD model in order to nudge the CFD solution towards the WRF
model data.
In this research work, we use operational mesoscale model data as inflow condition for microscale grid
and address the problem of topography variation at the border between mesoscale and microscale
grid. Three methods (Translation, Extrapolation and Cressman interpolation) are developed to impose
inflow condition on the microscale grid. Then we use Newtonian relaxation data assimilation technique
to assimilate the field measurement inside the microscale grid. This is an ongoing research work
performed in the framework of FP7/ITN Marie Curie WAUDIT project. It’s a consortium which aims at
developing new methodologies of added value to the wind industry, such that they can contribute to
new levels of standardization of wind assessment activities.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 65/167
Book of Proceedings
2. Methodology
In this section we discuss the site selected for this application, field measurement campaign, inflow
condition from the operational mesoscale model and microscale model used.
2.1. Description of the site
The area of interest is located in southern France. The site is considered very complex with strong
slopes (locally larger than 30°), valleys and forest (more than 80%). The average altitude of site is
700m. The site is located at the centre of the figure 1 (left), with plain in South-East, valley on the
North-West and further to the North-West a plateau at altitude of 1100m is seen. Figure 1 shows the
elevation map of the site (40x40 km2), inner simulation domain of (20x20 km2) and masts location.
The digital elevation map and land use file have resolution of 25m and 50m respectively and are
obtained from IGN (Institut Géographique National, France).
M80
●
●
●
M
FP
M80
FP
●
●
●
Figure 1: Topography and position of the 3 masts on
the site (colour map in meters).
Figure 2: Location of the simulation domain and
operational ALADIN grid.
Table 1. Description of wind measurements.
2.2. Experimental campaign
A one-year field measurement campaign was led by EDF R&D and EDF Energies Nouvelles between
June 2007 and June 2008 in order to provide input and validation data. The measurement set-up
included 2 sodars, two 50m masts (M and FP) with cup anemometers and vanes, and a 80m mast
(M80) with cup and sonic anemometers, vanes, temperature and humidity sensors. A Remtech PA2
sodar was installed besides the 50m M mast in order to provide the vertical profile of wind and
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 66/167
Book of Proceedings
turbulence between 100 m and 600 m. A Scintec SFAS sodar, which provided the vertical profile of
wind and turbulence between 30 m and 130 m, was located near the M80 mast. Four sonic
anemometers installed on the M80 mast measured the three components of the wind averaged over
10 minutes, and their standard deviation. Table 1 shows the list of the measurement locations,
instrument device used, height of measurement and period of measurement. Access to electricity
network was not possible in this area, so an autonomous device of power supply has been used which
included 10 batteries, 10 solar cells and 4 small wind turbines. The analysis of the sonic and sodar
measurement data reveals that the SFAS sodar underestimated the wind speeds by about 12% in
average when compared to the sonic anemometers. However, the ratio between sodar and sonic wind
speed is highly dependent on the wind direction and is linked to the topography [16].
2.3. ALADIN mesoscale model
ALADIN (Aire Limitée Adaptation dynamique Dévelopement InterNational) is a mesoscale model
developed and operated by several European and North African countries under the leadership of
Météo-France. It maintains a NWP system for use on limited geographic areas, with small domains
and high spatial resolution. The important meteorological events at fine scale (local winds, breezes,
thunderstorms lines) are the main results of the dynamical adaptation to the characteristics of the
earth's surface. It receives boundary data from the French global model-ARPEGE. The horizontal
resolution of ALADIN grid is approximately 9 km with 31 vertical levels. In the present case, analyses
are available at 0h and 12h and forecasts between 1h and 11h and between 13h and 23h with a one
hour step. A new forecast is produced every 12 hours, while the longest range of the forecast is 48
hours (www.cnrm.meteo.fr/aladin/). Figure 2 shows the location of the simulation domain (20X20 km2)
and surrounding operational mesoscale (ALADIN) grid. Météo-France provided the hourly wind speed
and temperature profile for the year 2007, turbulent kinetic energy and dissipation are deduced from
the mesoscale profile using theoretical relations. The operational ALADIN runs provide the initial and
boundary condition for the CFD modeling. Since the beginning of 2012, ALADIN is no more used
operationally at Météo-France, however, the methodology developed in this framework can be applied
with any mesoscale model.
2.4. Microscale CFD modeling
Microscale model used for this research work is an open-source CFD code, Code_Saturne developed
by EDF R&D and CEREA [14]. It is a general purpose CFD code, which handles complex geometry
and physics. The atmospheric module takes into account the large scale meteorological conditions
and thermal stratification of the atmosphere. This module is used for wind energy engineering, urban
canopy and pollutant dispersion modeling. It is a finite-volume code, robust for application of Reynolds
Averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) models. Although RANS is less
accurate in comparison with LES, RANS is the most commonly used CFD model for simulation
involving turbulent flow in industrial and engineering application because it is computationally less
expensive. The turbulence in the simulation domain is parameterized by the standard k-ε turbulence
closure model as in the previous work [1]. The atmospheric stability is not taken into account in the
simulations.
Inlet turbulent kinetic energy, k and dissipation ε, are generated using the frictional velocity, u*
deduced from the ALADIN wind profile. A study was conducted to assess the grid resolution for this
domain. The model applies no-slip boundary condition on the ground with roughness value deduced
from land use file, velocity inflow condition is prescribed on the north and west face for the
meteorological situation considered here, outflow condition on the east and south face and symmetry
condition at the top. The simulations are run until a stationary state is reached in the domain.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 67/167
Book of Proceedings
3. Coupling and data assimilation
In this section, methods developed for imposing mesoscale model wind profile on microscale grid and
assimilating field measurement into the microscale grid are discussed.
3.1. Imposing mesoscale boundary condition on microscale grid
The differences in horizontal (∆xy) and vertical resolution (∆z) between mesoscale grid (∆xy ≈ a few
km and ∆z ≈ some tens of meters) and CFD grid (∆xy ≈ some tens of m and ∆z ≈ some meters) are
very large, whereas the topography variations are strong at the border of the simulation domain. The
reliefs seen by the mesoscale and microscale codes are different. In particular, there are some
unavailable values of the model variables when the relief altitude is higher in the mesoscale grid than
in microscale grid. Hence imposing the inlet boundary condition on the CFD domain is a difficult task
and different strategies need to be implemented. Methods used to impose mesoscale data on
microscale grid are: translation, extrapolation and Cressman interpolation (refer to Figure 3).
Translation method was developed and implemented in [1]. In translation, the mesoscale wind profiles
are translated along the absolute co-ordinates of the microscale terrain. In this method wind speed
varies along the height above ground level which is not reliable far from the ground. In extrapolation,
the mesoscale wind profiles are extrapolated between the lowest level of the mesoscale profile and
the absolute altitude of the local ground level in the microscale grid. Thus, above a certain level, the
inlet wind speed obtained in extrapolation method is function of height above sea level.
Translation
Extrapolation
Cressman interpolation
Figure 3. Methods of imposing mesoscale wind profile on microscale grid boundaries.
Cressman interpolation is a method which is widely used in NWP [6]. Using the Cressman
interpolation, the wind speed on a boundary face of the microscale grid is calculated as a linear
combination of the values provided by the nearby mesoscale data. Figure 3 shows example of
Cressman interpolation, the observations O1 and O2 influence the grid point P, O3 doesn’t. These
values are weighted depending only upon the distance between the CFD grid point and the mesoscale
grid points. Thus, the inlet horizontal wind components are calculated as combinations of mesoscale
data following:
Vint erpolate =
naladin
naladin
i =1
i =1
∑ ViWi ∑ Wi
(1)
Where naladin is the number of available ALADIN verticals, the weights Wi are functions of the
distance between the CFD grid point where the wind speed is calculated (coordinates xvalue, yvalue,
and zvalue) and the nearby ALADIN grid points (coordinates xi, yi,and zi):
 0

r2
 2 if i ≥ 700 
Wi =  ri

2
−
e 4 else



WAUDIT
Contract#
238576
(2)
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 68/167
Book of Proceedings
where
2
ri
2
2
2
x
− xi   yvalue − yi   zvalue − zi 
=  value
 +
 +

rL
rL
rZ

 
 

(3)
rL and rZ are longitudinal and vertical radius of influence fixed at the beginning of the computation.
Thus, Cressman interpolation is able to calculate the mesoscale model variables on the microscale
grid.
3.2. Field measurement data assimilation into CFD model
Data assimilation was originally developed for NWP, in which meteorological observations enter the
models forecast and analysis cycles. Nudging, a Newtonian relaxation data assimilation technique is
used to incorporate the measurement data into the CFD simulations. It consists in adding to the
prognostic equation for the wind components a term that nudges the solution towards the observations
[6, 15]. This term that has a negative sign "keeps" the calculated state close to the observations. For
example, a primitive equation model, the velocity forecast equation is written as and similarly for all
other equations.
∂τ ij  uobs − us 
∂ui
∂u
1 ∂p
∂ 2u i
+ uj i = −
+v
−
+
 * W ( x, y , z, t )
ρ ∂xi
τu
∂t
∂x j
∂ x j ∂x j ∂x j 

u
u
(4)
τ
where obs , s , u and W ( x , y , z, t ) are observed velocity, predicted velocity, relaxation time scale
and cressman type spatial and temporal weighting function respectively. The relaxation time scale is
chosen based on empirical consideration and may depend on the variables. Then we use Cressman
type volumetric interpolation in the horizontal and vertical direction in order to calculate the observed
−r / 2 R
values on the grid points. A radial drop-off rate of e
is used for W ( x , y, z , t ) , where r is the radial
distance from observation point and R is arbitrarily set at 10m for a smooth drop-off. This type of
function is used in NWP and it doesn’t vary in time since time integration is very small in CFD
τ
simulation. The relaxation time scale, u is set here to 50 s. This type of modeling approach is not
common in CFD simulation and Code_Saturne is modified to incorporate this additional forcing term.
4. Results
4.1. Case description
The case chosen for initial analysis is typical of the most dominant wind direction (North-Westerly) and
is on 17th November 2007 and the specific time for the CFD simulation is 20:00 CET. The surface
wind speed and temperature are 6 m/s and 10 °C respectively. Figure 4 shows the wind rose from the
mast M80 location.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 69/167
Book of Proceedings
Figure 4: Wind rose from the M80 mast location
Figure 5: ALADIN u and v velocity components and Cressman
interpolated velocities for all microscale inlet faces for rL=8500
m and rZ=100 m.
4.2. Coupling method
CFD simulations are carried out for inflow conditions computed using 3 methods - translation,
extrapolation and Cressman interpolation. For the translation and extrapolation methods, ALADIN
profiles either upstream of the flow or at the centre of the domain are chosen to compute inflow
conditions. Cressman interpolation method uses ALADIN profiles in and around the simulation domain
to compute the inflow conditions. Sensitivity of the horizontal and vertical radius of influence are
tested, for horizontal radii 2000, 4250, 8500 and 10000 m and for vertical radii 100, 200 and 500 m.
Figure 5 shows u and v velocity components of all ALADIN grid points and Cressman interpolated u
and v velocity components for all inlet microscale grid faces. For horizontal radius (rL) 4250 m and
beyond no difference is seen in the imposed inlet velocities at the boundaries and horizontal radius (rL)
2000 m lead to use only the closest ALADIN vertical. Horizontal radius (rL) 8500 m and vertical radius
(rZ) 100 m are chosen for any further simulations.
The results of CFD simulations using translation, extrapolation and Cressman interpolation (as inlet
boundary) were compared and they show Cressman interpolation performed better compared to
translation and extrapolation. Translation and extrapolation use a single mesoscale profile as inlet
boundaries and results vary depending on the chosen mesoscale grid point. Cressman interpolation
uses combination of mesoscale grid points taking into account the altitude of the microscale grid and
mesoscale grid. However, hourly mesoscale inlet data are unable to predict strong variations in
velocities at the measurement locations. This variability cannot be reproduced by the CFD models, as
the results of the simulation correspond to a stationary state for given boundary conditions. On the
whole, using Cressman interpolation method to compute the inflow at lateral boundary seems to be
the better option for initializing and maintaining the wind speed level over the topography.
4.3. Comparison of CFD simulation without nudging and with nudging
Firstly we run CFD simulation using Cressman interpolated mesoscale wind fields as inlet boundary
condition and see how they predicted wind flow fields over the complex terrain. We have very good
data set at M80, M and FP mast location for this situation. Figures 7, 8 and 9 show the on-site field
measurements plotted for masts M80, M and FP respectively. At M80 location, three consecutive 10
minutes average velocity of cup anemometer and sonic anemometer measurement and three
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 70/167
Book of Proceedings
consecutive 20 minutes average velocity of short range sodar are plotted. Cup and sonic
measurements are in good agreement with each other. The sodar measurements at M80 show the
velocity profile until 100 m and are in good agreement with cup and sonic measurements. At M mast,
three consecutive 10 minutes average velocity of cup anemometer are available but sodar data are
not available for this situation. At FP location, three consecutive 10 minutes average velocity of cup
anemometer are plotted. It is to be noted that we apply hourly mesoscale wind speeds at inlet
boundaries and compare with 10 and 20 minutes average velocities at the three locations.
Then CFD simulation with data assimilation (nudging) is carried out for the same situation. At M80
location three data sets are available and we choose to nudge a data set and validate with remaining
data. Figure 6 shows the imposed Cressman weighting function at the M80 mast location,
W ( x , y, z , t ) = 1 at exact location of M80 and it drops exponentially outwards to W ( x, y, z , t ) = 0 . We
nudge M80 sonic measurements on 17/11/2007 at 20:00 CET for this situation and monitor the
influence of nudging at the other 2 nearby mast locations.
Figure 6: Cressman weighting function imposed at mast
M80 location.
Figure 7: Comparison with measurements at M80 of
velocity profile calculated with and without data
assimilation.
Figures 7, 8 and 9 show the comparison of measurements, CFD (CFD simulations without nudging)
and CFD+nudging (CFD simulation with nudging at M80) at mast M80, M and FP locations. At M80
mast location, CFD simulation predicted well and CFD+nudging simulation nudges the 4 levels of
sonic measurements into the simulation and improvement is noticed in comparison with CFD
simulation without assimilation. Also it is noticed that ALADIN mesoscale grid point closest to the
masts location over predicted velocity, which leads also to an overestimation of the wind speed in the
CFD simulation without assimilation. At M location, variability in three consecutive 10 minutes average
is noticed, and CFD+nudging better compares to measurements at 20:00 than CFD simulation without
nudging. At FP location, the improvement with assimilation is even more obvious, although there was
not any FP measurement assimilated.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 71/167
Book of Proceedings
Figure 8: Comparison with measurements at M of
assimilation.
Figure 9: Comparison with measurements at FP of
assimilation.
Figures 10 and 11 present the zoom-up of the velocity magnitude and vectors cross-section at mast
locations for CFD simulation without nudging and CFD simulation with nudging respectively. They
show the significant effect of data assimilation on the wind speed within all the zone of interest, but
also an important effect on the direction, as a result of the separated assimilation on u and v
components. Thus data assimilation may allow to correct at least partly some errors in wind speed and
direction calculated by the mesoscale and CFD models.
Figure 10: Zoom up of velocity magnitude and vector
cross-section for CFD simulation without nudging.
Figure 11: Zoom up of velocity magnitude and vector
cross-section for CFD simulation with nudging.
Simulations performed for other situations confirmed that assimilating data at M80 mast allows to
improve the comparison with measurements not only at M80 but also at the two other masts locations.
5. Conclusions
This research work implemented the operational mesoscale data into the microscale grid using 3
coupling methods and incorporated field measurement into the CFD model using nudging as data
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 72/167
Book of Proceedings
assimilation technique. Cressman interpolation method better represented the velocities at the inlet
boundaries compared to translation and extrapolation method. The results from the CFD with
assimilation indicate that incorporation of the field measurement data into Code_Saturne can nudge
the wind speeds and direction towards the measurements while keeping mass consistency of the
model. The improvement is noticed not only at the mast where data have been assimilated, but also
on the two other masts locations, leading to a more realistic wind field in the whole region concerned
by wind turbines installation. In the future, thermal stratification, and forest canopy model will also be
introduced. Annual energy prediction will then be calculated from the number of CFD simulations
deduced from a clustering process.
Acknowledgements
This is an ongoing research work performed in the framework of FP7/ITN Marie Curie WAUDIT
project. It’s a consortium which aims at developing new methodologies of added value to the industry,
such that they can contribute to new levels of standardization of wind assessment activities.
References
[1] [Laporte L, Dupont E, Carissimo B, Musson-Genon L and Sécolier C 2009 Atmospheric CFD
simulation coupled to mesoscale analyses for wind resource assessment in complex terrain Proc.
European Wind Energy conf (Marseille, France)
[2] Sanz Rodrigo J, Garcia B, Cabezon D, Lozano S and Marti I 2010 Downscaling mesoscale
simulations with CFD for high resolution regional wind mapping Proc. European Wind Energy conf
(Warsaw, Poland)
[3] Rodrigues V, Santos S, Palma J, Castro F, Miranda P and Rodrigues A 2008 Short-term
forecasting of a wind farm output using CFD Proc. European Wind Energy conf (Brussels,
Belgium)
[4] Politis E S and Chaviaropoulos P K 2008 Micrositing and wind turbine classification in complex
terrain Proc. European Wind Energy conf (Brussels, Belgium)
[5] Coirier W J, Kim S, Marella S, Mayes J, Chen F, Michalakes J, Miao S and Bettencourt M 2007
Progress towards coupled mesoscale and microscale modeling capability (Bettencourt Consulting,
LLC, USA) (CFD Research Corporation - National centre for Atmospheric Research)
[6] Kalnay E 2008 Atmospheric Modeling, Data Assimilation and Predictability (Cambridge University
Press)
[7] Haupt S E, Zajaczowski F J and Schmehl K J 2011 A preliminary study of assimilating numerical
weather prediction data intp computational fluid dynamics model for wind prediction J. Wind Eng.
Ind. Aerod. 99 320-329
[8] Liu Y, Warner T, Liu Y, Vincent C, Wu W, Mahoney B, Swerdlin S, Parks K, Boehnert J 2011
Simultaneous nested modeling from the synoptic scale to the LES scale for wind energy
applications J. Wind Eng. Ind. Aerod. 99 308-319
[9] Yamada T and Koike K 2011 Downscaling mesoscale meteorological models for computational
wind engineering application J. Wind Eng. Ind. Aerod. 99 199-216
[10] Tewari M, Kusaka H, Chen F, Coirier W J, Kim S, Wyszogrodzki A A, Warner T T 2010 Impact of
coupling a microscale computational fluid dynamics model with a mesoscale model on urban scale
contaminant transport and dispersion Atmos. Res. 96 656-664
[11] Cionco R M, Luces S A 2002 Coupled Mesoscale-Microscale model to compute neighborhood
scale wind fields 4th Symp. Urban Environment
[12] Berge E, Gravdahl A R, Schelling J, Tallhaug L and Undheim O, 2006 Wind in complex terrain. A
comparison of WAsP and two CFD-models. Proc. European Wind Energy conf (Athens, Greece)
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 73/167
Book of Proceedings
[13] Badger J, Hahmann A, Larsen X G, Diaz A P, Batchvarova E, Gryning S E, Floors R and
Jørgensen H E 2011 Comprehensive utilization of mesoscale modelling for wind energy
applications Proc. European Wind Energy conf (Brussels, Belgium)
[14] Code_Saturne 2.0.6, Theory and programmer’s guide, EDF R&D, http://research.edf.com/
research-and-the-scientific-community/software/code-saturne/download-code-saturne-80059.html
[15] Warner T T 2011 Numerical weather and climate prediction (Cambridge University Press)
[16] Dupont E, Lefranc Y and Sécolier C 2009 A sodar campaign in complex terrain for data quality
evaluation and methodological investigations Proc. European Wind Energy conf (Marseille,
France)
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 74/167
Book of Proceedings
The Anisotropic Multifractal Model and Wind Turbine Wakes
G. Fittona,1, T. Tchiguirinskaiaa, D. Schertzera, S. Lovejoya
a
Universitè Paris Est, Ecole des Ponts ParisTech, Marne-la-Vallée, France
b
McGill University, Physics Department, Montreal, Quebec, Canada
Presented at the 7th EAWE PhD Seminar on Wind Energy in Europe, TUDelft, Delft, Netherlands, October 2011.
Keywords: Universal Multifractals, Spectral Analysis, Wind Velocity Fluctuations and Power Estimation.
1. Introduction
A typical routine in wind field resource assessment, at the most basic level, consists of first to third
order statistics of times series data. The quality of the time series data can range between 0.05 to 600
seconds. More often than not the frequency of data will be the latter of the two since it is the
cumulative power over long periods of time that define the financial return from turbines and thus highresolution data is deemed unnecessary. It is now evident that such coarse time series data are no
longer sufficient for a representative assessment of the wind and that estimations based on such data
are associated with inaccurate power curve prediction and turbine damage. In particular it has been
suggested that such problems are due to a lack of understanding of the somewhat intermittent nature
of the wind velocity fields and the small-scale fluctuations thus associated. In order to address this
there has been a significant increase in research involving coupled mesoscale-microscale models and
stochastic downscaling methods. Our contribution is a demonstration that a good knowledge of smallscale variability is essential for a better understanding of the atmospheric boundary layer. We discuss
the applicability of the stochastic anisotropic multifractal model to the complex conditions of wind farm
potential and operational sites.
1. Data
Available to us is six-months of wind velocity and temperature measurements at the heights 22, 23
and 43m.
Figure 1: Schematic of turbine positions and wake effect due to North-Westerly winds (map courtesy of Julien
Richard).
1
Corresponding author: George Fitton
Universitè Paris Est, Ecole des Ponts ParisTech, LEESU, 6-8 avenue B. Pascal, Citè Descartes, 77455, Marne-la-Vallée
cedex 02, France; tel.: +33 1 64 15 36 07, fax: +33 1 64 15 37 64, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 75/167
Book of Proceedings
The measurements came from 3D sonic anemometers with a 10Hz data output rate positioned on a
mast in a wind farm test site subject to wake turbulence effects (see Fig. 1). The quality of the data
was of utmost importance so thorough pre-processing and verification was implemented to assure the
reliability of the results.
3. Analysis
3.1. The Energy Spectrum and Scaling
A typical first-step-method to determine the overall scaling behaviour is the transformation of the
velocity field into Fourier space. We ‘should’ then be able to observe powerlaw behaviour of the
spectrum such that
E (ω ) = Aω −β
(1)
where ω is the frequency, E(ω) is the energy at a given frequency, A is a coefficient of proportionality
and b is the scaling exponent. The review of [Marusic et. al., 2010] discusses the existence of a -1
power law sub-range over small frequencies, adjoined by a classical Kolmogorov inertial sub-range
with β = 5/3.
We will present shortly a more in-depth discussion on how our results compare to Kolmogorov’s
predictions however before this we would like to discuss the fact that there is no unique scaling regime
i.e. there are three common scaling features, instead of the predicted universal law (see Figs. 2 and 3
also), that are:
•
•
•
High frequency scaling range (RHF: ~0.1 secs to ~5 mins) in which all three velocity
components, u, v and w, follow (approximately) the same scaling law.
Mid-frequency w-component departure from scaling at ~5 minutes. Mid-Frequency, RMF,
corresponds to the ranges ~5 mins to ~1 hour.
Low frequency scaling reunification (RLF: ~1 hr to ~1 day) for all three velocity components at
about an hour. The power law is not the same as that for small scales as will be discussed
later.
The focus therefore of our more in-depth analysis is the behaviour of the horizontal u- and vcomponents over the midfrequency-ranges i.e. ~5 mins to ~1 day. In fact what we found was that our
data fell into two categories; days (i.e., independent samples of 219 measurements [≈14.5 hours] per
day) without a mid-frequency perturbation (Fig. 2) and days with a mid-frequency perturbation (Fig. 3).
In the next section we will consider the simpler of the two regimes that are the non-perturbed days.
3.2 Non-perturbed Days & The Anisotropic Multifractal Model
The results from spectral analysis on non-perturbed days confirm a unique power law for all three
velocity components over higher frequencies up to approximately 40 seconds at which the vertical
wind w-component shows a clear scaling break followed by a -1 power law subrange as described in
the previous section.
Moreover, such a clear separation of power law subranges allows us to obtain an estimate of the
integral length scale for the vertical wind component as suggested in [Monin & Yaglom, 1975], which
in turn leads to an estimate of the Reynolds number of about 60,000. Thus, from dimensional analysis
one may obtain a minimum Reynolds number of about 14,000. These estimates confirm that the
investigated wind field exhibits fully developed turbulence.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 76/167
Book of Proceedings
Figure 2: Averaged spectra for 11 non-perturbed days where the velocity component u is blue, v is green and w is red.
The high-frequency range from ~0.1 sec to 5 mins has spectral slope ~1.4, less than the predicted 5/3. In addition we
have highlighted the -1 adjoining range, from 5 mins to an hour, with the scale break being predictable based on the
mast height (see [Fitton et. al., 2011] for more details). Low frequency scaling region is compatible with the -11/5
scaling law.
Over the high-frequency range Fig. 3 displays spectral exponents that differ from Kolmogorov’s -5/3
law. The difference corresponds to an intermittency correction of spectral slopes and can be taken into
account using the universal multifractal framework (Schertzer and Lovejoy, 1987), where:
•
•
the energy density flux is a conserved (at any scale ratio λ ) multifractal field proportional to a
power law with singularity, g, i.e.
ελ ∝ λ γ ,
(2)
the statistical moments of the energy density flux are defined by:
ε λq ∝ λ K ( q ) ,
•
(3)
and the scaling moment function K(q) is defined by:
C
K (q ) = 1 q α − q .
α −1
(
)
(4)
Here, q, is the order of moment, C1 is the codimension of the mean singularity and a is the multifractal
Lévy index. The spectral exponent of Eq. 1 now becomes
β = 2H + 1 − K (2) .
(5)
where H = 1/3 quantifies the degree of non-conservation of velocity increments. For spectra (i.e. for
second order statistics), we estimated K(2) = 0.27. Such high intermittency corrections are expected
over high frequencies in areas with high Reynolds numbers and complex terrain.
In addition we observed the Bolgiano-Obuhkov -11/5 power law at low frequencies illustrating the
influence of largescale vertical motions specific to the topography of our wind farm test site [Faggio &
Jolin, 2003].
To take into account the dominant role of the vertical motion of large scale atmospheric structures,
one may consider that the buoyancy force variance flux, f, plays the same role as the energy flux, e, in
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 77/167
Book of Proceedings
3D turbulence but only along the vertical [Schertzer & Lovejoy, 1984]. This is contrary to the classical
‘buoyancy subrange’ that postulates an isotropic turbulence [Bolgiano, 1959, Obukhov, 1959] with two
different (horizontal and vertical) scaling regimes. Thus we have the coupled sets of scaling equations
[Schertzer & Lovejoy, 1984, Lazarev et. al., 1994]:

∆x1/ 3 

d
1/ 5
3 /5 
∆ V ( ∆ z ) = (φ ( ∆ z ) ) ∆ x

d
∆V ( ∆x ) = ( ε ( ∆x ) )
⇒ ( ε ( ∆x ) )
1/ 3
1/ 3
≈ (φ ( ∆z ) )
1/ 5
when ∆x1/ 3 ≈ ∆z3 / 5
(6)
(7)
where ∆V(∆x) and ∆V(∆z) denote the horizontal and vertical shears of the horizontal wind respectively
d
and the symbol = means equality in probability distribution. Because the scaling fluctuations of both
fluxes are not neglected (due to their explicit scale dependency) we can define anisotropic scaling (as
defined by the anisotropic multifractal model [Schertzer & Lovejoy, 1984]) at all significant scales
instead of two isotropic regimes, separated by a scaling break (see [Fitton et. al., 2011] for more
details).
Figure 3: Averaged spectra for 11 perturbed days where the velocity component u is blue, v is green and w is red. The
high-frequency range from ~0.1 sec to 5 mins has spectral slope ~1:6 which is much closer to the predicted 5/3. We
have highlighted the fluttering for the horizontal components over RMF. We can also see the fluttering of the vertical
component is accentuated to a plateau. The 11/5 low frequency scaling regime remains, although with a lower
coefficient of proportionality A (Eq. 1).
3.3. Perturbed Days, Wakes and Power Estimation
In [Fitton et. al., 2011] we put forward the argument that the non-perturbed days were a result of lack
of influence of wind turbines justified by the low frequency power law (crossdiagonal mean wind) of the
integrated cospectral analysis. The same argument allowed us to select days that were highly
perturbed. By this we mean days where the midfrequency range, RMF, in which the scaling of
horizontal velocity components remained the same as described in the previous section, now have
significant fluttering (see below [Fig. 3]).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 78/167
Book of Proceedings
To see the effect of the turbines we can do a direct comparison of the integrated spectra, ωE(ω), in
log-linear coordinates of perturbed and non-perturbed days (11 of each see Fig. 4).
Figure 4: Comparison of perturbed and non-perturbed, u-component averaged integrated spectra, ωE(ω), in loglinear
coordinates; blue is perturbed days with lightblue moving average, green is non-perturbed with red moving average
and purple is the differences of the moving averages.
This gives us a quantification of the energy per frequency increment making the overall evaluation of
the energy gains and losses much easier. We have selected the horizontal u-component since there is
no -1 adjoining range for nonperturbed days making it easier to make the comparison.
Note the behaviour of the horizontal v-component is very similar (evidence of asymmetry at larger
scales). From Fig. 4 we can draw the following intermediate conclusions based on the ranges defined
in Section 3.1:
•
•
•
High frequency scaling range (~0.1 secs to ~5 mins) has an injection of energy since
perturbed days (blue integrated spectra, light-blue moving average in Fig. 4) have more
energy than the unperturbed days (green integrated spectra, red moving average in Fig. 4).
This is confirmed by the positive difference of the moving average of the integrated spectra
(purple curve of Fig. 4). If we consider the most basic approximation to a turbine, the actuator
disc, then we can assume any eddy larger than the disc will be split into smaller eddies. This
may explain the increase in high frequency energy. In fact, we can further confirm this idea
since the transition of energy peaks at ~5 mins highlighted again in Fig. 4 correspond to the
size of the wake shown in Fig. 1.
Mid-frequency u-component (~5 mins to ~3 hours) shows evidence of energy pumping from
the turbines for the perturbed days. This is more obvious when looking at the negative
difference of the two integrated spectra over this range.
Low frequency (~3 hours to ~1 day [mesoscales]) shows that although there is similar scaling
behaviour the energy for the perturbed days (red curve) is greater than the non-perturbed
(light-blue curve) since the difference of the two (purple line) is positive. In [Fitton et. al., 2011]
we suggested this was because the two particular types of wind the site was typically subject
were strong North-Westerlys and weak South-Easterlys. This meant only the stronger winds
would interact with the turbines (see Fig. 1).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 79/167
Book of Proceedings
In addition we see at ~3 hours the energy of the nonperturbed days becomes greater than
perturbed. In the previous section we discussed how topographical features can change the
scaling power law over the lower frequency data. This suggests there are similar topographical
influences causing the loss of energy e.g. higher mean winds dissipate more energy over
complex terrain.
Fig. 5 displays a schematic diagram that illustrates the corresponding inter-relations of different
scaling ranges of the energy spectra. Over each of these ranges, two distinct power laws describe the
corresponding scaling behaviour, with and without wake effects. Thus, from Eq. 5 we get:
E1 ( ω ) = A1ω − β1 ,
(8)
E2 ( ω ) = A2ω − β2 .
(9)
Since the estimates of the multifractality parameter, α, remain stable for both perturbed and nonperturbed fields, the ratio of the energy spectra is defined by the second order structure function:
E1 ( ω )
E2 ( ω )
=
A1 −ζ ∆ (2)
ω
A2
(10)
where ζ ∆ = 2 ( ∆H ) − ( ∆C1 /(α −1) ) ⋅ ( 2α −2) from Eqs. 4 and 5.
Figure 5: Schematic of the inter-relations of different scaling ranges of the energy spectra in a log-log plot.
From Fig. 5, Eq. 4 and the above equation (Eq. 10) we see an empirical spectral exponent closer to
the theoretical values of β = 5/3 (over small scales) or β = 11/5 (over large scales), correspond to a
smaller intermittency correction K(2). Figs. 4 and 5 therefore suggest that by taking the energy over
large scales, wind turbines create additional small-scale eddies and re-inject them as part of the
energy over smaller scales, making the turbulence more homogeneous.
4. Conclusions
The aim of this study was to explore the scaling behaviour of atmospheric velocity measurements in a
wind farm test site subject to wake turbulence effects. Based on this study we can make the following
conclusions:
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 80/167
Book of Proceedings
•
•
•
•
•
Using long time series, 10Hz data, we identified (depending on the direction of the mean wind)
two or three scaling sub-ranges.
Through spectral analysis we found possible relations between wind velocity scaling breaks
and associated theories of fully developed turbulence in the atmospheric surface-layer and
used the universal multifractal framework to deal with the strong intermittency of the field.
We have discussed how the anisotropic multifractal model can be applied to near wall
atmospheric turbulence over complex terrain how it can be fully validated for days with no
interaction with the wind turbine wakes.
We found empirical evidence of the influence of wakes and suggested reasoning and scaling
techniques that enable us to quantify the loss of energy with the potential of taking this into
account using the anisotropic multifractal model.
And finally, we discussed how the pumping of energy from wind turbines over mid-frequency
scales, creates additional small-scale eddies which are re-injected as part of the energy over
smaller scales. This makes the turbulence more homogeneous over the smaller scales in an
analogous way to grid-generated homogeneous turbulence.
References
[1] [Bolgiano, 1959] BOLGIANO, R. 1959 Turbulent spectra in a stably stratified atmosphere, J.
Geophys. Res. 64, 2226.
[2] [Faggio & Jolin, 2003] FAGGIO, G. & JOLIN, C. 2003 Suivi ornithologique sur le parc dèoliennes
dErsa- Rogliano (Haute Corse) - Rapport final-SIIF/AAPNRC-GOC, 100p.
[3] [Fitton et. al., 2011] FITTON, G. F., TCHIGUIRINSKAIA, I., SCHERTZER D., & LOVEJOY, S.
2011 Scaling Of Turbulence In The Atmospheric Surface-Layer: Which Anisotropy?, Journal of
Physics: Conference Series (in review) Warsaw, ETC13.
[4] [Lazarev et. al., 1994] LAZAREV, A., SCHERTZER, D., LOVEJOY, S. & CHIGIRINSKAYA, Y.
1994 Unified multifractal atmospheric dynamics tested in the tropics: part II, vertical scaling and
generalized scale invariance, 115-123.
[5] [Marusic et. al., 2010] MARUSIC, I., MCKEON, B. J., MONKEWITZ, P. A., NAGIB, H. M., SMITS,
A. J. & SREENIVASAN, K. R. 2010 Wall-bounded turbulent flows at high Reynolds numbers:
Recent advances and key issues Phys. Fluid., 22, 065103.
[6] [Monin & Yaglom, 1975] MONIN, A. S. & YAGLOM, A. M. 1975 Statistical Fluid Mechanics,
Cambridge, MITPress, Vol. 2, pp. 874.
[7] [Obukhov, 1959] OBUKHOV, A. N. 1959 Effect of Archimedian forces on the structure of the
temperature field in a temperature flow, Sov. Phys. Dokl. 125, 1246.
[8] [Pinus et. al., 1967] PINUS, N. Z., REITER, E. R., SHUR, G. N. & VINNICHENKO, N. K. 1967
Power spectra of turbulence in the free atmosphere, Tellus 19, 206.
[9] [Schertzer & Lovejoy, 1984] SCHERTZER, D. & LOVEJOY, S. 1984 On the Dimension of
Atmospheric motions. In: T. Tatsumi (Editor), Turbulence and Chaotic phenomena in Fluids,
Amsterdam, Elsevier Science Publishers B. V., pp. 505-512.
[10] [Yaglom & Kader, 1989] YAGLOM, A. M., KADER, B. A., & ZUBKOVSKII, S. L. 1989 Spatial
Correlation Functions of Surface-Layer Atmospheric Turbulence in Neutral Stratification, Bound.Lay. Meteorol. 47, pp. 233-249.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 81/167
Book of Proceedings
Model evaluation methodology for wind resource assessment
Heather A. Holmesa,1, Javier Sanz Rodrigob, Daniel Cabezónb, Michael Schatzmanna
a
Center for Marine and Atmospheric Sciences (ZMAW), University of Hamburg, Hamburg, Germany
b
National Renewable Energy Centre (CENER), Sarriguren, Spain
Keywords: Wind Resource Assessment; Wind Turbine Wake, Validation, Verification, Model Evaluation.
Abstract
Literature shows a large range of wind turbine power output predictability, with under and over
prediction, from blind comparisons of numerical simulations with both field and wind tunnel test data.
Standardized quality assurance procedures need to be developed to improve power output
calculations, especially because the economic feasibility assessment for wind energy projects relies
on these models. The intent is not to evaluate and rank individual models, but to work simultaneously
with modelers and experimentalists in developing a consensus to establish quality assurance methods
to be utilized in the wind energy community. Numerical models undergo evaluation by the developers
to determine the reliability of the results. However, each modeler utilizes a different procedure, and
has access to different experimental datasets. Model evaluation for wind energy assessment is
difficult due to the complex processes that exist in the atmosphere, i.e., boundary layer stratification
and non-stationary time series data. To improve the results obtained from wind energy models a
database of quality checked experimental data, from physical models and field data, must be
accessible to numerical modelers. The objective is to develop methods that provide guidance and
outline quality check procedures for experimentalists and numerical modelers to ensure consistency
and improve the modeled results. Due to the different scales being modeled, varying model
applications and computational advancements several types of models are implemented to evaluate
wind energy potential. It is expected that each type of model will require a separate evaluation
procedure. However, a general evaluation procedure with purposed validation metrics is given and
applied to examples for terrain flow and two wake flow cases.
1. Introduction
Blind comparisons from the National Renewable Energy Laboratory (NREL) in the United States show
a large range of power output predictability for the numerical modeling of wind turbines compared to
wind tunnel test data. With a range of 60% underprediciton to 150% overprediction for the power
output from a wind turbine with simple unyawed, unstalled operating conditions [1]. This is evidence
that standardized quality assurance methods need to be developed for the numerical modeling of wind
energy applications to improve power output calculations. The estimations from these models are
used for economic justification, determination of wind farm locations and the siting of individual wind
turbines within a wind farm. With such high uncertainty in the numerical models for the most idealized
scenario (single turbine, simple operating conditions, neutral boundary layer and steady flow) a large
amount of error exists in the economic predictability for wind energy assessment.
1
Corresponding author: Heather A. Holmes
School of Civil and Environmental Engineering, Georgia Institute of Technology, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 82/167
Book of Proceedings
The application of quality assurance methods to wind energy assessment models is difficult because
of the different scales and modeled output variables required. Due to the scale differences, a few
meters up to more than 100 km, different types of models are used based on the application of the
modeled results. The first type, wind flow models are the first assessment tool used to determine the
economic viability of wind energy at a specific location. These are mesoscale models, also called
micrositing models, that predict mean wind flow patterns to determine the potential wind energy
extraction for a given location. Currently, the standard model being used is WAsP (Wind Atlas
Analysis and Application Program), which estimates wind conditions at a specified location based on
long term meteorological records collected at another location (reference location) [2]. While it is a
useful tool, the WAsP model is often used outside of its recommended operating conditions, which
requires neutral atmospheric stability and mild terrain with gentle slopes. With the advancement in
computing capabilities the implementation of more detailed computational fluid dynamics (CFD)
models for wind flow analysis are being used to address these limitations.
Wake models are the microscale fluid mechanics models used to investigate the flow behavior behind
a wind turbine (wake), and the interaction of the turbine wakes in wind farm situations. These models
are used to investigate two wake ranges, near and far wake, with the separation between the two
occurring approximately one to two rotor diameters downstream from the turbine [3]. The purpose of
the near wake investigation is to study the performance and power output of the rotor design for an
individual wind turbine, therefore airfoil aerodynamics are investigated. Where for the far wake,
incident flow is modeled to determine wind farm layout for maximum power extraction based on
topography, turbulence and turbine wake interactions [4].
Model developers must verify and validate the numerical model output, but currently a standardized
procedure for this model evaluation in regard to wind energy does not exist. To improve the results
obtained from wind energy models a common evaluation methodology and database of quality
checked experimental data must be accessible to numerical modelers. The objective is to develop
methods that provide guidance and outline quality check procedures for experimentalists and
numerical modelers to ensure consistency, improve the modeled results and reduce uncertainties in
wind energy predictions. The intent is not to evaluate and rank individual models, but to work
simultaneously with modelers and experimentalists in developing a consensus to establish quality
assurance methods. However, standardized quality assurance procedures are needed to improve the
reliability of power output calculations, especially because the economic feasibility assessment for
wind energy projects relies on these models.
This work will focus on the methodology for model evaluation and the importance of establishing a
common procedure for validation. Additionally, proposed quantities of interest and validation metrics
will be given, including three examples using these metrics for terrain and wake flow test cases. The
purpose is to present an initial framework of the validation processes as part of the evaluation
methodology specific to wind energy to encourage the community to adopt a standardized procedure.
Of primary importance for this validation procedure is the emphasis on blind comparison. The
objective of model validation is not to use experimental data for model improvement, but to blindly test
the outputs from numerical models with experimental data. Additional critical tasks for standardization
are listed below, and will be used to guide the discussions in this paper and create an initial framework
for the model evaluation procedure.
1.
2.
3.
4.
5.
Develop guidelines for proper documentation and post-processing of field data
Develop procedures for physical modeling to ensure quality data collection
Create database with high quality test data from experimental studies
Using the above database, design evaluation procedures for numerical models
Implement a tool for evaluation to ensure accurate comparison and provide documentation
Notice that this outline includes protocols to ensure proper data collection methods so high quality
datasets will be obtained. Also, proper documentation for experimental studies (i.e., field and wind
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 83/167
Book of Proceedings
tunnel) is required for the data to be useful for numerical model validation. Both will influence the
quality of data used in each step of the model evaluation, and it is recommended that databases of
models and experimental data be created to simplify the record keeping and documentation required
for model evaluation. For the numerical models, this can be done in the format of a modeling inventory
with information collected using a standardized questionnaire with all relevant model information
documented. A data repository is suggested for the experimental data to keep the validation datasets
in one location and to prevent manipulation of the datasets.
The purpose of this paper is to provide model evaluation background information for model developers
and end-users, state the intent to develop validation methods and initiate discussion within the wind
energy community. In the following section background information is given that includes general
terminology, and experimental and numerical background specific to wind energy research. Next, a
general model evaluation methodology is presented based on previous studies that incorporated
numerical modelers and experimentalists in urban fluid mechanics. The methodology is adapted for
use in wind energy applications, with specific attention to the validation procedure. Three test cases
for wind energy are shown to demonstrate the application of the validation procedures for terrain and
wake flow situations. Finally, a discussion and summary are given to provide future recommendations
for the wind energy community to consider when developing experimental databases and
implementing a uniform model evaluation methodology.
2. Background
2.1. Terminology
The following is a list of recommended definitions for commonly used evaluation terms in numerical
modeling. The purpose of this list is to provide clarity when model results are being discussed within
the wind energy community. Defining terms associated with computational simulations is not unique
and references for these definitions can be found in the literature [5, 6, 7, 8]. Note that the list is
ordered alphabetically and a term that appears in a definition may be defined further down the list.
Benchmark: Typically in literature this is defined as an analytical or highly accurate numerical
solution for use in verification [6]. However, this term is often being used to describe experimental
datasets for use in validation, therefore care should be taken when using this term to clarify the
accurateness and purpose of the dataset.
Blind Test: Comparison of numerical results with experimental data, where modelers are not
allowed access to the experimental dataset.
Error: Inaccuracy of the numerical model i.e., insufficient time-step resolution or spatial grid
convergence. This can be known error due to limitations in implementing the mathematical
equations (Acknowledged Error) or unknown error from mistakes (Unacknowledged Error).
Evaluation (Scientific Evaluation): Determining the appropriateness of the conceptual model in
describing the real world application, includes three parts: scientific review, verification and
validation.
Extrapolation: Using a numerical model to simulate a process outside the range of which it was
previously validated.
Conceptual Model: System of mathematical equations, governing laws, initial and boundary
conditions that describe the physical process of interest in the selected real world application.
Computational Model: Implementation of the conceptual model into computer code.
Metric: Variable used to quantitatively compare results from a numerical model with experimental
data, typically with specified criteria for validation.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 84/167
Book of Proceedings
Numerical Calibration: Utilizing field measurements, ensuring the proper scaling and units, as
input parameters to the numerical model that are not a priori known.
Numerical Model: Another term for conceptual or computational model, this term is provided to
distinguish between wind tunnel data and computer simulations.
Physical Model: Non-numerical modeling of a real world process; i.e., using a wind tunnel or water
tunnel to model a real world process to provide a high quality dataset for the validation of
computational models.
Prediction: The output from a validated numerical simulation, for a specific real world process that
is within the modeling capabilities deemed acceptable from the numerical model validation.
Quantity of Interest: Output variable from numerical model to compare directly with experimental
data, the metric is used to quantitatively compare the two results.
Real World: Determination of the physical process to be investigated, examples for wind energy
applications include wind flow patterns and flow around a wind turbine.
Scientific Review (Assessment): The first step in model evaluation, it is an investigation of the
scientific basis of a numerical model, which physical processes are included, how they are
modeled, assumptions, approximations, solution techniques and the interface and resources
available to the user.
Tuning: Making adjustments to parameters in the numerical model based on the comparison
between the model output and field measurements, not considered orthodox validation since it is
not a blind test.
Uncertainty: Recognizable inaccuracies of the model that are not due to a lack of knowledge. This
can be due to inherent variability in the physical process (Aleatory Uncertainty) or from a lack of
scientific understanding (Epistemic Uncertainty). Epistemic uncertainty can be improved by
increasing modeling skill or understanding.
Validation: Ensuring the physical processes are accurately modeled, this involves a comparison of
the computational results with experimental data.
Variability: In this case of wind energy this is the aleatory uncertainty attributed to the irregularity
of turbulent processes in the atmosphere.
Verification: Ensuring the mathematical accuracy of the computational model, including accurate
implementation of equations (Solution Verification) and checking the computer code for errors
(Code Verification).
2.2. Experimental considerations
Improving models for wind energy applications requires knowledge from several academic disciplines
to study the properties of wind in the atmosphere. The mesoscale phenomena are necessary to
evaluate the mean surface winds at a location to justify wind turbine siting, and the microscale physics
are required to determine turbine power output and reliability of the turbine design. Thus, requiring
several types of numerical models and experimental datasets to accurately describe and model the
physical processes that occur in the real world application. Therefore, a fitness for purpose mentality
should be followed to ensure that the model being used is accurate for describing the real world
process. Accompanying this, is the fitness for purpose of the datasets used to compare with
numerical simulations, where data requirements to validate a microscale model differ from those of a
mesoscale model.
As stated in Section 1, it is important to create a database of experimental data for the model
evaluation process. This not only provides a consistent comparison, but also makes certain that the
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 85/167
Book of Proceedings
same datasets are used for comparison and not manipulated prior to the model evaluation process.
Given the complexity of the experimental data with regard to uncertainty, variability and differing
requirements for the various types of numerical models, it is suggested that each dataset undergo
rigorous consideration and justification for inclusion into the experimental database. It is
recommended that the wind energy community set selection criteria to outline the requirements for
accepting an experimental dataset as an evaluation test case.
There are several obstacles that exist in the collection of atmospheric data, the first is the
instrumentation limitations (spatial and temporal), also the post-processing steps required to interpret
the data are not straight forward. To ensure quality datasets are collected during field studies a
general methodology for these post-processing steps should be implemented, as well as guidelines for
experiment documentation to accurately and completely record the field campaign. In addition to
these post-processing and documentation methods, field experiments intended for model validation
would benefit from an experimental design that incorporates numerical models from the beginning.
Experiments designed in this manner will be more complete and therefore more useful for validation
datasets, where several factors must be considered (i.e., experimental requirements, input and
boundary conditions, measurement uncertainty, experiment documentation, etc.).
In an effort to address the meteorological uncertainties associated with measurements for wind energy
applications the European UpWind project designated a meteorology working group. The first
deliverable reports state of the art measurement techniques and methods, including the required
accuracies and sampling frequencies, and resulted in the development of a Meteorology Database [9].
This report also discusses data from wind farms for wake flow investigations, typically from the wind
turbine Supervisory Control and Data Acquisition (SCADA) system collected operationally as a
standard part of wind farm monitoring. The SCADA data is used for wake analysis because the
number of meteorological sensors and towers in wind farms is limited, however several data filtering
steps are required to use the data for model evaluation. Additionally, this data does not always
include flow variables but monitors wind turbine performance (i.e., power and yaw position), requiring
the implementation of two models (flow and turbine performance) before the numerical and
experimental data can be compared, making it impossible to determine which model the uncertainties
are coming from.
While it is recognized that the data from wind tunnel experiments may only be applicable to idealized
atmospheric scenarios, the information is still useful for numerical modelers. The microscale physics
of atmospheric processes can be investigated with higher spatial and temporal resolution data if
physical models are utilized. For example flow in complex terrain, where wake regions exist and
separation can occur, can be modeled in a wind tunnel at a specified scale, to study the fundamental
physics of the flow. This can then be modeled numerically, with simplified conditions (e.g., neutral
stability, no Coriolis force) to verify that the flow physics are modeled accurately. In addition to
complex terrain, wake flow behavior can also be modeled in the wind tunnel. Although, matching all
similarity criteria is difficult and the results should be interpreted with caution since wake flows are
generally Reynolds number dependent.
To obtain a high quality dataset from wind tunnel experiments it is important to understand the
difference between wind tunnel boundary layers and the atmospheric boundary layer (ABL). When
using a physical model to represent real world scenarios, obtaining the correct approach flow is
important. Typically, this is investigated with vertical profiles of the mean velocity. However, this does
not provide an understanding of the turbulence in the approach flow which influences the physical
behavior of the flow phenomena. When characterizing the inlet flow profile it is important to not only
match the mean velocity, but to consider several additional turbulent boundary layer properties; i.e.,
turbulent kinetic energy, integral length scales of turbulence, wind direction fluctuations and spectral
density. To accurately model a real world field study, the wind tunnel model must not only match the
physical scale, but also the scale of the boundary layer turbulence. A good reference for these
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 86/167
Book of Proceedings
considerations and their application to an urban environment can be found in Schatzmann and Leitl
[10]. The approach flow characteristics are also important to quantify as they are used as inputs in
numerical models.
To properly model the ABL, vortex generators and surface roughness
modifications are typically used to obtain known approach flow conditions obtained from field data.
The dimensions of the wind tunnel will dictate the boundary layer scaling relationship, and the ability to
properly model the scales of motion in an ABL flow.
There are also difficulties in wind tunnel modeling when applying the proper geometric and dynamic
scaling to physically model the real world process. This is especially difficult in wind energy
applications when trying to model wind turbine wake behavior and power output. Due to these
difficulties it is increasingly important to develop quality assurance methodologies for wind tunnel
modelers to use in their experimental design. Not necessarily in regard to the exact setup of the
experiment, but in the experiment documentation, justification for scaling and inlet flow considerations
and quantification of the measurement uncertainties.
2.3. Numerical modeling
The simplest of the numerical models used for wind condition assessment are linearized flow models,
where the equations governing the physical process of the wind (flow) are assumed to behave linearly.
While these models are extremely simplified in describing the physical processes, they allow for
cheaper computational costs and for mean surface wind predictions over large areas, typically up to
10 km at 50 m horizontal resolution. The linearized flow model currently being used in the wind
energy industry is WAsP, and has been used since the introduction of the European Wind Atlas in
1989 [2]. The WAsP model was inspired by the linear theory of Jackson and Hunt [11] and can be
used reliably in neutral atmospheric conditions over mild terrain, with sufficiently gentle slopes to
ensure fully attached flows. Nevertheless, due to its simple usage and the increasing experience of
the users with the model, WAsP is often used out of its range of applicability.
With the increase in computational capacity there have been several advancements in the use of CFD
solvers to investigate the more complex processes in the real world. These CFD simulations account
for the non-linear behavior of the flow equations and can be used to account for thermally driven
effects in the atmosphere. While the application of these CFD models to atmospheric flows provides
more detailed investigations of the complex processes, the model may not be accurately simulating
the atmospheric behavior due to the simplifications and approximations necessary to solve the
governing equations. Therefore, iterative steps for model development, that are constantly evolving to
incorporate the best practice guidelines for CFD model development, are recommended to ensure the
accuracy of using these complex fluid mechanics models to simulate atmospheric flow phenomena
[12].
These models can be divided into the following categories: Reynolds Averaged Navier-Stokes (RANS)
which models ensemble statistics, Direct Numerical Simulation (DNS) which resolves all eddies and
Large Eddy Simulation (LES) which is between RANS and DNS and requires a comparison of time
series data and statistical means. Some LES studies have been done for wind energy, but in rather
small sites and for academic purposes only [13, 14]. While the CFD models based on RANS [15, 16]
are being developed for wind resource assessment to complement linear models in complex terrain
and other complex flow situations (wakes, forests, obstacles, etc). Therefore, the application of CFD
in wind resource assessment is still largely based on RANS turbulence models because LES remains
far more costly. For both, the phenomena investigated is the wake behind wind turbines and flow in
complex terrain, where LES requires a more detailed time resolved experimental dataset for
validation.
The operational use of CFD in wind resource assessment has occurred in the last 10 years, and there
is currently a large variety of commercial and research models available. The transition from traditional
linear models to CFD requires significant training and experience from the user due to the extended
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 87/167
Book of Proceedings
degrees of freedom of the CFD solver, making the results highly dependent on the user. To overcome
this difficulty, commercial CFD software developers are designing user friendly interfaces that can
emulate, to some extent, the traditional way of working with linear models. On the contrary, research
CFD models, based on generic commercial CFD solvers, in-house or open-source codes, are used by
researchers for the ease of implementing site specific topographic and atmospheric conditions. This is
especially useful when generating the mesh of the computational domain, a pre-processing step of
utmost importance in any CFD simulation with complex geometries.
As with the modeling for wind conditions, wake modeling for the flow behind wind turbines originated
in the 1980s [17, 18]. These models, which are still widely used for wind farm layout today, are based
on simple momentum and fluid dynamic similarity theories, or simplified solutions to the Navier-Stokes
equations. The problem with these models is that they lack many of the required physical processes
needed to predict wind turbine wake behavior, resulting in the under prediction of wake losses by 10%
in many operational wind farms. Wake model development has focused on two areas that influence
wake creation and propagation: turbine models and models for the surrounding atmospheric flow. The
atmospheric flow models largely follow the list of models used to predict the wind conditions described
above.
There are many different varieties of turbine models, with a range of complexity for calculations at
different scales. Thorough reviews of different turbine wake models can be found in Crespo et al. [19],
Vermeer et al. [4] and Sanderse et al. [3]. The simplest model consists of a drag element (typically an
actuator disk) that extracts momentum and injects turbulence over a few simulation grid points. These
models are often used in mesoscale models with larger domains to determine macro influences of
large wind farms. They are also the most widely used approach for far-wake modeling within wind
farms due to the low requirements in terms of input data, all disclosed by the manufacturer. The next
model in terms of complexity, is the blade element momentum (BEM) models that calculate blade
forces and the wake influence using a global momentum balance. BEM methods require knowledge
of the blade lift and drag coefficients, information which is typically not provided by the manufacturers.
To overcome the lack of input data, reverse engineering methods are used to infer these coefficients
by matching the power curve from flow modeling of a reference turbine model [20]. The forces in
these models are then distributed around a disk, and the influence of axial and rotational momentum is
propagated into the wake. Such a model can also be coupled with a wake meandering model that
predicts the unsteady oscillation of the wake as it moves downstream. As turbine models get more
complicated, the details of the blade aerodynamics become more prevalent. Recent calculations of
multiple turbine interactions have used actuator line methods, where the blades are treated as airfoils
distributed along rotating lines. Various other inviscid calculations of blade aerodynamics can also be
used, such as panel methods and boundary element methods, that directly calculate the blade forces
instead of using airfoil lookup tables.
With the need to calculate viscous aerodynamics of the blades, researchers have moved toward CFD
modeling. As with the models used for wind assessment, researchers have utilized RANS, unsteady
RANS, DES (detached eddy simulations, a hybrid between RANS and LES) and full LES of the
rotating blades. Researchers have also created computational domains where the rotor plane is
treated as a viscous area and the downstream region treated as inviscid, which can lead to significant
computational time savings. Although, typically the more detail contained in the turbine model, the
smaller the simulation due to constraints of computing resources.
2.4. Literature Review of Existing Experimental Data
Test cases for the evaluation of wind farm models comprise a wide variety of flow, topographic and
wind farm layout and wind turbine conditions characteristic of the complex system of a wind farm. A
hierarchy of increasing modeling complexity is required in order to approach model evaluation in a
systematic way, trying to focus on unit and subsystem elements of the model chain before evaluating
the full system.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 88/167
Book of Proceedings
Tables 1 and 2 present a non-exhaustive but rather complete list of test cases for model validation that
are documented in the literature. The dates refer to the publication of the reference, typically some
years after the measurement campaign. The test case list is divided in two categories: Wind for wind
turbine wake-free experiments where the objective is the characterization of the background wind
conditions of the atmospheric boundary layer (Table 1), and Wake for the characterization of the flow
conditions within wind turbine wakes and the associated power deficit in wind turbine arrays (Table 2).
The test cases are subdivided in three classes: theory, wind tunnel and field experiments. Scaling
theories for atmospheric boundary layer and wake flows is a good starting point in order to validate the
fundamental physics of the system. Then, wind tunnel experiments are the logical choice for exploring
unit elements of the model chain since they allow high temporal and spatial resolution measurements
on idealized geometries within a controlled environment. Parametric testing in the wind tunnel is very
useful for sensitivity analysis on key flow drivers like, for example, turbine spacing or hill geometry.
This is used to characterize specific aspects of the model chain and allow a more comprehensive
approach to field measurements, where the boundary conditions are not controlled and the
experiments are more limited and costly.
The wind test cases are generic to any boundary layer meteorology model and, in this context, they
aim at characterizing the wind conditions in sites suitable for wind energy development. Hence,
relevant topographic elements are: flat terrain onshore/offshore, complex terrain (hills and mountains),
roughness changes and forest canopies. Concerning wake test cases, the aim is to reproduce the
modified wind conditions behind a wind turbine or within a wind farm array embedded in an
atmospheric boundary layer. Both wind and wake test cases are defined for a specific atmospheric
stability class which determines the turbulence structure of the flow.
The cases presented here are especially suited for the evaluation of wind resource assessment
models. Aerodynamic models require further instrumentation of the wind turbine which is normally not
included in operational wind farms. Resource assessment models typically simulate steady-state
conditions which are then compared with ensemble averaged observations. Hence validation data is
generated by binning meteorological data and wind turbine SCADA data versus wind direction, wind
speed and atmospheric stability at a reference site where the velocity is undisturbed (free-stream
velocity). Concerning wake models, the SCADA data from the wind farms is used to calibrate the
operational power curve of each turbine and extract the power deficit with respect to a reference wind
turbine which is not affected by wake effects. Unsteady models are used to reproduce some
dynamical aspects of the wind conditions such as the diurnal cycle of atmospheric stability or the
meandering of the wake behind a wind turbine.
3. Model Evaluation Methodology
The procedure for this was adopted form the general findings from the COST Action 732, which
determined model evaluation guidance and protocols for urban dispersion modeling [21, 22]. While
the objectives for modeling wind energy differ from those of urban dispersion, a similar methodology is
used to develop the physical and mathematical models. Therefore, the procedures outlined in COST
732 can be utilized as a starting point in the development of the quality assurance protocols for wind
energy models. Examples of validation for air quality and urban dispersion models can be found in
Chang and Hanna [23] and Schatzmann et al. [24]. In both cases, the validation procedure requires a
high quality experimental dataset, including quantification of the uncertainty, for comparison and
suggests that the computational result be within recommended quantitative criteria. These procedures
provide a method for modelers to utilize that improves simulations and provides documentation for the
reliability of the modeled results.
A significant difference between the evaluation procedures for urban dispersion models and wind
energy is the critical parameters being modeled. In wind energy scalar transport is not the process of
interest. Instead, power generation from a wind farm or turbine is the variable of concern to determine
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 89/167
Book of Proceedings
the cost associated with implementing wind energy in a specific scenario. The evaluation of models
must be adapted appropriately for wind energy modeling and care should be taken when selecting the
primary variables of interest. Therefore, it is critical that a consensus be made within the community to
direct new research and ensure appropriate mathematical and physical models are being used.
Table 1: Test cases for flow-over-terrain (wind) model evaluation. (V: virtual/theory; F: field; L: wind tunnel; U:
unstable; N: neutral; S: stable; E: flat (empty); H:hilly; C: complex terrain; R: roughness/forest; T0: no wind turbines;
T1: single rotor; T2: line of turbines; T3: several lines of turbines).
Test Case
Wind
Stability Topography
Wake s
Reference
Date
Description
V
F
L
U
N
S
E
H
C
O
R
T0
T1
T2
T3
Exp
Theory (Idealized)
Classical theory for 1D Steady-state surface
boundary layer models in flat terrain
Stable ABL in flat terrain, 9hr of uniform surface
cooling and gesotrophic forcing with no radiation.
Verification with LES simulations
Monin-Obukhov
x
x x x x
x
Monin and Obukhov [40]
1954
GABLS1
x
x x
x
Cuxart et al. [41]
2006
x
Kim et al. [42]
1997
2D sinusoidal hills with different slopesand heights
Wind Tunnel
POSTECH 2D
isolated and doble
hills
x
x
CSIRO homogeneous
canopy model
x
x
x
x x
Brunet et al. [43]
1994
Model of waving wheat crop. X-wire measurements of
statistical moments and spectra
VKI 2D Forest
Clearing
x
x
x
x x
Sanz Rodrigo et al [44]
2007
2D forest clearing inside homogeneous 'foam' forest.
2D PIV measurements of the flow in the clearing for
3 different upstream canopy fetchs and porosities
CSIRO 2D Furry Hill
x
x
x x
Finnigan and Brunet [45]
1995
EnFlo 2D Stratified
Hill
x
x x
x x
Ross et al. [46]
2004
Field
Leipzig profile
Bradley's Roughness
Change
x
Letau [47]
1950
Bradley [48]
1968
x
Svensson et al. [49]
2011
x x x x
x
Verkaik and Holtslag [50]
2007
x
x x x x
x
Gryning et al. [51]
2007
160m mast. ABL in flat terrain on a coastal site
x
x x x
x
Sanz Rodrigo [52]
2011
100m mast. ABL in offshore open-sea conditions
Askervein smooth hill
x
x
x
x
Taylor and Teunissen [53]
1987
Cooper's Ridge
x
x x x
x
x
Coppin et al. [54]
1994
Bolund complex hill
x
x x x
x
Berg et al. [55]
2011
x x
Chen et al. [56]
1995
x x
Dellwik et al. [57]
2009
GABLS2 - CASES99
Cabauw onshore
mast
DTU Høvsøre coastal
mast
Alpha Ventus Fino1
offshore mast
x
x
x
x
x
x
x
x x
x
x x x x
x
2D polynomial hill with a slope of 0.36 covered by
waving wheat crop model of Brunet et al. (1994).
Vertical velocity/turbulence profiles along the hill
2D sinusoidal hill with slopes 0.3 and 0.7, covered
with roughness elements, in neutral and stablystratified conditions
UBC 2D Forest Edge
Falster 2D Forest
Edge
x
x
x
x x x
WAUDIT
Contract#
238576
x
x
x
Classical reference for neutral 1D ABL in flat terrain
Smooth (0.002m) to rough (0.25m) transition along
an array of 8x1m masts over 15m
Diurnal cycles for three days from the CASES99
campaign, no radiation
200m mast. ABL in flat terrain surrounded by
heterogeneous roughness
Isolated 126m-high gentle hill in neutral conditions.
50x10m mast along three arrays, 3x50m masts in
upstream reference and hill-top
Nearly-2D smooth ridge. Array of 7x4m and 2x30m
masts along the windward side of the hill
Isolated 12m-high and 150m-long hill with
escarpment, 10x9m masts equiped with sonic and
cup anemometers
Measurements downwind of a homogeneous forest
edge
2x45m masts and Lidar before and after a
homogeneous forest edge with different
winter/summer LAI profiles and stabilities
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 90/167
Book of Proceedings
Table 2: Test cases for wind turbine wake (wakes) model evaluation. (V: virtual/theory; F: field; L: wind tunnel; U:
unstable; N: neutral; S: stable; E: flat (empty); H:hilly; C: complex terrain; R: roughness/forest; T0: no wind turbines;
T1: single rotor; T2: line of turbines; T3: several lines of turbines).
Test Case
Wakes
Stability Topography
Wake s
Reference
Date
Description
V
F
L
U
N
S
E
H
C
O
R
T0
T1
T2
T3
Exp
Theory (Idealized)
Self-Similar Turbulent
x
Circular Wake
x
x
x
Wygnanski [58]
1986
Classic scaling laws of turbulent wakes
x x x
x
Chamorro and Porté-Agel [59,
60]
2009
Model wind turbine of 0.15m diameter and tip-speedratio of ~4, X-wire measurements on vertical planes
at x/D=[3,5,10,15], smooth and rough incoming
boundary layer in neutral and stable conditions
x x Chamorro and Porté-Agel [61]
2011
Model wind farm array of 0.15m diameter turbines in
aligned and staggered configuration with x/D=4 and
y/D=5 spacing. Measurements down to x/D=50
Wind Tunnel
UMN Single wind
turbine, full-rotor
model
x
UMN Wind farm
array, full-rotor model
x x x x x
PRISME Single wind
turbine, actuator disk
model
x
x
x
x
España et al. [62]
2011
Model wind turbine of 0.1m in neutral ABL using
porous mesh (no rotation). 2D LDV measurements
of vertical planes at x/D=[2,4,6,8,10] for different
mesh solidities
x
Cleijne [33]
1992
Onshore flat, 1x300kW, 30m diameter, 35m hub
height, met masts at x/D=[-2.8,2.5,5,8] for velocity
deficit and added turbulence intensity
Machielse et al. [63]
2007
Onshore flat, 5 x Nordex 2.5MW/80, 80m hub
height, 3.8D spacing, 2x100m + 1x108m met masts
Politis et al. [64]
2010
Field
Sexbierum single
wake experiment
x
x
x
Wieringermeer Test
Wind Farm (EWTW)
x
x
x
UpWind complex
terrain wind farm
x
x
Vindeby offshore
wind farm
x
x
Middlegrunden
offshore wind farm
x
Horns Rev offshore
wind farm
x
x x x
x
Nysted offshore wind
farm
x
x x x
Lillgrund offshore
wind farm
x
Egmond aan Zee
offshore wind farm
x
x
x
x
43 turbines, 48.4m diameter, 45&55m hub height, 5
lines ~13Dx1.5D, 1 met mast
Offshore(coastal) 11 x Bonus 450kW/35.5, 38m hub
height, 2 lines. 2 met mast + 1 coastal met mast +
ship-mounted SODAR
Offshore 20 x Bonus 2MW/76 in arc-layout, 64m hubheight, 2.4D spacing
x
x
Barthelmie et al. [65]
2003
x
x
2007
x
Hansen et al. [67]
2010
Offshore 80 x Vestas 2MW/80, 70m hub height,
7Dx7D matrix, 1 upwind and 2 downwind met masts
x
x
2011
Offshore 72 x Bonus 2.3MW/82, 69m hub height,
10.5Dx5.8D matrix, 1 upwind and 2 downwind masts
x
x
x
Dhalberg et al. [68]
2009
Offshore 48 x Siemens 2.3MW/92.6, 65m hub
height, 8 rows 3.3Dx4.3D with opening in the middle,
1 met mast
x
x
x
Eecen et al. [9]
2011
Offshore 36 x Vestas 3MW/90, 70m hub height,4
rows, 1 met mast. Study of added turbulence
intensity from single to triple wakes for x/D = 4 to 24
The concept of comparing numerical simulation results with experimental data is not novel, and
therefore it is advantageous to use previously developed evaluation procedures and apply them to
wind energy modeling. The primary sources for this work, specific to fluid mechanics or air quality
applications, include the following papers: American Institute of Aeronautics and Astronautics (AIAA)
guidelines [25], Oberkampf and Trucano [6], Schatzmann and Leitl [7] and Britter and Schatzmann [8].
Where a previously suggested six step methodology for model evaluation outlined in the findings from
the COST Action 732 will be used as a starting point [21, 8]. Below the six steps are listed and briefly
summarized, where the reference to model implies a numerical simulation.
1. Model Description
Brief description of the model and the purpose for which the model was developed. The theoretical
background of the model should be presented, including the limitations, assumptions and
applicable range. If the model is derived from experimental data this should be mentioned,
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 91/167
Book of Proceedings
explicitly stating the dataset used for development so it will not be used for evaluating the model
performance.
2. Database Description
Description and justification of the experimental datasets chosen to be compiled into an accessible
database for numerical modelers. The datasets selected to make up the database should be of
high quality, and must follow quality assurance methods for data collection and processing steps.
The uncertainty and variability in the data should be estimated and included in the database.
3. Scientific Review
Ensuring the equations used to describe the real world physical process are acceptable. This
includes a description of the equations, explanation and justification of the numerical model and its
limitations with respect to applications of use.
4. (Code) Verification
Evaluation of the implementation of the equations into a computer model. This step is done to
ensure there are no mistakes in the computer code or inaccuracies in the mathematical equations.
5. Model Validation
Comparing the model outputs with experimental data in a quantitative statistical manner. This step
is intended to determine how well the numerical model predicts the real world process. Quantities
of interest should be chosen, along with the validation metrics to quantitatively determine the
performance of the numerical simulation.
6. User-oriented Assessment
Providing documentation and information on the numerical model so an individual user is able to
read about the model development and applicability. This includes material that allows the user to
install, setup and operate the numerical code or software. Additionally, training or best-practice
guidelines for using the model are included in this step.
All six of these steps are necessary for the proper evaluation of a numerical model. Of the six steps
listed the most difficult and ambiguous is Step 5, Model Validation. This step does not have explicit
requirements, and can be approached with several questions regarding the criteria and methodology
to validate a numerical model. Determining the quantities of interest are the first priority, then deciding
if these should be considered as time or space dependent quantities follows. Along with the space and
time dependence, the question of averaging times and locations for the comparison arise. Once the
quantity of interest is identified and the appropriate spatial, temporal and averaging method used for
the comparison are selected the steps for statistical comparison are implemented. Again, this may
seem straight forward, but there are several statistical methods to compare the different quantities of
interest. These are the validation metrics, and are necessary to create a uniform method for
comparing numerical results with experimental data. As part of a validation procedure these metrics
must be clearly defined, and be accompanied with quantitative criteria for determining the success of
the numerical simulation. These validation metrics and criteria should not be viewed as ranking or
excluding numerical models, but rather as a way for the modeling community to compare the model
outputs in a uniform manner.
It is important to mention that there are several sources for error and uncertainty to arise throughout
the evaluation methodology. In every step of the evaluation procedure uncertainties influence the
accuracy of the model validation. The numerical model itself has uncertainty in the ability of the model
to accurately describe a real world process. The data used to develop the models, as inputs to the
model or for the model comparison, also have uncertainty due to limitations in the measurement
equipment or experimental setup. Finally, uncertainties can come from the variability inherent in the
physical process being modeled, for wind energy this is due to atmospheric processes and turbulence.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 92/167
Book of Proceedings
3.1. Validation Procedure
A general validation procedure is presented in this section, with suggestions for each step specific to
wind resource assessment models. Additionally, the use of numerical simulations in experimental
design are expected to be incorporated into this procedure as shown on the right hand side of Figure 1
in the schematic illustrating the suggested validation path. To begin, it is important to define the
quantities of interest and the proper metrics for validation so they are the most useful. This can be
referred to as the validation objective, and should be clearly defined prior to conducting the model
evaluation study. The quantities of interest and validation metrics given here are provided as an
example. It is expected that once the wind energy community defines the validation objective the
appropriate quantities of interest, validation metrics and acceptance criteria will be determined. The
following is a brief summary of each step required to develop a model validation methodology, and an
example for applying these steps to flow in complex terrain is given in the following section, as well as
two examples of wake model validation.
Figure 1: Suggested flow chart for model validation for turbulence models used in numerical simulations (CFD) for
wind energy.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 93/167
Book of Proceedings
Step 1. Determine Quantity of Interest
•
•
•
•
•
•
•
Wind Speed
Wind Direction
Turbulent Kinetic Energy
Turbulence Intensity
Power Output
Mean Quantities
Extreme Events
Step 2. Important Considerations Required for Validation
•
•
•
•
What time average should be used for comparison?
If extreme events are considered, what frequency of occurrence should be considered?
How many spatial locations are required for each case, where should they be?
Vertical profiles are modeled, what should experimentalists measure (difficult in field with
traditional instrumentation)?
Note: These considerations will greatly influence experimental data collection and should be
thoroughly considered when using experimental data as validation test cases. While
limitations exist in wind tunnel modeling, the data for comparison allows for a better spatial
and temporal resolution of experimental data. Therefore, these considerations should be
evaluated in the context of both field and wind tunnel experiments.
Step 3. Plot Quantities of Interest (Initial Comparison)
While plotting the quantities of interest from the numerical output and experimental data for
visual comparison is commonly used for validation and is a necessary tool for the first data
comparison, this type of investigation has no statistical relevance.
Step 4. Generate Plots to Visualize Comparison (Qualitative Comparison)
Scatter Plot: correlation between experimental data and model outputs for a quantity of
interest.
Quantile-quantile Plot: correlation between the probability distributions of a quantity of interest
Residual Plot: model performance (ratio of predicted to observed) as a function of independent
variables (i.e., stability, time of day, wind speed) to investigate the influence of these variables
on the results
Step 5. Use Validation Metrics for Statistical Comparison (Quantitative Comparison)
The following are commonly used validation metrics, where (P) denotes the output from a
numerical model and (O) is the observed value from experimental data. Once quantities of
interest are determined these equations should be reformulated to include the selected
variables. The overbar represents an average of that quantity and σ is the standard deviation.
Correlation Coefficient (R):
R=
(O − O )(P − P )
Fractional Bias (FB):
FB =
WAUDIT
Contract#
238576
σ Pσ O
(O − P )
0.5 (O + P )
(1)
(2)
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 94/167
Book of Proceedings
Normalized Mean Square Error (NMSE):
2
O − P)
(
NMSE =
(3)
OP
Geometric Mean (MG):
(
MG = exp lnO − ln P
)
(4)
Geometric Variance (VG):
2
VG = exp  (lnO − ln P ) 
(5)


Fraction of predictions within a factor of two of the observations (FAC2):
P
0.5 ≤ ≤ 2.0
(6)
O
Hit Rate (HR): Normalized quantities of interest are compared, indicated by the subscript i in
the equation below. Where, D is the allowed range from comparison data and accounts for
data uncertainty and W a threshold value that describes the repeatability of the experimental
data.

Pi − Oi
≤ D or Pi − Oi ≤ W
1 n
1 for
HR = ∑ Ni with Ni = 
(7)
Oi
n i =1

0 else

A perfect model would have R, MG, VG, FAC2 and HR equal to one and FB and NMSE equal to zero.
Also, the list presented here does not include all possible metrics and other measures that quantify
model performance should be considered. Determining validation metrics is not a trivial task and
several criteria must be established and met to ensure that the chosen metric is useful. Several key
components of a validation metric can be found in Oberkampf and Barone [26]. Mainly, the metric
should serve as a key variable in quantifying the agreement between the numerical simulation and
experimental data for the particular real world process of interest. In addition, the metrics should take
into account different sources of error and uncertainty that arise during the evaluation procedure.
Just as the fitness for purpose method applies to the experimental data used for the comparison, a
similar mentality should be applied to the selection of validation metrics. Each metric has a different
physical meaning and therefore will be a different indicator of the error or uncertainty in the numerical
model. For example, the FB and NMSE are representative of the mean bias and scatter of the data
based on a linear scale, where the MG and VG represent the same errors but for a logarithmic scale.
This implies that for datasets with extremely high or low values there will be a different influence on
each of these metrics. The linear scale metrics (FB and NMSE) are strongly influenced by seldom
occurring high values, while the logarithmic scale metrics (MG and VG) are influenced by low values
and undefined for a value of zero. Therefore, when selecting the metric that determines the
performance of a numerical model it is important to consider its physical relevance. The FAC2 metric
is typically considered a robust measure of model performance because it is not influenced by these
extreme high or low values [27]. Another metric that is not influenced by these extreme values is the
HR, which also has a threshold criteria directly implemented into the definition of the metric (D and W
in Equation 7).
A similar strategy, based on the physical relevance and interpretation of the metric, can be applied to
the visual metric plots (Step 4). These plots can be configured to include information about the
variables of influence or trends that exist in the comparison between predicted and observed data. By
plotting the residuals of scatter plot data points a structure or trend in the data can appear that was not
previously seen by visual inspection [28]. Another, more detailed residual analysis can be done using
box plots and plotting the residual data as a function of independent variables [27]. When plotted in
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 95/167
Book of Proceedings
this manner the data comparison can be done in a way that allows for the visualization of how these
independent variables influence the result. An example applicable to the wind energy community is
ABL data, where an investigation can be done with respect to atmospheric stability. The residual box
plot can be made to investigate the model performance as a function of boundary layer stability stable, unstable and neutral. This is important for wind energy where a significant amount of model
development is done for the case of neutral ABL, but the next phase of research includes scenarios
with more complex ABL stability.
The final step in the selection of validation metrics is establishing criteria, or a threshold value that
must be met for a numerical model output to be considered a prediction. While selection criteria are
not yet recommended in this work, the physical interpretations of the metrics presented above are
given in Table 3. The information in Table 3 shows that each metric provides a quantitative measure
for different physical attributes of the data comparison. Again, the physical meaning of the validation
metric criteria must be considered before establishing the threshold limit. Factors that play an
important role in the criteria are the uncertainty and variability in the experimental data, validity of the
threshold criteria to the real world application and realistic capabilities of the model application.
Table 1: Interpretation of the quantitative values for validation metrics presented in Equations 2 through 7
4. Example Cases - Terrain and Wake Flow
This section will illustrate the implementation of the Validation step in the model evaluation
methodology presented in Section 3. The first four model evaluation steps, Model Description,
Database Description, Scientific Review and Verification, will be discussed briefly. However, it should
be noted that the purpose of this work is to demonstrate the need for a thorough, statistically relevant
validation procedure, therefore the focus of this section is on the validation procedure and not the
entire model evaluation methodology. In the following sections three validation examples will be given
for three flow cases, terrain flow, single turbine wake and offshore wind farm wakes, where a blind
comparison is done using experimental data and results form one to three numerical models. The
references for the specific model used in each of the examples are given in the corresponding section,
ideally the reference includes the Model Description, Scientific Review and Verification as it is part of
the model development process. Where a more standardized documentation approach for these
three steps is often used when conducting a large investigation that includes results from several
numerical simulations [8, 22, 29].
The data used in each of these examples comes from field experiments with previously published
data, therefore the Database Description in these cases is in the related reference. The importance of
the quality of data used for validation is noteworthy, and it is necessary to quantifying uncertainties in
the experimental data for an accurate validation. In the case of the terrain flow case, the purpose of
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 96/167
Book of Proceedings
the field study was to collect data for comparison with numerical results, hence the data include error
bars as a quantification of the measurement variability. While for the single turbine case there are no
error bars making it an undesirable dataset for validation because it is impossible to determine
whether or not the numerical results are within the limits of the measurement uncertainty or variability.
It is still used in this work to provide an example of the validation procedure for wake flow behavior,
however it is recommended that a dataset with error bars be used in the future.
4.1 Validation Example: Terrain Flow
The following is an example of how to apply the above procedure to a test case using data from the
Bolund blind comparison study [29]. The Bolund study site was chosen by researchers at the
Technical University of Denmark (DTU)-Wind Energy to provide field data to improve flow modeling in
complex terrain, therefore flow over an isolated hill was the focus of the investigation. The east-west
elevation profile of the Bolund hill is shown in Figure 2, with the numbers along the top denoting the
locations of meteorological towers for measurements. For the validation example discussed here only
westerly winds are investigated, or flow moving from left to right in the figure.
Figure 2: Profile of Bolund hill showing elevation and monitoring locations for meteorological data, more detail
regarding the locations and Bolund experiment can be found in Berg et al. [55] and Bechmann et al. [29].
Step 1: Identify quantities of interest
In this case the mean wind speed and turbulent kinetic energy (TKE) are considered because they
were the quantities of interest chosen for the Bolund blind comparison. The mean wind speed in this
example is represented as the total mean wind speed obtained from vector scaling (S = (U2+V2+W 2)1/2,
where a capital letter denotes an average) and accounts for the wind speed in all three directions.
Step 2: Important considerations
For the Bolund example these considerations were already selected by the researchers conducting
the blind comparison: 10-min averaged field data and only time periods with a neutral atmospheric
boundary layer. The 10-min periods that met the criteria for wind direction (westerly, 270º ± 8º) and
stability were ensemble averaged for the comparison, therefore the focus is on a comparison of mean
data and not extreme wind events. Data collected at two heights from four meteorological masts
located on the hill centerline, with respect to westerly winds, were used for the comparison (M7, M6,
M3 and M8 in Figure 2). Additional towers, not on the same westerly center line, were determined to
be representative of the free wind conditions (M0 and M9) providing two additional points for
comparison. Thus, the spatial resolution for the data comparison is nine data points, where the two
measurements at M0 are used as reference data for input to the numerical model, hence they are not
considered validation data.
Step 3: Plot quantities of interest
Two quantities of interest are plotted for this analysis, shown in Figure 3 for the mean wind speed
represented as the fractional speed-up ratio (FSR = (S-S0)/S0) and in Figure 4 for the TKE. The data
point in front of the hill (x=-180m) is from the reference mast (M0), and the measurements at all
locations are normalized by data at this location (denoted with subscript 0). Two heights are shown in
the figures, 5 m in the top figure and 2 m in the bottom figure. For this analysis data from sonic
anemometers are utilized to allow for an investigation of the turbulence using TKE as the quantity of
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 97/167
Book of Proceedings
interest. For the wind speed comparison, data from three numerical models are shown, two
computational fluid dynamics (CFD) models and one linearized flow model. The CFD models are both
from the same model, with two different versions, one was developed as a surface layer model
(CFDWind 1.0, [30, 31]) and the other for the ABL (CFDWind 2.0, [30, 32]). The linearized flow model
is WAsP, where meteorological records are used at one location to predict the mean surface wind
speed at another specified location [2]. The WAsP model is considered suitable for modeling flow in
mild terrain, without flow separation, and is currently the standard model used in the wind energy
industry, therefore the results are included here for comparison purposes.
Figure 3: Quantity of interest plot showing the fractional speed-up ratio (FSR = (S - S0)/S0) for the Bolund field data
and three numerical simulations, top: z = 5m and bottom: z = 2m.
Figure 4: Quantity of interest plot showing the turbulent kinetic energy (TKE) for the Bolund field data and two
numerical simulations, top: z = 5m and bottom: z = 2m. Note: The top and bottom axis are plotted on different TKE
ranges for the y-axis.
For the mean wind speed (Figure 3) it appears that both CFD models give similar results for the flow in
front of and over the hill, while in the wake region the two models differ. A visual inspection of the
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 98/167
Book of Proceedings
model output compared to experimental data indicates that the model outputs for the 5 m height are
better than for 2 m, however in the region just behind the hill there are no experimental data points for
comparison. While WAsP seems to give reasonable values of wind speed for the incoming flow, once
the Bolund hill is encountered the numerical results differ significantly from the experimental data. All
models give a similar profile for the mean wind speed, with the speed up and wake regions occurring
at similar locations.
The TKE plot (Figure 4) does not show data from WAsP because it does not solve turbulence in the
simulation. Where the CFD models use a model for the turbulence to replace the parameterizations
typically used in linearized flow solvers. Note, that the y-axis scale used to represent the TKE in the
figures is different and can be misleading when comparing data between the two plots. It appears that
the numerical models give TKE values similar to the experimental data just in front of the hill (x ~ -80
m) at both heights, behind the hill (x ~ 100 m) at the 5 m height and does not match well with the
experimental data at all other locations. This investigation is important to ensure that the model
outputs are at least somewhat reasonable to the observations, and it allows for a spatial investigation
of where the simulation over or under predicts. However, a visual inspection does not provide
quantitative information required for a precise model validation.
Figure 5: Scatter plot comparing mean wind speed (S) for Bolund field data with three numerical simulations, the
correlation coe_cient for linear regression (R2) is also shown for each numerical model at two heights.
Step 4: Plot the visual metrics
The next step provides a more statistical comparison between the numerical output and experimental
data, as shown in the scatter plot (Figure 5) for the total Bolund mean wind speed (S). The first thing
to notice in the plot is the low number of data points, meaning that the statistical significance of the
comparison is low. This is a result of using field data for the comparison, here only nine data points
can be directly compared for the numerical output and experimental data. Next, the correlation
coefficient (R, R2) can be used as insight into how well the linear regression fits between the model
and observations, with one being a perfect correlation. In the case of wind speed, both CFD models
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 99/167
Book of Proceedings
indicate good linear correlation with high values for R2, while the R2 values for WAsP are low. This is
a good first step to statistically compare the information, but the correlation coefficient does not give
detailed information about the spread of the data and should not be used as a sole determination of
how well a numerical model performs.
Plotting the quantities of interest (Step 3) and this type of scatter plot are the two most often used tools
currently being utilized for model evaluation. These two plots alone are not enough to ensure an
accurate comparison and provide minimal statistical basis for the model evaluation. Additionally, when
more data points are used visual plots with more statistical detail can be used to investigate the
differences between the numerical model and observations (see Section 3.1, Step 4).
Step 5: Compute the validation metrics
For a quantitative statistical comparison between the numerical models and experimental data,
validation metrics are computed. Not only does this allow for a model to observation comparison, it
also provides a uniform way of comparing the models so they can be evaluated side by side in the
same context. Five validation metrics are computed for the three numerical models, using two
different quantities of interest, shown in Table 4. These metrics show the importance of selecting the
proper quantity of interest and metric for a specific application. For example, both CFD models
behave similarly based on the metrics for S, but for TKE the CFDWind1 model performs better
according to the FB and NMSE, while according to the FAC2 the CFDWind2 model is better.
Therefore, the selection of these quantities of interests and validation metrics is not a trivial task, and it
is critical to reach a consensus for the selection of both in the wind energy community.
Table 4: Validation metrics to quantitatively compare the performance of three numerical models
with data from the Bolund field experiment.
While the results for the two CFD models are similar to each other, the metrics for WAsP are
significantly different from those of the CFD models when calculated for the mean wind speed. By
comparing the validation metrics calculated for the CFD and WAsP models an example is provided
that shows how the statistical quantities can improve the validation analysis and go beyond a visual
comparison. Based on the metrics calculated for S (Table 4), the two CFD models have a mean bias
of ~15 % based on a linear scale (FB) and ~10 % for the logarithmic scale (MG). With random scatter
significantly less than a factor of 2 of the mean for both scales (NMSE and VG) and a perfect metric
calculated for FAC2. The WAsP model, as expected from the visual inspections in Steps 3 and 4,
does not perform as well according to the metrics with 35\% and 27\% mean bias for the linear and
logarithmic scales. The scatter for both scales is still within a factor of 2 of the mean, but the
percentage of numerical results within a factor of two of the observations is only 89 % based on FAC2,
and not 100 % like the CFD models.
Validation Example: Single Wind Turbine Wake
Moving from terrain flow to a smaller scale of investigation, the following example will follow the same
validation procedure for the flow in the wind turbine wake, measured at three locations downstream of
the turbine. The purpose of the Sexbierum field experiment was to analyze the near and far wake
behavior downwind of a three-bladed, 300 kW wind turbine with a hub height of 35 m and rotor
diameter of 30 m [33]. The data was processed and binned to be used for comparison with numerical
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 100/167
Book of Proceedings
simulations for wake behavior, however there are no data uncertainties or quantification of the
variability.
For the case of a single wind turbine the mean wind speed (U) and the added turbulence intensity (TI)
are used to investigate the flow behavior in the wake behind the turbine. Unlike the Bolund example
shown above, the mean wind speed in this example is not from vector scaling, but is the longitudinal
wind component (U) normalized by the undisturbed velocity (U0). This is typically used as an indicator
of the wake deficit and measured at several locations downstream of the wind turbine and normalized
by the undisturbed values measured upstream of the turbine, indicated by the subscript 0.
Again, as this case was chosen for a comparison in a previous study these considerations were
already selected by the researchers conducting the experiment: 3-min averaged field data and only
time periods with a neutral atmospheric boundary layer. The 3-min periods were selected based on
undisturbed wind speed and wind direction criteria established by the researchers, resulting in 873
samples which were then binned according to wind speed and wind direction. These 873 records
were combined to create an ensemble average, therefore this dataset is indicative of mean data and
not extreme wind events. Data were collected at one location upstream (x = -2.8D) and three
locations downstream (x = 2.5D, 5.5D and 8D) from meteorological masts at varying positions in the
wake with sensors at three to five heights. In the case of wake flow investigations, the wind direction
is used to bin the data, effectively creating a lateral profile of U and TI, or wind speed deficit and
added turbulence intensity.
Figure 6: Quantity of interest plots showing the velocity ratio (U/U0) and added turbulence intensity (TI) at Left: 2.5D
and Right: 8D downwind from the Sexbierum wind turbine, numerical results using two wake models are shown.
Two quantities of interest are plotted for this analysis, shown in Figure 6 (top) for the wind speed
deficit (U/U0) and (bottom) the added turbulence intensity (TI) in the wake. Data are shown for two
downstream locations, 2.5D and 8D, and two models are shown for the comparison with experimental
data. Both models are wake models based on the same atmospheric surface layer model (CFDWind
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 101/167
Book of Proceedings
1.0) [31], with the differences between the two wake models being the model used for turbulence.
One uses the standard isotropic k-ε model (CFDWake k-ε), while the other uses additional equations
to solve all terms in the Reynolds stress tensor, hence it is referred to as the anisotropic Reynolds
stress model (CFDWake RMS) [30, 34].
For the wind speed deficit (Figure 6, top) it appears that the CFDWake RMS model does a better job
of predicting the deficit behind the turbine than the CFDWake k-ε. Closer visual inspection shows that
for the near wake behavior (2.5D) the CFDWake RMS model does not match the observations as well
as the far wake (8D), however since there are no error bars for the data there is no way of knowing if
the numerical result lies within the uncertainty of the observations. Both models appear to have
problems simulating the added TI in the wake of the turbine (Figure 6, bottom), especially in the near
wake location where they both over predict the added TI at the rotor center (WD = 0º) and under
predict as the wind direction indicative of the rotor tip is reached.
Figure 7: Scatter plot comparing the velocity ratio (U/U0) calculated using the CFDWake k-εε and CFDWake RMS models
at two locations (2.5D and 8D) downwind from the Sexbierum wind turbine, the correlation coefficient for linear
regression (R2) is also shown.
While the previous step of visually comparing the quantities of interest is insightful, this step begins the
process of statistical comparison between the numerical output and experimental data, in the form of a
scatter plot shown in Figure 7 for the normalized mean wind speed (U/U0). Like the terrain flow case
shown above, the scatter plot for the wake flow case has limited data points for comparison, with a
maximum of 25 points for each location analogous to a low statistical significance. The correlation
coefficient (R, R2) can still be useful to compare the numerical results, in addition to a visual inspection
to determine if the linear regression has a slope of one and an intercept of zero. For the CFDWake
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 102/167
Book of Proceedings
RMS model the slopes of the two linear regression lines (2.5D and 8D) are close to one, and the R2 for
the near wake is high, while the R2 for the far wake is low. These correlation coefficients can be
misleading, and should not be the only metric used for validation. For example, the CFDWake k-ε
linear regression shown in Figure 7 shows a high R2 for the near wake and a R2 higher than the
CFDWake RMS model for the far wake. If only R2 is used to determine model performance CFDWake
k-ε is better than CFDWake RMS, but closer inspection of Figure 7 shows that the slopes of both
CFDWake k-ε regression lines are much less than one and the intercepts are larger than zero, and the
plots in Step 3 visually show that CFDWake RMS predicts better than CFDWake k-ε. Therefore, it is
necessary to calculate more metrics for comparison prior to determining which model performance is
better.
For a quantitative statistical comparison between the numerical models and experimental data,
validation metrics are computed. Five validation metrics are computed for the two numerical models,
using two different quantities of interest, shown in Table 5. For the case of modeling wake flow using
two CFDWake models, the performance metrics for each of the models are similar for the mean wind
speed and turbulence intensity. The CFDWake RMS model performs slightly better than the
CFDWake k-ε model in the case of the wind speed deficit (U/U0) for four of the five validation metrics
listed in the table (FB, NMSE, MG and VG), and in the case of FAC2 the two models both achieve
100% of the numerical results being within a factor of two of the observations. Again, the importance
of selecting the proper validation metrics and quantity of interest for the application is stressed here. If
only the correlation coefficient is considered the CFDWake k-ε model performs better, but close visual
inspection of the quantity of interest and scatter plots show better results for the CFDWake RMS
model. The validation metrics computed in Table 5 for U/U0 confirm that the CFDWake RMS model
performs better than the CFDWake k-ε.
Table 5: Validation metrics to quantitatively compare the performance of two numerical models
for single turbine wake ow with data from the Sexbierum field experiment.
Validation Example: Offshore Wind Farm Wakes
The next step in wake flow investigations is to determine the effect of the turbine wake on the wind
turbine performance for turbines located downstream in a wind farm. Typically, there are limited
meteorological measurements located in warm farms, therefore SCADA data is used for the wind farm
wake analysis. In the following example, SCADA data is collected operationally for an offshore wind
farm and provides power data for each wind turbine in the Horns Rev wind farm. In the case of Horns
Rev, the objectives for the data processing were to obtain datasets for model evaluation and to better
understand wake behavior in a wind farm. There were 80, 2 MW wind turbines in the offshore wind
farm, with a hub height of 70 m, 80 m rotor diameter and arranged in a 10 (E-W) x 8 (N-S) array with
7D spacing in both directions [35].
In this case there is limited data available, therefore the selection of the quantity of interest is
simplified. There is one meteorological measurement mast located upstream to characterize the
incoming flow conditions, while the SCADA data measures the turbine power and yaw conditions,
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 103/167
Book of Proceedings
hence there is no data available to characterize the wake flow behavior. For this reason, the power
deficit (P/Pref, turbine power normalized by the turbine power in the first row or free-stream power) or
turbine performance is typically used to investigate wake behavior in a wind farm, requiring the
implementation of two numerical models prior to the validation process. Even though a flow model is
not being shown, the power of a turbine is directly influenced by the wake behavior of the turbines
located upstream and when using the power deficit as the quantity of interest the errors in the flow
model will propagate through to the power deficit, but are not distinguishable.
There are ten rows of turbines with eight turbines in each row at the Horns Rev wind farm, where the
predominant winds are from the west, researchers for the study set strict data binning and selection
criteria for the wake flow investigations [35]. The 10-min SCADA data were used to select periods
with a wind direction where the flow was down an exact row (westerly, 270º ± 2.5º) and a free-stream
velocity close to maximum rotor load (6 ms-1 ± 0.5 ms-1) to achieve the maximum wake signal.
Events with reduced power levels (i.e., idling, starting, etc.) and when the five upwind turbines were
not operational were excluded. The records remaining after implementing the data selection criteria
were combined to create an ensemble average of the mean power data for a row of ten turbines,
computed as the average power deficit of all eight turbines in each row.
The quantity of interest plot is shown in Figure 8 (left), where the normalized power deficit (P/Pref) is
compared to the results from one numerical simulation that utilizes CFDWake k-ε as the wake model
for the power deficit calculation [30]. All of the numerical results appear to be within or close to the
range of uncertainty in the observational data. Due to the limited meteorological measurements taken
in the wind farm, the datasets selected for this analysis do not account for atmospheric variability
caused by stability or turbulence, thus resulting in a large range of variability in the measurement data
[35].
Figure 8: Left: Quantity of interest plot showing the normalized power de_cit (P/Pref ) for each row of turbines in the
Horns Rev offshore wind farm (270º wind direction case, 7D spacing). Right: Scatter plot comparing the normalized
power deficit calculated using the CFDWake k-εε model, the correlation coefficient for linear regression (R2) is also
shown.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 104/167
Book of Proceedings
The scatter plot for the normalized power deficit is shown on the right hand side of Figure 8. The
statistical significance is low, due to the limited amount of data being used for the comparison.
However, the linear regression results show a line with a slope and intercept near one and zero,
respectively, and a high R2 indicating that the numerical model performs well. Still, the validation
process is continued to provide quantitative performance measures.
For the quantitative comparison between the numerical model and experimental data five validation
metrics are computed, shown in Table 6. All of the metrics are near perfect, with a perfect FAC2
meaning that all of the predictions are within a factor of two of the observations. These metrics
confirm what was seen in the previous two validation steps, where the qualitative analysis using the
quantity of interest plot and scatter plot for visual validation illustrate good model performance for the
power deficit calculated in the Horns Rev wind farm using the CFDWake k-ε flow model results for
wake effects.
Table 6: Validation metrics to quantitatively compare the performance of one
numerical model for a wind farm with data from the Horns Rev offshore wind farm.
5. Discussion
The three validation examples shown provide a starting point to develop a validation methodology for
wind energy applications. However, due to the limited amount of data and limitations in the spatial and
temporal resolution of the field studies, the use of wind tunnel data for these investigations arises.
While the atmospheric stability is difficult to capture in wind tunnel investigations, modeling of neutral
ABL processes can be done to obtain experimental datasets with high spatial and temporal resolution.
A scatter plot (Figure 9) taken from the work done in COST 732 shows a comparison of mean wind
speeds (U) from a numerical simulation with wind tunnel data, using 566 data points for the
comparison. While there appears to be a significant amount of scatter in the data and at first glance
the validation metrics (Table 7) appear to perform worse with more data points, particularly for the
FAC2 for U compared to FAC2 for S in the Bolund CFD model example. The better spatial resolution
ensures a better estimation of the small scale flow physics, therefore considerations for metrics that
include the use of more data points should be included to weight the incorporation of flow complexity
in the model evaluation.
Table 7: COST 732 Validation metrics to compare numerical
model performance with wind tunnel data.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 105/167
Book of Proceedings
Figure 9: Left: Locations for comparing wind tunnel data (black closed circle) and numerical output (gray open circle),
each point has up to 29 vertical locations. Right: Scatter plot and correlation coefficient (R2) comparing mean wind
speed from wind tunnel data with a numerical simulation, using 566 data points.
The implementation of a unique validation of metric of this type would be beneficial to the wind energy
community because of the economic and political demand on results obtained with numerical models.
Due to these sociological factors the numerical model development and improvement in the wind
energy community not only have a scientific motivation, but also hinge on fast paced progress to
assist in the development of wind farms and economic planning. An example for a novel visual
validation metric is shown in Figure 10 for the COST 732 example presented above. Here, the
percentage of data points with a positive Hit Rate (HR = 1) is calculated as a function of the total
number of data points for both U and TKE. To generate the plot data points where removed from the
dataset based on values of TKE, where those with higher TKE were removed first. This figure shows
that for minimal data points, less than 30 in this case, when selected at locations with minimal
turbulence effects the HR for both U and TKE are perfect. Therefore, if only these 30 points were
used for validation it would result in metrics that indicate a seemingly accurate model that realistically
performs poorly for particular flow scenarios.
Figure 10: Suggestion for new validation metric that incorporates the number of data points used in the model
validation.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 106/167
Book of Proceedings
Another consideration for wind energy applications is to modify the validation flow to include two
validation paths, one for a general flow model evaluation and the other for the applicability of the
model in operational scenarios, thus incorporating the fitness for purpose mentality into the validation
process. The terrain flow case presented in Section 4.1 evaluates the overall quality of the flow
model, it includes all measurement points and uses wind speed and TKE to assess the model
performance with the aim to improve the scientific understanding of the fundamental flow physics.
This can be modified for an operational validation in the selection of the quantities of interest and
important considerations, by including only the relevant measurement data for wind energy
assessment. In the case of terrain flow, specifically Bolund, this can be done by choosing the mean
wind speed measured at the 5 m height as the quantity of interest and only considering the locations
on top of the hill. Where in the context of wind energy assessment the most important quantity of
interest is the mean wind speed (related to the annual energy production), and for turbine placement
the critical locations are the hill top and the wind speed representative of hub height. Therefore, the
operational validation flow would only consider these limited data points because they are the most
crucial to the assessment of wind energy potential at a given location. This is also important in the
context of wake modeling where two numerical models, wake flow and turbine power, are
implemented prior to the final validation step. The single wake validation in Section 4.2 is
representative of the first type of validation where the flow physics are investigated, while the wind
farm case in Section 4.3 is a direct example of the operational validation, where the power deficit from
the SCADA system is used to evaluate the numerical results from a combined flow and turbine power
simulation.
6. Summary
Currently several types of numerical models and statistical processes are utilized in wind energy
assessment. Initially, a wind atlas is used to identify locations with high energy potential, where a
regional wind map is calculated over an area with a 1 km spatial resolution. Then, micrositing studies
are done to provide more site specific information regarding the expected power output from wind
turbines placed at the location. Typically, this is done with one to five years of measurements taken
from one meteorological mast and a microscale model, with resolutions of 10-100 m, to extrapolate
the measurement data to other locations within a reasonable distance. This process assumes that the
locations for extrapolation and the mast measurements are located in close enough proximity so that
the wind climate is similar. The numerical tools used for this micrositing process are WAsP and steady
state CFD models. Future research aims to improve the numerical tools by implementing models that
couple the scales of motion, which can be accomplished by using a mesoscale model at high spatial
resolution coupled with an unsteady CFD model. In its current state, this type of modeling is too
advanced for the wind energy industry to be used in practice. Although, it is important to understand
the research progress of numerical modeling to know where validation efforts should be focused.
There is currently no standardized procedure for model evaluation in regard to wind energy
assessment, and to improve power predictions a common evaluation methodology and database of
high quality experimental data is needed. Implementing a database of experimental data not only
provides a consistent comparison, but also makes certain that the same datasets are used for
comparison and not manipulated prior to the model evaluation process. Given the complexity of the
experimental data with regard to uncertainty, variability and differing requirements for the various
types of numerical models, it is suggested that each dataset undergo rigorous consideration and
justification for inclusion into the experimental database. Another consideration when trying to reduce
the uncertainty of numerical models to within a certain percentage is whether or not the data for
comparison is within a similar uncertainty range. The numerical models can not provide predictions
that are better than the experimental measurements, this includes data from both field and the wind
tunnel experiments. For future wind energy projects, where a sufficient community of wind energy
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 107/167
Book of Proceedings
researchers is established, it is recommended that selection criteria be developed to outline the
requirements for accepting an experimental dataset as an evaluation test case.
Along with outlining criteria for the acceptance of experimental data as validation data and establishing
a common model evaluation methodology, the wind energy community can benefit from a
collaborative working environment to design future experiments. This can be in the case of field
studies or physical modeling investigations where the needs of the numerical modeler can be
addressed in the experimental design and setup. For example, the inlet and boundary conditions
required for the numerical models can be properly measured in the experiment if the criteria are
known beforehand.
The numerical models can also be used in the design of large scale field studies to aid in the location
selection for measurement towers, which are limited due to the cost of the equipment and tower
installation. In the case of the Bolund field experiment, CFD modeling with a RANS solver was used
to determine the placement of the meteorological instrumentation [36]. Additionally, because the CFD
model can calculate the turbulence intensity in a spatial domain, the determination of the type of
measurement sensor also came from the numerical simulation data. Specifically, cup anemometers
versus sonic anemometers, where for Bolund the placement of the sonic anemometers was found to
be most beneficial up to a height of 5 m, because this is where the simulation showed the greatest
deviation of wind speed and turbulence values from the upstream flow.
Another application of using a model to develop a more comprehensive field experiment can be found
in Kastner-Klein et al. [37] and Leitl et al. [38], where wind tunnel data was used to design a large
scale urban dispersion study in the United States. In their study, a 1:300 scale model of the central
business district of Oklahoma City was made and measurements were taken in the atmospheric
boundary layer wind tunnel (WOTAN) at the University of Hamburg, Meteorological Institute. Care
was taken to properly model the ABL at the 1:300 scale using turbulence generators and floor
roughness elements. The results from the wind tunnel study were useful in the design of the large
scale field study because it showed that two distinct dispersion regimes exist depending on the wind
direction and source location, which dictated the final location selection for concentration
measurements.
In addition to providing useful information for the design of field experiments, it has been found that
wind tunnel data significantly enhances the capability of field data to serve as validation data for
numerical models [10]. This is due to the ability of the wind tunnel experiments to be conducted with
precisely controlled steady state flow characteristics that do not exist in the real world because of
atmospheric variability. While wind tunnel data was not available for the analysis presented in Section
4, the benefit of incorporating wind tunnel data into the model evaluation procedure for wind energy
applications was discussed. This type of building block investigation, using wind tunnel data, would be
useful to the wind energy community to investigate microscale flow phenomena occurring during
periods of neutral ABL stability. A specific example, is to make a physical model of Bolund and use
wind tunnel measurements to investigate the flow physics to better understand the wind behavior in
the wake of the hill. The data presented in Figure 3 shows a discrepancy in the results from the two
CFD models for the flow behavior behind the hill. Since field measurements are not available at that
location it is not possible to determine which numerical result is more accurate.
This paper gives a brief background of terminology and procedures used for model evaluation in
engineering applications, outlines procedures for validation and provides example validation
procedures for three scales of flow typical in wind energy applications. While the focus for this work
was on validation, it is also important that the other five steps in the model evaluation methodology
(Section 3) not be overlooked. Considerations specific to validating numerical models used in wind
energy assessment were given to provoke discussion among the wind energy research community.
Current and future international projects, such as the International Energy Agency (IEA) Task 31
WAKEBENCH [39] are considering establishing a protocol for an evaluation methodology specific to
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 108/167
Book of Proceedings
wind energy. Establishing a consensus among modelers in the wind energy community for the
development of these methods is desired, and input from researchers is critical to ensure the use and
applicability of the quality assessment protocols that will be developed. In addition to outlining the
methods for model evaluation, the final objective for the wind energy community should be to establish
threshold criteria for model validation and define best practice guidelines that can ultimately lead to
industry standards.
Acknowledgments
The authors are grateful for financial support from the EU FP7-PEOPLE program, under WAUDIT
Marie-Curie Initial Training Network.
References
[1] Leishman JG. Challenges in modelling the unsteady aerodynamics of wind turbines. Wind Energy
2002; 5:85-132
[2] Troen I, Petersen EL. European wind atlas. Ris_ National Laboratory: Roskilde, 1989.
[3] Sanderse B, van der Pilj SP, Koren B. Review of computational fluid dynamics for wind turbine
wake aerodynamics. Wind Energy 2011; , doi: 10.1002/we.458.
[4] Vermeer LJ, S_rensen JN, Crespo A. Wind turbine wake aerodynamics. Progress in Aerospace
Sciences 2003; 39:467-510.
[5] Schlesinger S. Terminology for model credibility. Simulation 1979; 32:103-104.
[6] Oberkampf WL, Trucano TG. Veri_cation and validation in computational fluid dynamics. Progress
in Aerospace Sciences 2002; 38:209-272.
[7] Schatzmann M, Leitl B. Validation and application of obstacle-resolving urban dispersion models.
Atmospheric Environment 2002; 36:4811-4821.
[8] Britter R, Schatzmann M ( (eds.)). Background and justi_cation document to support the model
evaluation guidance and protocol document. COST Office: Brussels, 2007.
[9] Eecen P, Wagenaar J, Stafanatos N, Pedersen T, Wagner R, Hansen K. Final Report UpWind 1A2
Metrology February 2011. ECN-E-11-013.
[10] Schatzmann M, Leitl B. Issues with validation of urban ow and dispersion CFD models. Journal of
Wind Engineering and Industrial Aerodynamics 2011; 99:169-186.
[11] Jackson PS, Hunt JCR. Turbulent wind ow over a low hill. Quarterly Journal of the Royal
Meteorology Society 1975; 101:929-955.
[12] Blocken B, Gualtieri C. Ten iterative steps for model development and evaluation applied to
computational fluid dynamics for environmental uid mechanics. Environmental Modelling and
Software 2012; 33:1-22.
[13] Silva Lopes A, Palma JMLM, Castro FA. Simulation of the Askervein ow. Part 2: Large-eddy
simulations. Boundary Layer Meteorology 2007; 125:85-108.
[14] Bechmann A, Sörensen NN. Hybrid RANS/LES applied to complex terrain. Wind Energy 2010;
13:36-50.
[15] Undheim O, Anderson HI, Berge E. Non-linear, microscale modelling of the ow over Askervein hill.
Boundary Layer Meteorology 2006; 120:477-495.
[16] Palma J, Castro FA, Ribeiro LF, Rodrigues AH, Pinto AP. Linear and nonlinear models in wind
resource assessment and wind turbine micro-siting in complex terrain. Journal of Wind Engineering
and Industrial Aerodynamics 2008; 96:2308-2326.
[17] Ainslie JF. Calculating the ow_eld in the wake of wind turbines. Journal of Wind Engineering and
Industrial Aerodynamics 1988; 27:213-224.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 109/167
Book of Proceedings
[18] Katic I, H_jstrup J, Jensen NO. A simple model for cluster e_ciency. European Wind Energy
Association Conference, Rome, Italy, 1986.
[19] Crespo A, Hernandez J, Frandsen S. Survey of modeling methods for wind turbine wakes and
wind farms. Wind Energy 1999; 2:1-24.
[20] Curch_eld M, Lee S, Moriarty P, Mart__nez L, Leonardi S, Vijayakumar G, Brasseur J. A
largeeddy simulation of wind-plant aerodynamics. 50th AIAA Aerospace Sciences Meeting
including the New Horizons Forum and Aerospace Exposition, Nashville, Tennessee, U.S.A., 2012.
[21] Franke J, Hellsten A, Schl unzen H, Carissimo B ( (eds.)). Best practice guideline for the CFD
simulation of ows in the urban environment. COST O_ce: Brussels, 2007.
[22] Britter R, Schatzmann M ( (eds.)). Model evaluation guidance and protocol document. COST
Office: Brussels, 2007.
[23] Chang JC, Hanna SR. Air quality model performance evaluation. Meteorology and Atmospheric
Physics 2004; 87:167-196.
[24] Schatzmann M, Olesen H, Franke J ( (eds.)). COST 732 model evaluation case studies: approach
and results. COST O_ce: Brussels, 2010.
[25] AIAA. Guide for the Veri_cation and validation of computation uid dynamics simulations. American
Institute of Aeronautics and Astronautics: Reston, VA, 1998. AIAA-G-077-1998.
[26] Oberkampf WL, Barone MF. Measures of agreement between computation and experiment:
Validation metrics. Journal of Computational Physics 2006; 217:5-36.
[27] Chang JC, Hanna SR. Technical descriptions and user's guide for the BOOT statistical model
evaluations software package, version 2.0 2005.
[28] Huges IG, Hase TPA. Measurements and their uncertainties: A practical guide to modern error
analysis. Oxford University Press: Oxford, 2010.
[29] Bechmann A, S_rensen NN, Berg J, Mann J, R_ethor_e PE. The Bolund experiment, Part II: Blind
comparison of microscale ow models. Boundary Layer Meteorology 2011; DOI:10.1007/s10546011-9637-x.
[30] Sanz Rodrigo J, Cabezóon Martínez D, García Hevia B. A systematic validation procedure for
wind farm models in neutral atmospheric conditions. 13th International Conference on Wind
Engineering (ICWE), Amsterdam, The Netherlands, 2011.
[31] Sanz Rodrigo J, Cabezón D, Martí I, Patilla P, van Beeck J. Numerical CFD modelling of
nonneutral atmospheric boundary layers for o_shore wind resource assessment based on MoninObukhov theory. EWEC-08 scientific proceedings, Brussels, Belgium, 2008.
[32] Sanz Rodrigo J, Cabez_on D, Lozano S, Martí I. Parameterization of the atmospheric boundary
layer for o_shore wind resource assessment with a limited-length-scale k-_ model. EWEC-09
scientific proceedings, Marseille, France, 2009.
[33] Cleijne J. Results of the Sexbierum wind farm: Single wake measurements 1993. TNO Report
C19.3.
[34] Cabezón D, Migoya E, Crespo A. Comparison of turbulence models for the computational fluid
dynamics simulation of wind turbine wakes in the atmospheric boundary layer. Wind Energy 2011;
14:909-921.
[35] Barthelmie R, Pryor S, Frandsen S, Hansen K, Schepers J, Rados K, Schlez W, Neubert A,
Jensen L, Neckelmann S. Quantifying the impact of wind turbine wakes on power output at
offshore wind farms. Journal of Atmospheric and Oceanic Technology 2010; 27:1302-1317.
[36] Bechmann A, Johansen J, Sörensen NN. The Bolund experiment: Design of measurment
campaign using CFD December 2007. Riso-R-1623(EN).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 110/167
Book of Proceedings
[37] Kastner-Klein P, Leitl B, Pascheke F, Schatzmann M. Wind tunnel simulation of the Joint Urban
2003 tracer experiment. 84th AMS Annual Meeting, Seattle, USA, 2004.
[38] Leitl B, Pascheke F, Schatzmann M, Kastner-Klein P. Wind tunnel experiments within the scope
of the Oklahoma City tracer experiments. International Workshop on Physical Modelling of Flow
and Dispersion Phenomena (PHYSMOD2003), Prato, Italy, 2003.
[39] Sanz Rodrigo J, Moriarty P. WAKEBENCH: Benchmarking of wind farm ow models 2010.
Proposal for Task 31 of the IEA-Wind Implementing Agreement.
[40] Obukhov ASMAM. Basic laws of turbulent mixing in the surface layer of the atmosphere. Tr. Akad.
Nauk SSSR Geo_z. Inst. 1954; 24:163-187.
[41] Cuxart J, Holtslag A, Beare R, Bazile E, Beljaars A, Cheng A, Conangla L, Ek M, Freedman F,
Hamdi R, et al.. Single-column model intercomparison for a stably stratified atmospheric boundary
layer. Boundary Layer Meteorology 2006; 118:273-303.
[42] Kim H, Lee C, Lim H, Kyong N. An experimental and numerical study on the ow over
twodimensional hills. Journal of Wind Engineering and Industrial Aerodynamics 1997; 66:17-33.
[43] Brunet Y, Finnigan J, Raupach M. A wind tunnel study of air ow in waving wheat: singlepoint
velocity statistics. Boundary Layer Meteorology 1994; 70:95132.
[44] Sanz Rodrigo J, van Beeck J, Dezs o-Veidinger G. Wind tunnel simulation of the wind conditions
inside bidimensional forest clear-cuts. Application to wind turbine siting. Journal of Wind
Engineering and Industrial Aerodynamics 2007; 95:609-634.
[45] Finnigan J, Brunet Y. Turbulent airow in forests on at and hilly terrain. Wind and Trees, Coutts
MP, Grace I (eds.). Cambridge University Press: London, 1995.
[46] Ross A, Arnold S, Vosper S, Mobbs S, Dixon N, Robins A. A comparison of wind-tunnel
experiments and numerical simulations of neutral and strati_ed ow over a hill. Boundary Layer
Meteorology 2004; 113:427-459.
[47] Lettau H. A re-examination of the Leipzig wind pro_le considering some relations between wind
and turbulence in the frictional layer. Tellus 1950; 2:125-129.
[48] Bradley E. A micrometeorological study of velocity profiles and surface drag in the region modified
by a change in surface roughness. Quarterly Journal of the Royal Meteorology Society 1968;
94:361-379.
[49] Svensson G, a M Holtslag A, Kumar V, Mauritsen T, Steeneveld GJ, Angevine WM, Bazile E,
Beljaars A, De Bruijn EIF, Cheng A, et al.. Evaluation of the diurnal cycle in the atmospheric
boundary layer over land as represented by a variety of single-column models: The second GABLS
experiment. Boundary Layer Meteorology 2011; 140:177-206.
[50] Verkaik JW, a M Holtslag A. Wind profiles, momentum fluxes and roughness lengths at Cabauw
revisited. Boundary Layer Meteorology 2007; 122:701-719.
[51] Gryning S, Batchvarova E, Br ummer B, Jorgensen H, Larsen S. On the extension of the wind
profile over homogeneous terrain beyond the surface layer. Boundary Layer Meteorology 2007;
124:251-268.
[52] Sanz Rodrigo J. Flux-profile characterization of the o_shore ABL for the parameterization of CFD
models. EWEC scientific proceedings, The Netherlands, 2011.
[53] Taylor P, Teunissen H. The Askervein hill project: Overview and background data. Boundary
Layer Meteorology 1987; 39:15-39.
[54] Coppin P, Bradley E, Finnigan J. Measurements of ow over an elongated ridge and its thermal
stability dependence: The mean _eld. Boundary Layer Meteorology 1994; 69:173-199.
[55] Berg J, Mann J, Bechmann A, Courtney MS, Jörgensen HE. The bolund experiment, part i: Flow
over a steep, three-dimensional hill. Boundary Layer Meteorology 2011; DOI:10.1007/s10546-0119636-y.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 111/167
Book of Proceedings
[56] Chen J, Black T, Novak M, Adams R. A wind tunnel study of turbulent air flow in forest clearcuts.
Wind and Trees, Coutts MP, Grace I (eds.). Cambridge University Press: London, 1995.
[57] Dellwik E, Bing ol F, Mann J, Sogachev A. Wind and turbulence at a forest edge. EWEC
scientific proceedings, Marseille, France, 2009.
[58] Wygnanski I, Champagne F, Marasli B. On the large scale structures in two-dimensional, smalldeficit, turbulent wakes. Journal of Fluid Mechanics 1986; 168:31-71.
[59] Chamorro L, Porté-Agel F. A wind-tunnel investigation of wind-turbine wakes: Boundary-layer
turbulence effects. Boundary Layer Meteorology 2009; 132:129-149.
[60] Chamorro L, Porté-Agel F. Effects of thermal stability and incoming boundary-layer flow
characteristics on wind-turbine wakes: A wind-tunnel study. Boundary Layer Meteorology 2010;
136:515-533.
[61] Chamorro L, Porté-Agel F. Turbulent ow inside and above a wind farm: A wind-tunnel study.
Energies 2011; 4:1916-1936.
[62] España G, Aubrun S, Loyer S, Devinant P. Spatial study of the wake meandering using modelled
wind turbines in a wind tunnel. Wind Energy 2011; 14:923-937.
[63] Machielse L, Eecen P, Korterink H, van der Pijl S, Schepers J. ECN test farm measurements for
validation of wake models. EWEC scientific proceedings, Milan, Italy, 2007.
[64] Politis E, Prospathopoulos J, Cabezón D, Hansen K, Chaviaropoulos P, Barthelmie R. Modelling
wake effects in large wind farms in complex terrain: the problem, the methods and the issues. Wind
Energy 2012; 15:161-182.
[65] Barthelmie RJ, Folkerts L, Ormel F, Sanderho_ P, Eecen P, Stobbe O, Nielsen NM. Offshore wind
turbine wakes measured by SODAR. Journal of Atmospheric and Oceanic Technology 2003;
20:466-477.
[66] Barthelmie R, Frandsen S, Nielsen N, Pryor S, Rethore P, J_rgensen H. Modelling and
measurements of power losses and turbulence intensity in wind turbine wakes at Middelgrunden
offshore wind farm. Wind Energy 2007; 10:217-228.
[67] Hansen K, Barthelmie R, Jensen L, Sommer A. The impact of turbulence intensity and
atmospheric stability on power deficits due to wind turbine wakes at Horns Rev wind farm. Wind
Energy 2012; 15:183-196.
[68] Dhalberg J, Norling J, Loman G, Thor S. Assessment of the Lillgrund windfarm power
performance wake losses 2009. Lillgrund Pilot Project Report, Vattenfall Vindkraft AB.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 112/167
Book of Proceedings
CFD model of stratified atmospheric boundary-layer flow
Tilman Koblitza,1, A. Bechmanna, A. Sogacheva, N. Sørensena and P.-E. Réthoréa
a
DTU Wind Energy, Roskilde, Denmark
From Wind Energ. doi: 10.1002/we.1684, published online 15 Nov 2013.
Abstract
For wind resource assessment, the wind industry is increasingly relying on Computational Fluid
Dynamics models of the neutrally stratified surface-layer. So far, physical processes that are important
to the whole atmospheric boundary-layer, like the Coriolis effect, buoyancy forces and heat transport,
are mostly ignored. In order to decrease the uncertainty of wind resource assessment, the present
work focuses on atmospheric flows that include stability and Coriolis effects. The influence of these
effects on the whole atmospheric boundary-layer are examined using a RANS k-ε model. To validate
the model implementations, results are compared against measurements from several large-scale field
campaigns, wind tunnel experiments, and previous simulations and are shown to significantly improve
the predictions..
Keywords: Atmospheric Boundary-Layer; k-E turbulence model; Coriolis effect; Atmospheric Stability; CFD; RANS.
1. Background
Wind flow modeling software is widely used for wind resource assessment, and mostly based on either
the Computational Fluid Dynamics (CFD) approach or the linear WAsP approach [1]. They focus
primarily on modeling of the neutrally stratified atmospheric surface-layer (ASL) which typically covers
the bottom 10% of the atmospheric boundary-layer (ABL). In the ASL the logarithmic wind profile is a
justified approximation and the models account for the effects of roughness and topography changes.
Atmospheric stability and Coriolis effects are mostly ignored, or like in WAsP are treated as small
perturbations to the neutral background flow that can be added after solving the model equations. In
order to decrease the uncertainty of predictions, especially in complex terrain, stability and Coriolis
effects on the whole atmospheric boundary layer should be included in such models.
Turbulence within the ABL covers a wide range of scales (from less than a cm up to several km [2]).
Since the solution of the full Navier-Stokes equations is not computationally feasible, high Reynolds
number flows can be based on the solution of the Reynolds-Averaged Navier-Stokes (RANS)
equations. Recently, Sogachev et al. [3] developed an atmospheric model for flows over flat terrain
that accounts for stability and Coriolis effects: the energy equation in terms of the potential
temperature is solved in parallel to the RANS equations and a consistent two-equation turbulence
model is used to close the equations. Using a two-equation closure method to describe the whole ABL
allows the flow to be computed at a much lower computational cost than e.g. using large-eddy
simulations (LES) [4]. The aim of the present work is to develop and validate a RANS ABL model
framework describing the whole ABL that can be applied for flows over complex terrain. The starting
point is the DTU Wind Energy CFD solver EllipSys3D [5, 6, 7]. The solver was initially developed for
1
Corresponding author: Tilman Koblitz
DTU Wind Energy, Technical University of Denmark, Department of Wind Energy, DTU Risø Campus, Frederiksborgvej 399,
Building 115, 4000 Roskilde, Denmark: Tel.: +45 51801495; e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 113/167
Book of Proceedings
simulating the near-ground surface-layer flow inside a neutrally stratified domain (from now on referred
to as the ASL model) and has under these conditions been validated against field experiments [8, 9].
To model ABL flows more appropriately the solver is modified (from now on referred to as the ABL
model) and the two-equation turbulence closure from Sogachev et al. [3] is used.
To get a better understanding of the physical processes involved in ABL flows and to validate ABL
models, more data sets from atmospheric experiments on full scale are necessary. Various
experiments that focus on neutral flow over complex terrain are available, e.g. the Askervein Hill
experiment [10], or more recently the Bolund experiment [11, 8]. Existing benchmark literature for nonneutral ABL flows mostly focuses on flat terrain [12, 13, 14, 15], and test cases for complex terrain are
scarce. Ross et al. [16] present a wind-tunnel study of neutral and stably stratified boundary-flow over
a steep hill. The wind-tunnel experiment was designed to represent a realistic non-neutral ABL flow
and analyses stability effects over terrain under controlled conditions. Although not real ABL flow at full
scale, this test case was chosen to test the applicability and performance of the ABL model for flows
over well defined but steep terrain.
The central goal of the present study is to examine how well the ABL model performs in representing
neutral and non-neutral ABL flows, and to set the starting point to apply the ABL model for flows over
complex terrain. In section 2 the modeling approach is presented, followed by section 3 where
implementation aspects (3.1) and the simulation methodology (3.2) are described. In section 4 MoninObukhov Similarity Theory (MOST) is briefly described. Section 5 presents results from four test cases
that are used to validate the ABL model.
The simulations are divided in three parts. First, sections 5.1 and 5.2 focus on neutral ABL flow over
rough flat ground. Simulation results are compared against experimental data from the Leipzig wind
profile [17] and the Cabauw site [18]. Second, non-neutral ABL flow over rough flat ground is
considered in section 5.3. Experimental data from the GABLS2 test case [15], that analyses a diurnal
cycle in the ABL, is used to validate the ABL model for non-neutral conditions. Additionally, MOST is
used to compare simulation results against experimental data from several large scale field campaigns
[12, 13, 14]. Third, in section 5.4 the wind-tunnel experiment analysing stratified boundary-layer flow
over a steep hill [16] is used to assess the applicability of the ABL model for non-neutral flows over
terrain. Concluding remarks are given in section 6.
2. Modeling ABL flows
2.1. Goberning equations
The high Reynolds number atmospheric flows considered in this study are based on the solution of the
RANS equations. The continuity and momentum equations read:
∂
ρU j = 0 ,
∂x j
(
)
(1)
 ∂U ∂ U j
∂ρUi ∂ρUi U j
∂ 
+
−
( µ − µ t )  i +
 ∂x j
∂t
∂x j
∂x j 
∂xi


  ∂ Pˆ
 +
= Sv ,

  ∂xi
(2)
where xi (x1 = x, x2 = y, x3 = z) are the longitudinal, lateral and vertical directions. Ui is the mean
velocity component along xi, P̂ is the pressure and µt is the turbulent eddy viscosity.
Since the unsteady term is retained in equation 2 it is possible to simulate transient phenomena with
RANS. This is based on the assumption that time averaging of the RANS equations is performed on a
time scale similar to the turbulent fluctuations, while the low frequency variations of the mean flow
(e.g. diurnal simulations of stratified flows in the ABL, as considered in this study) can be properly
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 114/167
Book of Proceedings
resolved by the unsteady RANS equations. Transient RANS allows to simulate the unsteady
phenomena of a diurnal cycle in the ABL, and is basically the only option besides the more
computationally expensive LES approach.
2.2. Basic RANS turbulence closure for ASL flows
The RANS equations are used to describe the air flow in the lowest part of the ABL where the
logarithmic wind profile a justified approximation. To close the given set of equations the popular k-ε
turbulence model is used [19]. The eddy-viscosity is obtained by solving the two differential transport
equations for the turbulent kinetic energy k and the dissipation rate ε :
∂k
∂k
∂
+Uj
−
∂t
∂x j ∂xi
∂ε
∂ε
∂
+Uj
−
∂t
∂x j ∂xi
 µt ∂k 

 = Pk − ε ,
 σ k ∂xi 
 µt ∂ ε  ε

 = (Cε 1Pk − Cε 2ε ) ,
 σ ε ∂xi  k
(3)
(4)
where σk and σε are the Schmidt numbers for k and ε respectively and Pk is the rate of shear
production of k. Cε1, Cε2 are model coefficients. The resulting mixing length l and the eddy viscosity µt
are expressed in terms of k and ε as:
l = Cµ3 / 4
µt = ρCµ
k2
ε
k3 / 2
ε
,
(5)
= ρCµ1/ 4k 1/ 2l .
(6)
2.3. Adaptation of RANS Turbulence Closure for ABL Flows
When modeling the full ABL, thermal stratification and Coriolis effects (caused by the rotation of the
earth) should be included. These effects are introduced into the RANS equation system via additional
source/sink terms on the right hand side of the momentum equations 2:
Sv = gi ( ρ − ρ0 ) + ε i fc ρU j + Svol ,
(7)
where gi is the gravitational acceleration giT = (0, 0, g), ρ is the varying density and ρ0 is a reference
density. εiT = (−1, 1, 0) and fc = 2Ωsinλ is the Coriolis parameter (with the earth’s rotation rate Ω and
the latitude λ), and are added explicitly to the momentum equations as an external force. The Coriolis
force in vertical direction is neglected since it is small compared to the gravitational acceleration.
To include buoyancy effects an equation for the potential temperature is solved in addition to the
RANS equations:
∂
∂
∂ 
µt
ρU jθ −
( ρθ ) +
 µ +
∂t
∂x j
∂x j 
σθ
(
)
 ∂θ

 ∂xi

 = Sθ .

(8)
The potential temperature equation couples with the momentum equations via vertical buoyancy
forces gi(ρ − ρ0) that act in the direction of the gravitational acceleration. Density variations as a result
of pressure variations are assumed to be small so that the flow is treated incompressible. Hence, the
density is not a function of pressure, and temperature and density vary linearly as required by the
Boussinesq approximation:
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 115/167
Book of Proceedings
ρ=
M
,
RT
(9)
where M = 29 g/mol is the molar mass and R =8.313 J/molK is the universal gas constant.
To close the given set of equations Sogachev et al. [3] recently developed a consistent closure
method for the two-equation turbulence model given above by considering the behavior of the ε
equation in homogeneous turbulent flow when using a source/sink term associated with stability. The
potential temperature equation now couples with the turbulence model via an additional source/sink
term B that is added to the two turbulent transport equations [20, 21]:
∂k
∂k
∂
+Uj
−
∂t
∂x j ∂xi
 µt ∂k 

 = Pk − ε + B ,
 σ k ∂xi 
(10)
∂ε
∂ε
∂  µt ∂ ε  ε *
+Uj
−

 = Cε 1Pk − Cε 2ε + Cε 3B + D .
∂t
∂x j ∂xi  σ ε ∂xi  k
(
)
(11)
Cε1* is a modified Cε1 coefficient (see section 2.3.1) and D is an additional diffusion term (see section
2.3.2). The buoyancy source/sink term B is added to the turbulent kinetic energy equation and also
appears in the dissipation equation together with the coefficient Cε3 and depends on the eddy viscosity
µt, the gravitational acceleration gi and the density gradient:
B = µt g i
∂ρ
.
∂xi
(12)
In unstable conditions B is positive and supports the generation of turbulent kinetic energy in the k
equation, while B turns negative in stable conditions and suppresses turbulence. Details about the
specification of the coefficient Cε3 are given in section 2.4.2.
2.3.1. Length-scale limitation
The standard k-ε model, when applied to ABL flows, is known to be too diffusive, leading to a strongly
overestimated turbulent length scale l that grows continuously with height and results in a very large
ABL height [22, 23]. In real ABL flows the maximum size of turbulent eddies is limited e.g. by the finite
ABL height or by stratification (see also [24]). Using a length-scale limiter, as initially proposed by
Apsley and Castro [23], allows to reduce the maximum mixing length in the model and the resulting
ABL height is effectively reduced. The modified Cε1* coefficient in the length scale determining
equation 11 is described by:
Cε*1 = Cε 1 + (Cε 2 − Cε 1 )
l
.
le
(13)
Apart from the maximum global mixing length le no additional coefficients are introduced. When the
local mixing length l reaches the specified global maximum mixing length le, Cε1* equals Cε2 and the
production and destruction terms in the dissipation equation are in balance, which limits the local
length scale l to le. On the other end, when l << le, Cε1* equals Cε1 and the modification still satisfies the
logarithmic wind profile in the surface-layer close to the ground.
For neutrally stratified ABL flows over a flat rough surface, le is estimated by an expression from
Blackadar [25]. To provide a suitable solution for stratified flows an expression from Mellor and
Yamada [26] is used that depends on the vertical distribution of turbulent kinetic energy k in the ABL,
and reflects variations in ABL depth induced by thermal stratification:
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 116/167
Book of Proceedings
G

l0 = 0.00027 , constant in neutral ABL

fc

∞
lε = 
,
z kdz
∫
0
l
=α ∞
, reflecting variable ABL depth
 MY
kdz

∫0
(14)
where G is the geostrophic wind. The coefficient α is chosen so that both length scales are identical
for a neutrally stratified ABL flow (lMY = l0). For several test cases an empirical value of α =0.075
results in both length scales to agree reasonably well (lMY ≈ l0) [3]. To calculate lMY with the above
expression the turbulent kinetic energy k needs to be integrated over the domain height. For
horizontally homogeneous flow it is sufficient to perform this integration every time step at one location
since the flow is horizontally homogeneous. For complex terrain domains with a curvilinear grid this
integration would have to be performed at every location individually. This is computationally not
feasible, and instead a precursor simulation over flat terrain can be used to determine the time varying
values of lMY that are then used within a complex terrain domain.
2.3.2. Diffusion term
An additional diffusion type term D is introduced into the dissipation equation 11 of the k-ε model as
proposed by Sogachev et al. [3]. Numerical experiments have shown some differences in the behavior
of the standard k-ε model of Laundner and Spalding [19] and the k-ω model of Wilcox [27]. Sogachev
and Panferov [28] reported that for example in forest canopies the k-ω model is performing slightly
better than the k-ε model. The ABL model in the present study is developed using the k-ε model, but to
obtain consistent results between the two closures the k-ε model can be transformed to behave
similarly to the k-ω model, by including an additional diffusion term:
 1
1  ∂2 k  1
1  k ∂ε ∂k
2 ∂k ∂k 
.
D = Cµ 
−
+
+
k 2 −

  σ k σ ε  ∂x j  σ k σ ε  ε ∂x j ∂x j σ k ∂x j ∂x j 
(15)
3. Model coefficients
2.3.1. Specification of general constants
The model constants Cε1 and Cε2 in equations 4 and 11 are chosen to be consistent with experimental
observations for decaying homogeneous, isotropic turbulence. To ensure that the model solution
agrees with the constant-stress logarithmic wind profile near the ground the relation below has to be
satisfied, which follows from considering constant-flux flow with ∂k/∂z =0 [20]:
σε =
κ2
Cµ1/ 2 (Cε 2 − Cε 1 )
.
(16)
For neutral flows the constant Cµ is typically adjusted to set a desired turbulence level and Cµ =0.09 is
a typical value for industrial flows [19] while in atmospheric research a value of Cµ =0.03 is often used
[29].
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 117/167
Book of Proceedings
2.4.2. Specification of stability related coefficients
The additional coefficient Cε3 in the dissipation equation 11 has to be specified, and an optimal value is
unknown [30]. Using the recently developed consistent closure method for two-equation turbulence
models from Sogachev et al. [3], Cε3 is modeled using a stability-related coefficient αB (see table 1):
Cε 3 = (Cε 1 − Cε 2 )α B + 1 .
(17)
αB is specified based on the standard coefficients Cε1 and Cε2 of the production and destruction terms
in the ε equation respectively:
l

1 − l
MY

αB = 
1 − 1 + Cε 2 − 1  l
  Cε 2 − Cε 1  lMY

for Rig > 0
,
with Rig = B / Pk .
(18)
for Ri g < 0
αB now depends on the local gradient Richardson number Rig and on the local ratio of l/lMY , and hence
is a function of stability.
Table 1: Typical values for the coefficients in the k-ε turbulence model for
industrial flows (Cµ = 0.09) and for atmospheric flows (Cµ = 0.03).
In the model the turbulent Prandtl number σθ in equation 8 is approximated as a function of a slightly
modified gradient Richardson number RiG:
for Ri g > 0
0.74
σθ = 
 0.74 (1 − 15RiG )
−1/ 4
for Rig < 0
,


α
with RiG = B  Pk + B B  .
σθ 

(19)
The additional term in the denominator of the gradient Richardson number RiG is used to stabilize
equation 19: RiG and the resulting σθ are now limited during convective cases where Pk approaches
zero.
Sogachev et al. [3] have shown numerically that the developed model framework is suitable for three
flow regimes: grid turbulence, wall-bounded flow and homogeneous shear flow. In contrast to an
earlier proposed description of Cε3 [30] the form given in equation 18 is universal and needs not to be
specified for each case. Compared to the ASL model no additional coefficients are introduced into the
ABL model that need to be calibrated. The formulation does not allow for any tuning of the model and
only depends on the closure coefficients given in table 1.
3. Simulating ABL flows
3.1. Implementation into EllipSys3D solver
To simulate ABL flows the EllipSys3D solver is modified using the ABL model equations presented
above.
To solve the convection-diffusion equations, the third-order accurate QUICK differencing scheme is
used. All equations are transformed into general curvilinear coordinates. This allows the model to be
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 118/167
Book of Proceedings
applied to complex geometries, like ABL flows over terrain, where the use of Cartesian or rectangular
coordinates is often not possible.
The buoyancy forces gi(ρ − ρ0) in the momentum equation 2 are added explicitly as external volume
forces. When implemented into the solver this can create numerical problems: the implementation via
a discrete body force generates a numerical decoupling between the pressure and the velocities. This
problem was identified to cause oscillations in the solution, especially close to boundaries and under
strongly stratified conditions. To avoid this, an algorithm for allocating discrete forces is used following
Réthoré and Sørensen [31]. This approach solves the problem by spreading the buoyancy forces on
the neighbouring cells and by applying an equivalent pressure jump at the cell faces.
In order to improve convergence for small mixing lengths, ambient floor values for the turbulence
variables are imposed. Especially during strongly stable conditions the mixing length l and the eddy
viscosity µt approach values close to zero, or even negative values. This typically occurs within the
stable temperature layer (inversion) in the upper part of the ABL. To avoid numerical convergence
issues, k and ε are not allowed to drop below a predefined limit. The ambient values kamb and ε amb are
defined a priori and set a minimum turbulent mixing length via equation 5. They are set using a
minimum limiter on the turbulence variables: ε = max(ε ,εamb) and k = max(k, kamb). Values below the
ambient level are simply overwritten. For the present study the ambient turbulence values are chosen
to be kamb =1 · 10−4 m2/s2 and ε amb =7.208 · 10−8 m2/s3. Together with the model constants in table 1
and equation 5 this leads to a minimum turbulent mixing length of 1 m. The modification is only active
in the upper part of the ABL where the turbulence variables approach their predefined ambient levels.
3.2. Model Setup
All presented calculations use the set of consistent closure coefficients for ABL flows stated in table 1.
Specific simulation parameters like roughness length z0, geostrophic wind G, Coriolis parameter fc and
surface temperatures θ0 are summarized in table 2.
Table 2: Values for the simulation parameters associated with each model run: geostrophic wind G that is used to
specify the pressure gradient to drive the flow, roughness length z0, Coriolis parameter fc, maximum global
mixing length scale le and potential surface temperature θ0. Four test cases are presented covering neutral ABL
flow over flat terrain (Leipzig, Cabauw), non-neutral ABL flow over flat terrain (GABLS2) and neutral and stable
wind-tunnel flow over a steep hill (here le and fc are not applicable).
To simulate ABL flows over flat terrain, initial conditions are typically specified by the logarithmic wind
profile over flat terrain according to a specific surface roughness z0. However, since cyclic boundary
conditions are used for the present simulations the final results do not depend on their initial
conditions. To drive the flow in the ABL simulations a pressure gradient is applied that results in the
desired geostrophic wind G. The Coriolis force balances the pressure gradient force at the ABL top,
where friction by definition is zero: G = −1/ρfcδp/δy.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 119/167
Book of Proceedings
3.2.1. Computational domain
For all flat terrain simulations the same computational domain is used: it is 6 km high and uses a grid
of 384 stretched cells in vertical direction. The bottom cell is equal to the size of the roughness length
z0 at the wall, and the mesh is stretched hyperbolically towards the top resulting in cell heights of
about 70 m at the top boundary. In horizontal directions the domain is 1 x 1 km long with a grid of 32
evenly distributed cells. However, as the modeled flow over flat surfaces is horizontally homogeneous,
the horizontal grid structure is irrelevant and the flow variables are therefore functions of the height z
alone. Rough wall boundary conditions are used at the bottom of the domain [9], and a symmetry
condition (no-gradient) is used on top. All vertical boundaries are cyclic. Note that grid independent
results could already be obtained using a grid of around 100 cells in the vertical direction.
The wind-tunnel experiments of Ross et al. [16] were conducted at the Environmental Flow Research
Laboratory (EnFlo), University of Surrey, U.K. The wind tunnel has a working section of 20 m length,
3.5 m width and 1.5 m height and the shape of the two-dimensional steep hill is given by:
H cos2 ( π x / l )
z=
0
for − l / 2 ≤ x ≤ l / 2
elsewhere
,
(20)
where H = 0.229 m is the maximum hill height, x is the distance from the centre of the hill and l =1 m is
the width of the hill. The computational domain covers 10 m of the wind-tunnel test section, and is 1.5
m high and wide and the hill is placed 2.5 m behind the upstream boundary. The wind-tunnel flow is
solved by the EllipSys3D solver and therefore the domain is 3-D, although the model hill and the
resulting flow is 2-D. The grid has 192 grid points in the horizontal, 24 in the lateral and 96 in the
vertical direction. Stretched cells are used in the vertical direction with a height of 0.27 mm at the wall
and 5 cm at the top of the domain. In horizontal direction the mesh is refined on top of the hill with
cells of 1.5 cm length and is stretched towards the inlet and outlet where cells are about 10 cm long. In
lateral direction the mesh is equally spaced with the cells being 6 cm wide. At the inlet and upper
boundary inlet conditions are used and at the downstream boundary outlet conditions are used.
4. Monin-Obukhov similarity theory
Monin-Obukhov Similarity Theory [32] expresses the vertical structure of the horizontally
homogeneous ASL as dimensionless universal functions and is often used to validate ABL models.
Based on dimensional analysis, all nondimensionalized mean flow properties within the ASL only
depend on a reduced set of key scaling parameters: the friction velocity u∗, the height above ground z,
and the vertical turbulent heat flux H:
H ≡ w 'θ ' ≈
µt ∂θ
.
σθ ∂z
(21)
From these parameters a universal length scale, the Obukhov length L, can be formed that describes
the exchange processes in the surface-layer:
L=−
u*3θ0
,
κ gH0
(22)
where θ0 is the potential temperature at the surface and H0 is the near-surface value of the vertical
turbulent heat flux H. L is proportional to the vertical potential temperature gradient and describes the
height at which buoyant production of turbulence first exceeds mechanical production due to shear.
The dimensionless height ζ = z/L is used as a stability parameter and has the same sign as the
Richardson number Ri: positive in stable conditions and negative in unstable conditions. Based on the
assumptions that the flow within the surface-layer is stationary, horizontally homogeneous, and that
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 120/167
Book of Proceedings
fluxes are independent of height, ζ is now constant throughout the surface-layer (in contrast to Ri) and
the normalized wind speed depends on the universal function ζ alone:
φm ( ζ ) =
∂U  κ z 

.
∂z  u* 
(23)
For ideal conditions of stationary and horizontally homogeneous flow MOST is valid in about the
lowest 10% of the ABL where the Coriolis effect is negligible. The forms of the stability function φm are
not known from dimensional analysis and is obtained empirically from field experiments over flat
terrain: measurements for different values of ζ are substituted into equation 23 and curves are fitted to
the resulting data. Despite numerous field experiments there are still discrepancies in the literature for
the exact forms of φm. The forms chosen here are widely used and known as the Businger-Dyer
relations [12, 33]:
1 + δ m ζ
−1/ 4
 (1 − γ m ζ )
φm = 
for ζ ≥ 0
for ζ < 0
.
(24)
In the literature the coefficient δm ranges from 4 to 10, and γm ranges from 15 to 28 (see e.g. [34, 13]).
During ideal conditions the relation 23 between dimensionless wind shear and dimensionless height ζ
is valid for any wind speed, height, roughness and stability condition in the surface-layer. Ideally
vertical profiles will then collapse on one line.
Integration of equation 23 for the velocity gradient yields a modified logarithmic wind profile for the
wind speed U:
U=

u*   z 
ln   − ψ m (ζ )  ,
κ   z0 

(25)
where Ψm is a stability function. Using the Dyer stability functions [35] Ψm is defined as:
ψm
 −5ζ

=   1 + x 2
 ln  2
 
for ζ ≥ 0
  1 + x 2 
π
 
  − 2atan ( x ) + 2
2
 

for ζ < 0
,
with x = (1 − 16ζ
)
1/ 4
. (26)
5. Results and discussion
5.1. Neutral ABL -Leipzig test case
In this case the Leipzig wind profile is modeled [17] with the ASL and the ABL model both run steadystate. The necessary simulation parameters are shown in table 2 and the computational grid is
described in section 3.2.1. Results are shown in figure 1 together with measurements [17].
Comparison of the ASL and ABL model results shows the influence of the Coriolis effect: the
additional body force in the ABL model induces a velocity component v perpendicular to the direction
of the geostrophic wind G and causes the wind to veer with height. In the ABL model also the height of
the ABL is now limited to about 1300 m, due to the applied length-scale limiter (seen in the middle of
figure 1, where the velocity component v approaches zero). With the chosen value of l0 = 41.8 m the
ABL height is however slightly overpredicted. This length scale is slightly larger than the one
suggested by Apsley and Castro [23] who used l0 = 36 m for their simulation. It is generally accepted
that the Leipzig experiment was actually conducted in slightly stable conditions [24], and when using a
lower length scale of l0 = 28 m the measured and simulated profiles agree perfectly (not shown here).
However, the goal was not to match the simulation to a single observation, and the ABL model
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 121/167
Book of Proceedings
predicts the flow reasonably well and simulated results are significantly improved compared to the
ASL model.
Figure 1: Results of the Leipzig test case using the ASL model (dashed blue) and the ABL model (solid red) shown
together with the Leipzig wind profile (grey symbols) [17]. Left: wind component u parallel to geostrophic wind G
plotted over height; Middle: wind component v perpendicular to geostrophic wind G plotted over height; Right:
turbulent mixing length scale l (see equation 5) plotted over height.
Figure 2: Results of the Cabauw test case using the ASL model (dashed blue) and the ABL model (solid red) shown
together with anual averages of the Cabauw site (grey symbols) [18] for three classes of geostrophic wind speeds (5,
10 and 15 m/s). Bars denote standard deviations. Left: dimensionless wind components (U − U g)/u∗ and (V − Vg)/u∗
plotted over non-dimensional height zfc/u∗ ; Right: turbulent mixing length scale l (see equation 5) plotted over height.
5.2. Neutral ABL -Cabauw test case
The neutral ABL over flat terrain at the Cabauw site in the Netherlands is simulated [18] with the ASL
and ABL models run steady-state. The necessary simulation parameters are shown in table 2 and the
computational grid is described in section 3.2.1. Results are shown in figure 2. The non-dimensional
geostrophic wind components (U − Ug)/u∗ and (V − Vg)/u∗ are shown as functions of the nondimensional height zfc/u∗, plotted with a logarithmic scale. Anual averages from the Cabauw site [18]
are shown for three classes of the geostrophic wind (G = 5, 10, 15 m/s) at heights 10, 80 and 200 m,
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 122/167
Book of Proceedings
together with simulation results using G = 10 m/s. When plotted using above non-dimensional form the
simulation results for other geostrophic winds collapse on the same line. As for the Leipzig test case,
the turning of the wind with height induced by the Coriolis force and the limitation of the ABL height by
the length-scale limiter can be seen when comparing the simulated profiles of the ABL model with the
ASL model. Results from the ABL model agree well with the measurements [18] and the chosen
modifications prove applicable.
5.3. Non-neutral ABL -GABLS2 test case
This section focuses on assessing how well the ABL model performs in representing non-neutral ABL
flows. Observations from the GABLS2 test case held in Kansas, USA [15] with a strong diurnal cycle
are chosen to validate the ABL model. In the study of Svensson et al. [15] the observational dataset is
compared against simulation results from 30 different models and simulating the described diurnal
cycle has shown to represent a challenging test case for ABL models.
Figure 3: Diurnal evolution of flow properties. a: surface temperature θ0 (black line) that is given as input to the ABL
model and the resulting potential temperature field within the first 300 m ; b: wind speed at 10 m (green) and friction
velocity u∗ at the surface (blue). Symbols and lines denote measurements and simulation results respectively; c: wind
profiles of the ASL model (dashed blue) and ABL model (solid red) compared against analytical profiles from MOST
(equation 25) for different times of the day.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 123/167
Book of Proceedings
The simulation uses the computational grid from section 3.2 and the ABL model is run transient, where
non-neutral conditions are induced by a prescribed time-varying ground temperature. A time step of 1
s is used. The initial conditions for the simulation of the diurnal cycle are given in [15].
Results are shown in figure 3. Figure 3a is showing the time varying ground temperature that is used
as a model input together with the resulting diurnal evolution of the computed potential temperature
field that adopts to the changing surface conditions. The surface stability conditions are influencing the
generation of turbulence and the turbulent mixing. Figure 3b shows the friction velocity u∗ at the
surface (on the left axis) and the velocity variation at the 10 m level (on the right axis) over one diurnal
cycle together with measurements (symbols) and the spread of the different model results from the
model intercomparison study of Svensson et al. (shaded region) [15].
In both plots, 3a and 3b, a clear transition between day-and nighttime is visible after sunrise around
8:00 and after sunset around 18:00. Stable nighttime conditions before 8:00 are characterised by
small turbulence levels and a low ABL depth of around 100 m. The stable stratification suppresses the
generation of turbulence and results in small values for the friction velocity as shown in figure 3b. The
air close to the ground is colder then the air above, and due to the small amount of mixing, only
penetrates up to heights of about 100 m, where a steep temperature gradient is visible in figure 3a.
During daytime between 12:00 and 18:00 unstable conditions are induced by the heating of the
ground. Large amounts of convective turbulence lead to a well mixed ABL with a greater depth. Due to
convection warm air is rising upwards and penetrates the strong stable temperature gradient that is
capping the ABL during night. After 12:00 the stable gradient is not existent anymore and the ABL
continues to grow in height. It is during this period that the turbulent length scales and the friction
velocity at the surface reach their maximum values. At around 14:00 the maximum temperature is
reached, and before returning to the stable nighttime regime, the ABL flow is close to neutral at around
18:00 where the potential temperature is nearly constant with height.
Also shown in figure 3b is the evolution of the wind speed at the 10 m level. Higher wind speeds are
observed during daytime, where the increased turbulence is effective at mixing momentum downward
close to the ground and vice versa. The shaded areas indicate the model spread of the 30 models that
were intercompared within the study of Svensson et al. [15]. Computed results generally are within the
observed range and a clear diurnal pattern is visible. Svensson et al. [15] report that all models
underestimate the 10 m wind speed after the morning transition and tend to overestimate the wind
speed towards the end of the day. One obvious reason for this is that the geostrophic wind during the
simulation was kept constant in space and time, while both, observations and mesoscale simulations
shown in [15] show a decrease of the geostrophic wind during the observational period. Also note that
the measured turbulent kinetic energy shows a sudden increase at about 3:00 which was reported to
be a local disturbance not included in the model forcing, and therefore not present in the computed
results.
Figure 3c shows the computed wind profiles in the first 100 m at different times of the day compared
against the observations and the standard logarithmic profiles from the ASL model. Also shown are
the theoretical profiles from MOST [36] where the computed surface heat flux H0 together with
equations 22 and 26 is used to determine the Obukhov length L and the modified logarithmic wind
profile. Stable conditions at night result in smaller wind speeds close to the ground and higher wind
speeds above, when compared against the logarithmic solution of the neutral ASL model. During
unstable conditions the wind speed increases rapidly over the first few meters while it is almost
constant with height above. The agreement is good, and the developed model captures the observed
and theoretical non-neutral behavior.
In figure 4 MOST (see section 4) is used to assess the performance of the ABL model. Theoretical φm
functions from equation 24 are shown together with simulation results and experimental data from
several field campaigns [12, 13, 14]. To decrease the spread of the experimental φm values, the data
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 124/167
Book of Proceedings
needs to be selected carefully. Especially during transitional periods in the morning and evening the
assumptions underlying MOST (stationary and horizontally homogeneous flow with constant ζ over
height) are violated in real ABL flows. The shown simulation results are therefore selected accordingly:
only cases for fully developed flow away from the transitional periods are shown. MOST was derived
for the range |ζ| < 2 [12] and for higher values it can be seen that experimental and simulation results
start to deviate from MOST.
Figure 4: φ m from observations (grey symbols: Businger et al. [12], Li et al. [13], Klipp and Mahrt [14]), analytical
expression from equation 24 with δ m = 5 and γm = 15 (solid black) and for the range of analytical solutions where δm
varies from 4 to 10 and γ m varies from 15 to 28 [3] (shaded area), and results from the ABL model. Note that results
during transitional regimes around 8:00 (sunrise) and 18:00 (sunset) are omitted, since during these conditions z/L is
not constant with height which is a necessary assumption for MOST to be valid.
Note that in this case the ABL model is run for several days, cyclically repeating the surface
temperature from the GABLS2 test case, until a cyclical soultion is reached. This ensures that the
whole flow field is in equilibrium with the model equations, and that the solution is independant of the
initial conditions. For the GABLS2 model intercomparison the spin-up time of the models given in [15]
(time before numerical and observational data is compared) is only 8 h. We find that the solution in
this case is still significantly dependant on the initial temperature field (also given in [15]), and when
compared against MOST, the agreement is not as good as for the fully converged results. This
indicates that the flow field in the ABL after 8 h of spin-up time is not yet in equilibrium with the surface
forcing at the ground. In a recent study, Sogachev also found [30] that a spin-up time of several days
is needed, depending on the initial conditions. In summary, we find model results to agree best with
MOST when allowing the solution to fully converge to a cyclical solution. Results for the GABLS2 test
case are best, when following the instructions from Svensson et al. [15]. This indicates that if
information on a large scale (like the geostrophic wind or the vertical temperature profile) is available
from measurements, those conditions should be used in the model to compare numerical and
observational data. However, conditions in the real ABL are often non-stationary and horizontally nonhomogeneous, and are therefore not necessarily in agreement with empirical theories like MOST or
with the model equations of numerical models.
5.4. Non-neutral flow over a steep hill
In this test case neutral and stably-stratified boundary-layer flow over a steep hill is simulated and
compared against simulations and wind-tunnel measurements [16, 37]. The wind-tunnel experiment
was designed to represent realistic ABL flow over a two-dimensional steep hill. Wind-tunnel flows
cannot fully resemble real ABL flow at full scale and the Coriolis effect is negligible. However, this test
case allows to study stability effects under controlled conditions, and is chosen to test the applicability
and performance of the ABL model for flows over well defined terrain.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 125/167
Book of Proceedings
The hill is steep enough to induce flow separation and represents a challenging test case for the
developed ABL model. The simulation is run in transient mode with a time step of 0.1 s.
Figure 5: Contour plots of non-dimensional streamwise velocity U/Uf in a vertical plane across the hill for neutral (left
column) and stable flow (right column). Simulation results are shown along with measurements and RANS results from
Ross et al. [16] and LES results from Wan et al. [37].
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 126/167
Book of Proceedings
Figure 6: Contour plots of momentum flux u'w' (in m2s-2) in a vertical plane across the hill for neutral (left column) and
stable flow (right column). Simulation results are shown along with measurements and RANS results from Ross et al.
[16] and LES results from Wan et al. [37].
Due to the small scale of the wind-tunnel the Coriolis effect is neglected in the model, hence equation
14 to determine a maximum length scale le cannot be applied, and no length-scale limitation is used.
The necessary input parameters to simulate the wind-tunnel flow are summarized in table 2. Two
cases are simulated: neutral and stably stratified flow with a relatively weak stratification of about 10
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 127/167
Book of Proceedings
K/m in the lowest 0.5 m and a much stronger stratification of about 40 K/m above. The neutral
simulations are solved steady-state and the stable simulations are run transient for 150 s of model
time (equivalent to 15 tunnel flow-through times) after which the computed flow has reached a quasisteady state.
The initial conditions for the neutral simulation are specified by the logarithmic wind profile with a
surface roughness of z0 =0.23 mm (see table 2). For the stably stratified flow the inlet profiles at the
upstream boundary are generated by running a precursor simulation: the experimental velocity and
temperature profiles from the wind-tunnel given in [16] are run through the ABL model using the wind
tunnel specified below in absence of the hill. This ensures that the inlet profiles are in equilibrium with
the model equations. Computed wind speed and turbulence properties of the neutral and stably
stratified flow are compared against experimental results [16] and simulation results [16, 37]. Figure 5
shows contour plots of the streamwise velocity u in a vertical plane perpendicular to the hill and figure
6 shows contour plots of the momentum flux u'w'. Results for neutral flow are shown in the left column,
and results for stable flow are shown on the right. In both cases flow separation occurs at the lee side
of the hill. For the stable case, the depth of the wake region is slightly increased because the stable
stratification acts to suppress vertical motion. Above 0.5 m the strong temperature gradient of about
40 K/m is effectively capping the flow and turbulence is limited to the lower part of the domain and is
significantly reduced when compared to the neutral case where the momentum fluxes are an order of
magnitude larger. The ABL model captures the general effects induced by stratification, although the
size of the wake is different. For both, neutral and stably stratified flow the recirculation region in the
lee side of the hill is significantly overpredicted compared to the wind-tunnel experiment, while it
agrees well with the LES results [37]. The velocity above of the hill is generally predicted well. The u'w'
values in figure 6 are generally predicted well and agree with both wind-tunnel and LES results from
[37]. However, for the neutral case the u'w' values upstream of the hill are found to be too high when
compared to the wind-tunnel values. Similar findings were reported by Wan et al. [37]: the model is
found to be too dissipative in this region which leads to an increased upwards deflection of the flow
induced by the hill and leads to a slower velocity recovery downstream of the hill and hence an
overestimated wake region.
Since the ABL model was developed for ABL flows at full scale, we cannot expect the model to
reproduce the wind-tunnel measurements. Due to the small scale of the wind-tunnel the Coriolis effect
is neglected and the length-scale limiter is not applied. The implemented turbulence closure has been
developed for steady ABL flows, and it cannot be expected that the unsteady wake region in the lee of
the hill is predicted correctly, and it is not the aim of the ABL model. The wake region has shown to be
sensitive to changes in the model constants. No coefficients were adjusted and all test cases are run
with the same set of constants from table 2. Although this test case is of limited value to verify the
developed ABL model, it is shown that the model can be applied on curvilinear grids without any
modification, and that general effects of stratification on the flow are captured correctly.
6. Conclusions
The present work presents an ABL model that aims at describing the wind flow within the whole ABL.
The model is successfully validated using four test cases. For neutral ABL flow, two test cases over
flat terrain are considered and the implemented Coriolis effect and the length-scale limited k-ε model
prove applicable. Computed profiles for the velocity components agree well with measurements from
the Leipzig and the Cabauw test case [17, 18].
For non-neutral ABL flow a diurnal cycle is simulated, where a time varying surface temperature
reflects different stability conditions that typically occur within the ABL throughout one day. The
implementation of the k-ε model developed by Sogachev et al. [3] and of the potential temperature
equation proved applicable and the ABL model that now accounts for stability effects performed well.
Finally a wind-tunnel test case is used to validate the ABL model for stably stratified flow over a steep
hill. Although this test case is of limited value to validate the ABL model, the applicability for flows over
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 128/167
Book of Proceedings
terrain using curvilinear grids was shown. For all test cases the computed velocity, potential
temperature and turbulence values compare reasonably well.
The advantage of the presented RANS model framework is its general applicability. All
implementations in the ABL model are tuning free, and except for the simulation parameters in table 2,
no additional model coefficients need to be specified a priori the simulation. In summary the test cases
show that the developed ABL model is applicable and gives significantly improved predictions for
neutral and non-neutral ABL flow compared to the ASL model. It presents a promising approach to
also applying the ABL model to complex topography.
Acknoledgements
This work has been carried out within the WAUDIT project (Grant no. 238576). This Initial Training
Network is a Marie Curie action, funded under the Seventh Framework Program (FP7) of the
European Commission. Computations were made possible by the use of the PC-cluster Gorm
provided by DCSC and the DTU central computing facility. We would also like to thank the Danish
energy agency (EFP07-Metoder til kortlægning af vindforhold i komplekst terræn (ENS-33033-0062)),
the Center for Computational Wind Turbine Aerodynamics and Atmospheric Turbulence (under the
Danish Council for Strategic Research, Grant no. 09-067216).
References
[1] Mortensen N, Heathfield D, Myllerup L, Landberg L and Rathmann O 2005 Wind atlas analysis
and application program: WAsP 8 help facility (online) ISBN 87-550-3457-8
[2] Jacobson M Z 2005 Fundamentals of Atmospheric Modeling 2nd ed (Cambridge University Press)
ISBN 0521548659
[3] Sogachev A, Kelly M and Leclerc M 2012 Boundary-Layer Meteorol. 1–21
[4] Bechmann A 2006 Large-eddy simulation of atmospheric flow over complex terrain Ris-phd-28(en)
Risø National Lab., Roskilde, Denmark
[5] Michelsen J A 1992 Basis3d-a platform for development of multiblock pde solvers Technical report
afm Technical University of Denmark
[6] Michelsen J A 1994 Block structured multigrid solution of 2d and 3d elliptic pde solvers Technical
report afm Technical University of Denmark
[7] Sørensen N N 1995 General purpose flow solver applied to flow over hills Technical report risø-r827(en) Risø National Lab., Roskilde, Denmark
[8] Bechmann A, Sørensen N N, Berg J, Mann J and Réthoré P E 2011 Boundary-Layer Meteorol.
141 245–271
[9] Sørensen N N, Bechmann A, Johansson J, Myllerup L, Botha P, Vinther S and Nielsen B S 2007
Identification of severe wind conditions using a reynolds-averaged navier-stokes solver
(Proceedings: The Science of Making Torque from Wind)
[10] Taylor P A and Teunissen H 1987 Boundary-Layer Meteorol. 39 15–39
[11] Berg J, Mann J, Bechmann A and Courtney M S 2010 Boundary-Layer Meteorol. 141 219–243
[12] Businger J A, Wyngaard J C, Izumi Y and Bradley E F 1971 28 181–189 ISSN 0022-4928, 15200469 URL http://adsabs.harvard.edu/abs/1971JAtS...28..181B
[13] Li X, Zimmerman N and Princevac M 2008 Boundary-Layer Meteorol 129 115–136 ISSN 00068314, 1573-1472 URL http://link.springer.com.globalproxy.cvt.dk/article/10.1007/ s10546-
008-9304-z
[14] Klipp
C
L
and
Mahrt
L
2004
130
2087–2103
ISSN 00359009,
1477870X
URL
http://adsabs.harvard. edu/abs/2004QJRMS.130.2087K
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 129/167
Book of Proceedings
[15] Svensson G, Holtslag A, Kumar V, Mauritsen T, Steeneveld G, Angevine W, Bazile E, Beljaars A,
de Bruijn E, Cheng A, Conangla L, Cuxart J, Ek M, Falk M, Freedman F, Kitagawa H, Larson V,
Lock A, Mailhot J, Masson V, Park S, Pleim J, Sderberg S, Weng W and Zampieri M 2011
Boundary-Layer Meteorol. 140 177–206 ISSN 0006-8314 10.1007/s10546-011-9611-7 URL
http://dx.doi.org/10.1007/s10546-011-9611-7
[16] Ross A N, Arnold S, Vosper S B, Mobbs S D, Dixon N and Robins A G 2004 Boundary-Layer
Meteorol
113
427–459
ISSN
0006-8314,
1573-1472
URL
http://link.springer.com.globalproxy.cvt.dk/ article/10.1007/s10546-004-0490-z
[17] Lettau H 1950 2 125129 ISSN 2153-3490 URL http://onlinelibrary.wiley.com.globalproxy.
cvt.dk/doi/10.1111/j.2153-3490.1950.tb00321.x/abstract
[18] Ulden A P V and Holtslag A A M 1980 The wind at heights between 10m and 200m in comparison
with the geostrophic wind vol 1 ed E C commission L (Proceedings: Seminar on Radioactive
Releases) pp 83–92
[19] Launder B E and Spalding D B 1974 Comp. Meth. Appl. Mech. Engineer. 3 269–289
[20] Pope S B 2000 Turbulent flows 1st ed (UK: Cambridge University Press)
[21] Pielke R 2002 Mesoscale meteorological modeling (Academic Press, San Diego) ISBN
9780125547666
[22] Spalart P R and Rumsey C L 2007 AIAA Journal 45 2544–2553
[23] Apsley D D and Castro I P 1997 Boundary-Layer Meteorol. 83 75–98 ISSN 0006-8314
[24] Zilitinkevich S S and Esau I N 2002 Boundary-Layer Meteorol. 104 371–379 ISSN 0006-8314
[25] Blackadar A K 1962 J. Geophysical Research 67 3095–3102
[26] Mellor G L and Yamada T 1974 J. Atmos. Sci. 31 1791–1806
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 130/167
Book of Proceedings
Turbulent fluxes, stability and shear in the offshore environment: WRF
mesoscale modelling and field observations at FINO1
D. Muñoz-Esparzaa,1, B. Cañadillasb, T. Neumannb and J. van Beecka
a
von Karman Institute for Fluid Dynamics (VKI), 1640 Rhode-Saint-Gènese, Belgium
b
DEWI GmbH - Deutsches Windenergie-Institut, Wilhelmshaven, Germany
From Renewable Sustainable Energy 4: 63-136 (2012), doi.org/10.1063/1.4769201.
Abstract
This paper is focused on the evaluation of five Planetary Boundary Layer (PBL) schemes in the
Weather Research and Forecasting model for offshore wind energy purposes. One first order scheme:
Yonsey University and four one-and-a-half order schemes: Mellor-Yamada-Janić, Quasi-Normal Scale
Elimination, Mellor-Yamada-Nakanishi-Niino and Bougeault-Lacarrère, are considered. Turbulent flux
measurements from the FINO1 platform in the North Sea are used to estimate the Obukhov length,
allowing the sorting of the data into different stability classes. In addition, wind LiDAR measurements
are used to analyze wind profiles up to 251.5 m, encompassing the heights where today’s wind
turbines operate. The ability of the different PBL schemes to forecast turbulent fluxes of heat and
momentum and surface stability is evaluated. Obukhov length results show that in general, PBL
schemes forecast more moderated stable stratifications and a reinforcement of the instability for
neutral and convective conditions, compared to FINO1 observations. The vertical structure of the wind
speed profile is thoroughly analyzed for stable, near-neutral, unstable and very unstable conditions by
using total shear stresses, eddy diffusivities and wind speed shears. The Mellor-Yamada-NakanishiNiino scheme presents the best agreement with measurements considering the different atmospheric
stabilities analyzed. Stable conditions are the most complicated scenario for the PBL schemes to
reproduce due to their overdiffusive formulations, which effect is to lower the vertical wind shear.
Under such conditions, Quasi-Normal Scale Elimination and Yonsey University outperform the rest of
the PBL schemes, the latest using a revised diffusion formulation.
Keywords: Atmospheric Boundary-Layer; k-E turbulence model; Coriolis effect; Atmospheric Stability; CFD; RANS.
1. Introduction
To date, there are approximately 4 GW of offshore wind power capacity installed in Europe, almost 6
GW under construction, 17 GW consented by EU Member States and future plans for a further 114
GW. Therefore, it is expected that during this decade, offshore wind power capacity will grow tenfold
to reach an estimated installed capacity of 40 GW by 2020 [1]. In order to help with such a
development, numerical models can be used for both the optimization of the location of wind farms
(wind resource assessment) and estimation of the daily energy production once the wind farm is in
operation (short term forecasting).
Microscale numerical models are currently used for wind resource assessment purposes. In the case
of Reynolds Averaged Navier Stoke models (RANS), they can go up to spatial resolutions in the order
1
Corresponding author: Domingo Muñoz-Esparza
Environmental & Applied Fluid Dynamics Department, von Karman Institute for Fluid Dynamics, 1640 Rhode-Saint-Gènese,
Belgium, Tel.: +32 2 359 97 60, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 131/167
Book of Proceedings
of 1 m which can be achieved with a relatively low computational cost [2]. However, these models
have two main limitations. Firstly, the conditions of the site are required as a model input. Thus, the
boundary conditions have to be obtained from onsite measurements which are normally acquired from
a meteorological mast. Therefore, their use becomes unfeasible in the case of lack of on-site
measurements. Secondly, these models cannot be used to forecast the expected wind/power
production a few days ahead.
Numerical Weather Prediction models (NWP) are designed for weather forecasting purposes [3, 4].
However, they have a strong potential to be applied by the wind energy community as they reproduce
the time evolution of atmospheric processes using realistic boundary and initial conditions. Moreover,
they can simulate across multiple spatial scales by use of nesting techniques (dynamic downscaling).
On the other hand, they parameterize all the microscale processes due to the coarse horizontal
resolution used (∼1 km). For wind energy applications, where the boundary layer processes are
crucial, the Planetary Boundary Layer (PBL) parameterization has to be analyzed in detail.
Vertical turbulent fluxes of momentum are the main driver of wind speed gradients in the PBL. Zängl et
al. [5] simulated an Alpine foehn with the mesoscale model MM5 in order to investigate the influence
of PBL parameterization. They found that vertical momentum eddy diffusivities, Km, can present
notable differences and that first-order closure schemes provide larger values than turbulent kinetic
energy (TKE) based schemes (1.5 order), particularly in strong wind conditions. Hu et al. [6] compared
three PBL schemes from the Weather Research and Forecasting (WRF) model: Yonsey University
(YSU), Asymmetric Convective Model (ACM2) and Mellor-Yamada-Janić (MYJ), against surface and
boundary layer measurements of temperature and wind speed. They observed that the 1.5 order
scheme (MYJ), provides too weak mixing under convective conditions whereas YSU and ACM2, which
are non-local schemes perform better. The main focus of the analysis was on the potential
temperature results and so the influence of vertical mixing on the wind speed was not sufficiently
analyzed. However, it was shown that the strong mixing of YSU was unable to simulate low level jet
events, as these are damped partially or even totally. Shin and Hong [7] performed a more detailed
analysis of five PBL parameterizations in WRF (YSU, ACM2, MYJ, Quasi-Normal Scale Elimination
(QNSE) and Bougeault-Lacarrère (BouLac)), for a one day period of onshore measurements. Surface
properties were evaluated from six 10 m high masts while profiles were provided by radiosoundings up
to 1.5 km. Higher biases of surface fluxes were found during nighttime (stable conditions) than during
daytime, providing similar results for all schemes. Regarding PBL structures, they attributed the better
performance of the YSU scheme under convective conditions to its non-local formulation. None of the
PBL schemes were able to properly simulate stable boundary layer features. However, the TKE based
schemes were in better agreement with measurements. All these studies focused on the physical
modelling of the entire boundary layer but, in the case of wind energy applications, a particular
evaluation needs to be performed with a focus on the lower part of the PBL, where wind turbines
operate.
In the last years, some research has started to be done on the use of WRF for wind energy purposes.
Storm and Basu [8] carried out a one year wind resource analysis with the YSU PBL scheme. The
wind speed distribution at 100 m was in a good agreement with tower observations at Sweetwater
(Texas), as was the mean wind shear based on 50 m and 100 m heights. However, when the shear
was averaged daily, it was shown that it was underestimated (overestimated) during nighttime
(daytime) hours. The effects of initial and boundary data on the estimation of the wind shear
exponents were found to be minimal among the four datasets used. Draxl et al. [9] evaluated forecast
results from seven PBL formulations in WRF at the onshore site Høvsøre during a one month period.
They found that all the schemes tend to underestimate the wind at hub height during nighttime and
overestimate it during daytime. The characteristics of the diurnal cycle and their transitions were in
general well captured. They noticed that TKE based schemes provided better results. A tendency
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 132/167
Book of Proceedings
within the YSU scheme to always reproduce neutral wind profiles was observed, which is in
agreement with previous results from Storm and Basu [8].
So far, contributions dealing with offshore wind energy applications are restricted due to a lack of high
quality data for validation. Suselj and Sood [10] studied the influence of a new formulation of the
master length scale (as used in Mellor-Yamada-Nakanishi-Niino (MYNN)) for the MYJ scheme on the
wind speed shear. Data from two offshore sites were used: the FINO1 mast in the North Sea and the
Östergarnsholm tower in the Baltic Sea. At the measurement heights of FINO1 (41.5-81.5 m), the new
formulation did not show any substantial improvement.
Wind turbines rated at 5 MW have been in operation since 2010 at the Alpha Ventus wind farm close
to FINO1, in the North Sea. Thus, the ability of mesoscale models to forecast “tall” wind profiles needs
to be properly addressed. In addition, special attention has to be paid to the role of atmospheric
stability, the effect of which is to significantly alter the wind shear. This causes a variety of velocity
distributions across the rotor swept area, which dramatically influences the power production and
fatigue loads on the wind turbine.
The paper is structured as follows. In Section 2 we describe the FINO1 test site and its
instrumentation. In Section 3 an optimum averaging time for turbulent fluxes is determined and a
stability classification of the data is made. The WRF setup is described in Section 4. Surface fluxes of
momentum and heat are evaluated in Section 5.1., together with Obukhov length results. Section 5.2.
addresses the vertical structure of the wind profile, with a discussion for each stability class in terms of
wind shear, eddy diffusivities and profile fluxes. Finally, Section 6 is devoted to the conclusions.
2. Test site and instrumentation
The German research platform FINO1 is located 45 km North of the Borkum Island (lat. 54°0.87’N,
lon. 6°35.24’E) in the North Sea (Fig. 1) and has been in operation since 2003 [11]. The mean water
depth of the platform is about 30 m. The platform is equipped with a tall meteorological mast up to
about 100 m, performing continuous multilevel measurements within the lower part of the atmospheric
boundary layer.
Figure 1: General view of the FINO1 mast.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 133/167
Book of Proceedings
In the present work, a one year period is investigated (January 2010 - December 2010).
Measurements from the sonic anemometers at 41.5 m, 61.5 m and 81.5 m (10 Hz) were used as the
primary source of data while slow response sensors (wind direction, relative humidity, air pressure and
temperature) are used mainly for data post-processing and/or comparison. The three ultrasonic
anemometers are located on north-westerly oriented booms. In order to minimize the influence of the
mast structure on the measurements of the sonic anemometers and to avoid the wakes of the nearby
Alpha-Ventus wind farm, we have selected wind conditions with directions comprised within the sector
(240º, 360º). This sector corresponds to open-sea conditions. In addition, a ground-based pulsed wind
LiDAR system, WindCube v1, developed and manufactured by the French company Leosphere has
been used in this study. The wind LiDAR system was positioned on a container roof at approximately
10 m distance to the north-west of the FINO1 mast. It performed continuous measurements from July
2009 to February 2011, scanning the atmosphere up to a height of 251.5 m. This wind LiDAR has
been proven to be applicable for wind speed and wind direction measurements by comparison against
mast-based sensors [12].
3 Turbulent flux measurements and stability
3.1 Calculation of turbulent fluxes
Before proceeding with further calculations, the quality of the data from all the sensors was verified
using a similar methodology to the one of Vickers and Mahrt [13]. To calculate the turbulent fluxes of
momentum and heat we use the eddy-covariance method, as in Cañadillas [14]. The natural wind
coordinate algorithm was used to remove the mean vertical wind component [15]. The choice of the
averaging time for the eddy-covariance method, taver, influences the magnitude of the averaged
turbulent flux. This has been shown by several investigations [16-18]. In addition, taver can vary with
several factors: location, wind speed, measurement height and atmospheric stability. An optimum
value of the averaging time should be used in order to include the low frequency contribution to the
flux. Oncley et al. [19] applied the Ogive function to determine the amount of low frequency content
that was included in the turbulent flux measured with the eddy-covariance method. The Ogive function
is the cumulative integral of the cospectrum, Co, from the highest to the lowest frequencies:
Ogw ',x (fi ) =
∫f Cow ',x (f ) df ,
fi
(1)
N
where fN is the Nyquist frequency and fi is the lowest computed frequency. In our case x is one of the
following fluctuating magnitudes: alongwind velocity, u', crosswind velocity, v', or potential
temperature, Θ'. In all the cases w' stands for the fluctuating vertical velocity.
We chose a time series length of 120 min, as a characteristic time where the largest turbulent
structures could contribute to the turbulent flux and the time series is detrended. The Ogive function
analysis was focused on two months which are representative of the stabilities occurring at FINO1.
This method allowed a robust time averaging period to be estimated. Sensitivity tests were performed
by varying taver: 5, 10, 20, 30 and 60 min.
In Fig. 2a, the Ogive function for different taver is plotted versus its maximum (i.e., the maximum
turbulent flux) on each time series. It can be observed that reduced taver, 5 or 10 min, deviates more
from the 1:1 slope, whereas for 30 and 60 min the agreement is almost perfect. However, the
observed scatter introduces some error on the flux calculation. To quantify the error obtained for each
taver, the integrated error distribution is shown in Fig. ¡Error! No se encuentra el origen de la
referencia.b, being the error calculated as the difference with respect to the maximum of the Ogive
function for each time series. For a cumulative frequency of 80% the error is about -45% and -25% for
taver of 5 and 10 min, respectively. For the rest of taver (20, 30 and 60 min), the error falls below -10%.
By comparing the Ogive function results with the flux errors and the correlations, we can conclude that
30 min is an accurate time averaging interval to compute turbulent fluxes at FINO1. Moreover, our
optimum value of taver is consistent with other sea-air flux studies [20-23]. Larger time intervals
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 134/167
Book of Proceedings
increase the risk of including non-steady effects without reducing the errors and shorter intervals will
produce a dramatic increase of the error. Similar conclusions were obtained for <w'v'> and <w'Θ'> and
other heights (61.5 and 81.5 m).
Figure 2: (a) Ogive function values for different taver vs. maximum of the Ogive function and linear regression. (b)
cumulative frequency of occurrence as a function of the flux error. Results correspond to <w'u'> at 41.5 m for the two
selected months.
3.2 Turbulent fluxes and stability
The total shear stress, τm, can be directly calculated from the alongwind, <w'u'>, and crosswind,
<w'v'>, kinematic momentum flux components:
τ m = ρ u*2 = ρ ( − w ' u '

) + (− w 'v ' )
2
2 1/ 2

,
(2)
where ρ is the density and u* is the friction velocity.
Figure 3: Momentum flux components as a function of wind speed at 41.5 m. (a) Alongwind momentum flux, <w'u'>41.5.
Dashed line represents the solution from the logarithmic wind profile equation under neutral conditions. (b) Crosswind
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 135/167
Book of Proceedings
momentum flux, <w'v'>41.5. Data is sorted by <w'u'>41.5 < 0 and <w'u'>41.5 > 0. Bin median values and correlation are
plotted.
A quadratic dependency of <w'u'>41.5 with the wind speed at 41.5 m, U41.5, was found (Fig. 3a). That
tendency agrees well with the expression derived from the logarithmic wind profile equation, where the
roughness length, z0, is parameterized using Charnock’s equation [24]: z0 = α cu*2/g, being g the
acceleration due to gravity and adopting α c = 0.016 [25]. The crosswind component, <w'v'>41.5,
remains close to zero, with an almost zero slope. Observations where <w'u'>41.5 > 0 occur mainly for
low velocities (U41.5 < 5 ms-1) and seem to be due to the uncertainty in the flux calculation and/or
strong swell conditions which could dominate the total shear for light winds where the turbulence
contribution to the flux is reduced. Observations where <w'u'>41.5 > 0 are discarded from further
calculations, representing 7.34% of the dataset.
The Obukhov length, L, can be directly estimated from the turbulent fluxes as:
L=−
u*3θ
,
g κ w ' θ 'v
(3)
where θ'v is the potential temperature, <w'θv'> is the kinematic virtual potential heat flux and κ is the
von Kármán constant (= 0.40). The sonic heat flux is used instead of <w'θv'>, which is slightly lower
(<w'θv'> -0.1 <w'q'>), and partly includes the contribution of the humidity flux, <w'q'>. The Obukhov
length is used to perform a detailed atmospheric stability classification. Different stability classes for
the offshore environment are considered according to van Wijk et al. [26], which propose the
classification presented in Table 1. The data correspond with the wind sector (240º, 360º), so that the
number of occurrences for each stability class collected in Table ¡Error! No se encuentra el origen
de la referencia. does not represent the overall conditions prevailing at FINO1. More weight is given
to unstable atmospheric stratification due to the fact that air masses advected from the North-West are
usually cooler than the sea surface temperature and so convective scenarios would be expected to
dominate.
Table 1: Stability classification based on the Obukhov length, L.
In Fig. ¡Error! No se encuentra el origen de la referencia.a the dimensionless wind speed profiles,
U/u*,41.5, corresponding to the predominant conditions on each stability class are plotted. The stability
classification method used sorts the profiles satisfactorily, showing an increase in the shear from
unstable to stable conditions. Under VS conditions, the boundary layer height is very shallow and
geostrophic conditions are reached above ∼180-200 m in some of the cases. For VS and some of the
S cases our reference sonic anemometer (41.5 m) is positioned very close to the top of the surface
layer. However, the stability classification remains consistent with the expected behaviour of the wind
speed profiles, as is shown in Fig. ¡Error! No se encuentra el origen de la referencia.a. In Fig.
¡Error! No se encuentra el origen de la referencia.b the probability distribution on the U-zL-141.5
plane is presented. The competition between mechanical and thermal turbulence is evident. For
higher wind speeds, u* dominates and the stability tends to be neutral. Only for lower velocities, with
an important contribution of <w'θ'>, the stability/instability starts to dominate. Strong stabilities (both
VUN and VS) are not possible to develop for U41.5 > 15 ms-1. We found that this pattern is bounded by
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 136/167
Book of Proceedings
symmetric elliptical curves (black solid lines in Fig. ¡Error! No se encuentra el origen de la
referencia.b).
Figure 4: (a) Dimensionless wind speed profiles, U/u*,41.5, corresponding with the predominant conditions of each
stability class. (b) Two dimensional histogram of the U-zL-141.5 plane. Colorbar indicates the number of samples on
each 2D bin.
Figure 5: WRF domain configuration. The horizontal resolutions of the three domains are 27 km, 9 km and 3 km, from
the parent to the innermost domain, respectively. Color bar indicates surface elevation in meters.
4. WRF mesoscale modelling
We use the Numerical Weather Prediction model of the National Center for Atmospheric Research in
USA (NCAR): Advanced Research WRF (ARW) v3.2, to perform mesoscale simulations of the
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 137/167
Book of Proceedings
conditions at the FINO1 platform. The ARW dynamics solver is a conservative finite differences code
that integrates the non-hydrostatic compressible Euler equations [27]. Our computational domain is
composed of 3 domains centered over the FINO1 platform. The parent domain has a horizontal grid
spacing of 27 km and covers an approximate surface of 3300 x 3300 km2, including most of Europe.
Grid spacing is refined progressively by a factor of 3 through two nested domains until 3 km resolution
is achieved for the inner most domain, which covers approximately 350 x 350 km2 (see Fig. 5). On the
vertical coordinate, 60 levels are placed. Grid spacing is of 10-20 m up to 300 m height to accurately
resolve the lower part of the boundary layer. Above 300 m, grid spacing is progressively stretched in
order to reduce the computational cost. Interactions of the meteorological fields between the domains
are accounted for by one-way nesting. The domain configuration is shown in Fig. 5. The time step is
consistently reduced from the parent domain to the most inner domain in order to respect numerical
stability constraints (the Courant-Friedrichs-Lewy (CFL) number was limited at every time step to
CFL<1). Single-Moment 3-class simple ice scheme microphysics [28], rapid radiative transfer in the
longwave [29], the Dudhia shortwave scheme [30], NOAH surface scheme [31] and cumulus KainFritsch scheme [32] (not applied into the innermost domain) were used. Horizontal eddy viscosities are
calculated using a first-order Smagorinsky closure, based on horizontal deformations. The role of such
diffusion is to introduce the effect of large scale two-dimensional turbulence structures present in the
atmosphere, while the vertical mixing due to small unresolved scales is accounted for by the PBL
scheme.
The effect of sub-grid scale turbulent fluxes in the vertical coordinate is accounted for by the solution
of the 1D diffusion equation. Five PBL formulations are analyzed; one first order scheme: YSU [33]
and four one-and-a-half order schemes: MYJ [34], QNSE [35], MYNN [36, 37] and BouLac [38]. We
implemented the explicit calculation of the turbulent fluxes for each of the PBL schemes analyzed. We
used a first-order centered finite difference scheme to discretize the vertical gradients, in the same
way that the implementation of the diffusion equation is performed in the code.
w 'ξ '
i
 ξ −ξ

= −Kξi  i +1 i − γ ξi  − w ' ξ '
 ∆zi

3
 zI 
,
h h 
 
(4)
where ξ is a generic prognostic variable (u, v or θ), ∆z is the distance between two neighboring nodes,
Kξ is the eddy diffusivity and h is the PBL height. The subscript i refers to full-levels whereas the
subscript I means half-levels in the vertical coordinate. Note that the contribution due to the flux at the
inversion layer (second term of the right hand side) and the non-local contribution to the flux, γξi, are
only considered in the case of the YSU scheme. For the rest of the schemes, only the local part of the
turbulent flux is computed. In addition, we implemented the revised YSU scheme for stable conditions
proposed by Hong [39]. This scheme keeps the original K-profile formulation from Hong et al. [33] but
changes the definition of the mixied-layer velocity scale (now only dependent on the mechanical
forcing) and removes the non-local contribution for stable stabilities. Each PBL scheme is tied to a
different surface layer formulation [27], being all based on the application of the Monin-Obukhov
similarity theory [40] at the first vertical grid point. A summary of the PBL and surface layer schemes
considered on this paper is presented in Table 2.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 138/167
Book of Proceedings
Table 2: Summary of the of the five PBL schemes analyzed and their respective surface layer options.
The parent domain is initialized and 6-hourly forced at the boundaries by meteorological fields derived
from the NCEP Climate Forecast System Reanalysis data, CFSR [41], with a horizontal resolution of
0.5º x 0.5º. Sea surface temperatures are updated every 6 h from interpolation of daily reanalysis over
a 0.25º x 0.25º grid (AMSR-AVHRR [42]). The first 24 hours are discarded as spin-up time of the
model to generate the proper energy content on the mesoscale wavelengths [43]. Grid nudging is
applied in the parent domain and is switched off in the boundary layer not to influence the behavior of
the PBL schemes.
5. Comparison of simulations and observational data
As shown in Section 3, wind speed profiles vary greatly depending on the atmospheric stratification. In
order to have statistical representativeness in our analysis, we selected 380 samples (30 min
averages) from each stability class and simulated their corresponding periods with WRF. The wind
speed is evaluated using 10 min averages for the same periods (1140 samples). Very stable
conditions are a challenging scenario for numerical models. They present low turbulence levels where
intermittent episodes of relaminarization take place. In addition, extreme events such as low-level jets
occur. These aspects are particularly difficult for mesoscale models to handle and thus need a specific
analysis. Therefore, we focus on the remaining stability classes ranging from very unstable to stable
conditions.
5.1. Surface turbulent fluxes
Figure 6 (first column) shows the 2D histograms of u*,s from WRF and u*,41.5 measured at the lowest
sonic anemometer at FINO1. The subscript s refers to surface conditions in WRF, i.e., surface fluxes
calculated from the surface layer scheme. The first vertical grid point for wind speed is located at z1 ∼
12 m, but it slightly varies during the simulation time due to the σ-coordinate formulation.
WRF predicts systematically higher values of u*,s with respect to the measurements at 41.5 m (Fig. 6,
first column). Both u*,s from WRF and u*,41.5 from FINO1 are well correlated for all the PBL schemes (r2
∼ 0.83-0.86, where r is the linear correlation coefficient between observations and model results). Part
of the overestimation of u* by WRF comes from the fact that the top of the surface layer height is
dictated by the height of the first grid point, z1 ∼ 12 m, whereas in the atmosphere it varies depending
on the stability. Above z1, the PBL formulation controls the turbulent mixing and produces a decrease
of the vertical fluxes that vanish at the PBL height. In order to have a fair comparison, u*,41.5 derived
from the vertical flux of momentum is presented in Fig. 6 (second column). The mean u* difference
with respect to observations was reduced by 0.05 ms-1 for YSU, by 0.03 ms-1 for QNSE and MYNN,
and by 0.02 ms-1 for the rest of the schemes when u*,41.5 was used instead of u*,s. It has been shown
for an onshore case by Shin2012 that the location of z1 can influence surface properties such as u*
and <w'θ'>. In terms of u*, they found little dependency on z1 during daylight hours whereas during
nighttime, u* from the TKE schemes decreased and YSU showed an opposite behaviour when z1 was
lowered. Since the influence under unstable conditions is negligible, we adopted a value of z1 ∼ 12 m
which is a rather good representative of the surface layer height in stable conditions.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 139/167
Book of Proceedings
Figure 6 (third column) compares observations of <w'θ'> to WRF results at 41.5 m. The discrepancies
between WRF and the sonic heat flux increase with the magnitude of <w'θ'>41.5 for positive values
(upward heat flux) for all the PBL schemes. As for u*,s, greater overpredictions were found when
<w'θ'>s was considered (not shown). For stable stratifications (negative <w'θ'>41.5), the agreement is
better for all the schemes due to the tendency of WRF to neutralize the stable stability for reduced
values of <w'θ'>41.5. Another contributing factor to this phenomenon is the reduced magnitude of the
heat flux for stable conditions compared to convective stability. Similar results were obtained by Suselj
and Sood [10] for a more stable case at FINO1 using a MYJ-type of PBL scheme with a modified
length scale. Also, Peña and Hahmann [47] obtained equivalent results using the YSU scheme for
another location in the North Sea, when comparing to a sonic anemometer at 50 m above mean sea
level. Two recently published articles highlight the problem of the overprediction of the surface heat
flux magnitude by the WRF model also over land [46, 48]. They both observed the same effect that we
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 140/167
Book of Proceedings
are reporting on <w'θ'>, using measurements at 10 m and 5.37 m above ground surface, respectively.
Figure 6: Two-dimensional histograms of : u*,41.5 at FINO1 with u*,s WRF (first column), u*,41.5 at FINO1 with u*,41.5 WRF
(second column), <w'θ'>41.5 at FINO1 with <w'θ'>41.5 WRF (third column) and L41.5-1 at FINO1 with L41.5-1 WRF (fourth
column). Colorbar indicates the number of samples on each 2D bin. The scale of the figures in the first three columns
(fourth column) is shown on the left (right) colorbar. The PBL scheme is pointed out on top of every figure.
In our measurements, the uncertainty on the calculation of the heat flux is about 15% (sonic
anemometer of the type SOLENT 1210R3-50 [49]), which cannot be the only factor contributing to the
mismatching of the results. Moreover, the use of sonic heat flux instead of sensible heat flux for the
measurements tends to reduce/increase the values of <w'θ'>41.5 during stable/convective conditions,
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 141/167
Book of Proceedings
which effect is to diminish the differences with respect to the WRF results depending on the magnitude
of the humidity flux.
Obukhov length results are compared in Fig. 6 (fourth column). Under unstable atmospheric
stratification (L-1 < 0), the overestimation of both momentum and heat fluxes by the numerical model
has opposite effects on L41.5-1. Simulated values of u*,41.5 tend to reduce the magnitude of L41.5-1
whereas <w'θ'>41.5 contributes to enhance it. This competition between u*,41.5 and <w'θ'>41.5 results in
an underestimation of L41.5-1 by all of the PBL schemes, due to the overprediction of <w'θ'>41.5. For
near-neutral conditions, the predicted heat flux is positive most of the time (Fig. 6, third column),
producing negative values of L41.5-1. In the case of stable stability (L-1 > 0), the overestimation of
<w'θ'>41.5 by WRF is much lower, since it depends on the magnitude of the heat flux. This gives more
weight to u*,41.5 and thus, L41.5-1 is underpredicted. This effect is more pronounced for MYJ and
BouLac. However, QNSE, MYNN and YSU agree better with the observations. There is a cold bias of
1-1.5 K for the range of observed temperatures between FINO1 measurements and WRF results at
41.5 m. Since it has a maximum contribution to L41.5-1 of 0.0037%, the influence of the θ41.5 bias is
negligible. In general, we found that WRF forecasts more moderated stable stratifications compared to
measurements whereas a reinforcement of the instability for neutral and convective regimes is
predicted. The above described effects displace the probability density function of L41.5-1 from positive
towards negative values.
5.2 Vertical shear and profile fluxes
Sonic anemometer data are used for the lowest three levels of comparison (41.5, 61.5 and 81.5 m) in
terms of wind speed and momentum fluxes, while the rest of the wind speed profile up to 251.5 m is
covered by the wind LiDAR. In order to analyze the vertical change of wind speed, the results for the
mean wind speed shear, ∂U/∂z, are collected in Fig. 7 (left column) for each of the stability classes
under consideration.
Table 3: Root-mean-square errors for wind speed difference, ∆U [ms -1], and wind speed U [ms-1]. Subscript indicates
the height and p stands for profile average (using 9 heights: [61.5:20:201.5, 251.5] m). Results are presented for each
PBL scheme and grouped by stability class. The two PBL schemes with best skills for ∆Up [ms -1] are emphasized in
bold whereas the worst one is underlined. The calculation of statistics is based on 1140 samples (10 min averages) for
each stability class.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 142/167
Book of Proceedings
Figure 7: Mean profiles of wind shear, ∂U/∂z (left column), total shear stress, τ m (middle column) and vertical
momentum diffusivity, KM (right column).The stability class is shown on the top of every figure. Dashed line is a linear
interpolation between sonic and LiDAR wind shear measurements.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 143/167
Book of Proceedings
For stable conditions (Fig. 7a), QNSE outperforms the other schemes but the shear is underestimated
below 100 m. YSU, MYNN and MYJ produce a more moderated shear whereas BouLac shows a
similar shear distribution to the expected for neutral or even unstable conditions. In order to quantify
the differences with respect to the FINO1 field observations, the root-mean-square error (RMSE) at
four heights and the average of the entire profile (9 heights from 61.5 to 251.5 m) are collected in
Table ¡Error! No se encuentra el origen de la referencia. for each stability class (hereafter RMSE is
also referred simply as error). Instead of directly using the wind shear, ∂U/∂z, we calculated an
equivalent magnitude, the wind speed difference, ∆U=U(z)-U41.5, which represents the variation of the
wind speed with height. We consider this magnitude more convenient since it is expressed in ms-1 and
an interpretation of the results can be easily performed. The previous comments are confirmed in
Table 3. QNSE and YSU have the lowests ∆Up (0.985 and 0.987 ms-1, respectively), closely followed
by MYNN. Further insight into the behaviour of PBL schemes can be achieved by examination of the
shear stress profiles. For stable conditions (Fig. 7b), there is an overestimation of measured total
shear stress, τm , which is maximum for BouLac. This is a consequence of a high level of eddy
diffusivity, Km (Fig. 7c). The BouLac scheme produces the largest level of Km, which yields to a high
level of shear stress. This induces an enhancement of the turbulent mixing in the lower part of the
boundary layer and thus, the suppression of the wind speed shear. This problem of the BouLac
scheme seems to be partly due to the formulation of Km which uses a constant value (=0.4) for the
stability-dependent term, φ (Km = φ⋅l⋅q0.5, being l the turbulent length scale and q the turbulent kinetic
energy). This lack of a link with stability makes it fail under stable conditions. Storm and Basu [8]
pointed out that YSU provides enhanced diffusion during stably stratified conditions. Similar behavior
was found for YSU in another onshore study carried out by Draxl, forecasting neutral conditions most
of the time independent from the observed atmospheric stability. In our case, we obtained low levels of
mixing and shear distributions that are more similar to observations thanks to the special treatment of
stable diffusion that we use [39], compared to previous results using the original version of the YSU
scheme [8, 9].
Under near-neutral conditions all the schemes satisfactorily reproduce the wind shear pattern except
for BouLac (Figs. 7d-f). The profiles of Km show a strong mixing of BouLac (also the lowest rate of
wind shear) while MYNN has the highest mixing among the others. From these results we would
expect MYNN to be the second most diffusive scheme. However, the wind speed shear depends also
on the level of total shear stress, τm, as it can be seen from Eq. ¡Error! No se encuentra el origen de
la referencia., and consequently, both parameters need to be analyzed simultaneously. MYNN and
QNSE have the lowest errors under near-neutral stratification (∼0.53 ms-1) with YSU also providing
satisfactory results (0.55 ms-1). To have an overall view of the performance of the different PBL
parameterizations, the RMSE of the mean wind speed profile, Up, has been also included in Table 3.
In this case, BouLac has the smallest Up, 2.05 ms-1, and the highest ∆Up, 0.77 ms-1. This evidences
the fact that Up can hide the error on the wind speed shear when the WRF-PBL profile crosses the
FINO1 measured profile, which may account for the bias between simulated profiles and observations.
Nonetheless, Up is a good complementary metric in order to determine the ability of the model to
follow observations.
A general good agreement is found for the unstable regime (Fig. 7g). Despite the fact that the PBL
schemes’ total stress is higher than the sonic measurements, the predicted wind shear is slightly
higher than the FINO1 observations. Similar distributions of Km are obtained (Fig. 7i), which explains
the overlapping of the different schemes in terms of wind speed shear. For convective scenarios,
BouLac minimizes the error (0.31 ms-1). We attributed this improvement to the fact that the constant
used for the stability-dependent term in Km seems to be more suited for convective regimes. MYNN
has a similar performance to BouLac, with a ∆Up of 0.32 ms-1. The worst results are obtained with
MYJ.
Under very unstable conditions the wind shear is extremely low. All the PBL schemes have more
pronounced gradients below 100 m, as it can be seen from Fig. 7j. In relation to the total shear
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 144/167
Book of Proceedings
stresses, FINO1 results reveal a quasi-uniform τm. This is a signature of the presence of the surface
layer, where the fluxes are expected not to vary with height. However, τm from the PBL schemes
decreases with height, as it occurs above the surface layer until the boundary layer height is reached.
This is a consequence of the physical model applied in mesoscale and generally in numerical models.
Forcing conditions in terms of fluxes of heat and momentum are imposed at the lowest level, which
are based on the Monin-Obukhov similarity theory. Hence, the top of the surface layer is forced to be
at the height of the first vertical grid point, while above, the PBL schemes provide the turbulent mixing
by distributing fluxes that linearly decrease with height. However, the surface layer extension can be
about 100 m in the case of a strong convective forcing. This issue would explain the lower wind shear
observed at FINO1, even if the rate of mixing decrease obtained by the PBL schemes is much
reduced than for any other stability regime. This effect has more importance for very unstable
conditions, where the surface layer is thicker.
The shape of the τm profile of YSU (Fig. 7k) is due to the non-local contribution (Eq. ¡Error! No se
encuentra el origen de la referencia.), which becomes more important for strongly convective
conditions and produces a very uniform wind speed profile above 100-120 m. MYNN provides the
lowest error (0.22 ms-2), YSU and BouLac showing similar results. From the momentum diffusivity
profiles presented in Fig. 7l, a more irregular Km profile is observed for MYJ. Such irregular Km profile
originates from a numerical instability which arises for extremely convective regimes. Numerical
instabilities of the MYJ scheme have already been observed by Pagowski\s\do5(2)004 in simplified
one-dimensional studies of convective conditions. In our computations, this instability causes sudden
drops to the minimum allowed threshold for TKE in the MYJ scheme (0.1 m2s-2) and consequently, in
the turbulent length scale. The eddy diffusivity drops and the shear stress gives extremely low values.
This instability is difficult to see directly from the wind speed, however, it can be observed in the wind
speed shear (irregular profile in the upper part, Fig. 7j). The effect of the numerical instability is partly
hidden when mean profiles are considered and mostly affects the turbulent kinetic energy and eddy
diffusivities. We tried to stabilize the computations by reducing the time-step to extremely low values
and to initialize MYJ with results from other schemes but the numerical instability developed in any
case. MYJ results presented in Sec. 5.1. where the instability showed up were not included.
6 Conclusions
In the present paper, the ability of five vertical turbulent flux parameterizations in the Advanced
Research WRF (ARW) v3.2: YSU, MYJ, QNSE, MYNN and BouLac, to account for the effects of
atmospheric stratification has been thoroughly evaluated at the offshore platform FINO1 for wind
energy purposes. Mesoscale simulations were performed using three nested domains and fine vertical
resolution in the lowest part of the boundary layer (∆z ∼ 10-20 m below 300 m). A comprehensive
validation based on turbulent fluxes, surface stability and wind profiles (up to 251.5 m) was carried out.
A spectral analysis of high frequency sonic anemometer measurements in terms of the Ogive function
revealed that 30 min is an appropriate averaging time, taver, to compute turbulent fluxes of heat and
momentum at FINO1. Larger intervals increase the risk of including non-steady effects without
reducing the errors and shorter intervals produce a dramatic increase of the error since they do not
properly account for low frequency structures, especially during convective stabilities. Turbulent fluxes
of momentum and heat allowed a detailed classification of stability based on the Obukhov length.
Surface friction velocity, u*s, is overpredicted by all the PBL schemes when the surface layer bottom
boundary condition is considered (z1 ∼ 12 m). The bias is nearly constant for the whole range of u*s
and the correlation is good for the five PBL schemes (r2 ∼ 0.83-0.86). The error is reduced by 0.020.05 ms-1 when u*41.5 is computed from the PBL momentum fluxes. This effect is due to the fact that
the FINO1 measurements are at 41.5 m and a decrease of the vertical flux is expected with height.
Regarding the sensible heat flux, the differences with respect to the FINO1 measurements grow with
the magnitude of <w'θ'>41.5. Under stable conditions there is a better agreement with the observations.
In general, we found that WRF-PBL schemes tend to forecast more moderated stable stratifications
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 145/167
Book of Proceedings
compared to measurements and a reinforcement of the instability under neutral and convective
regimes. However, QNSE, MYNN and YSU improve the stability prediction for stable conditions.
Stable conditions are the most complicated scenario for the PBL schemes to reproduce due to their
overdiffusive formulations. QNSE handles stable forcings better since it was developed for stable
stratification [51]. YSU with a revised stable PBL parameterizarion [39] provides also good results,
considerably improving the original performance of the scheme (not shown). More difficulties are
found in this regime for BouLac, even though this is also a TKE based scheme. We partly attribute this
issue to the use of a constant stability-dependent coefficient in the formulation of the eddy diffusivity.
For near neutral conditions the observed shear pattern is well reproduced by all the schemes with the
exception again of BouLac (highest Km and τm profile distributions). Under very unstable conditions,
the almost uniform wind speed profile proves difficult to reproduce by any of the formulations. In these
conditions, MYNN and YSU provide the best agreement, the latter due to its non-local formulation.
Differences below 100 m are partially due to the height of the surface layer in the simulations, imposed
by the first vertical grid point. Turbulent fluxes start to decrease from the second vertical grid point,
allowing wind speed gradients to appear. However, the uniform τm profile observed at FINO1 (41.581.5 m) produces a uniform wind speed profile. We have observed that the MYJ scheme develops a
numerical instability when strongly convective scenarios take place. Based on all of the validation
results and analysis, we found that MYNN provides the best results among the PBL schemes
evaluated here for the four different stability classes considered at the FINO1 offshore platform.
As a final remark, we want to emphasize the importance of a correct PBL formulation in order for
mesoscale models to be meaningful for the wind energy community. Even if the same formulation for
the surface layer could not be used, its influence was found to be minor. This can be clearly seen form
the results of MYJ and BouLac, for which a common surface layer formulation was used. As it has
been demonstrated here, stability effects are particularly important and cannot be properly evaluated if
measurements from a single height are considered. Profile measurements encompassing the rotor
swept area of operating wind turbines are a must, together with multi-height turbulent flux
measurements to adequately evaluate PBL formulations.
Acknowledgements
DME wants to acknowledge the support from the Marie Curie FP7-ITN-WAUDIT Project # 238576 and
from the Deutsches-Windenergie Institute during his research stay at DEWI Wilhelmshaven. The
FINO1 platform is one of three offshore platforms of the FINO Project, funded by the Federal Ministry
for the Environment, Nature Conservation and Nuclear Safety (BMU). The authors thank Cian
Desmond and Richard Foreman for the advice with the English language. We acknowledge our four
anonymous reviewers for their constructive comments that improved the quality of the manuscript.
References
[1] Wind in Our Sails: The Coming of Europe’s Offshore Wind Energy Industry, edited by S. Azau and
Z. Casey (European Wind Energy Association Report, Brussels, Belgium, 2011).
[2] Bechmann, N. N. Srensen, J. Berg, J. Mann, and P. E. R_ethor_e, “The Bolund experiment. Part
II: Blind comparison of microscale flow models,” Boundary-Layer Meteorol. 141, 245 (2011).
[3] Caldwell, H. N. S. Chin, D. C. Bader, and G. Bala, “Evaluation of a WRF dynamical downscaling
simulation over California,” Climatic Change 95, 499 (2009).
[4] Weisman L., C. Davis, W. Wang, K. W. Manning, and J. B. Klemp, “Experiences with 0-36-h
explicit convective forecasts with the WRF-ARW model,” Weather Forecast. 23, 407 (2008).
[5] Zngl, A. Gohm, and F. Obleitner, “The impact of the PBL scheme and vertical distribution of model
layers on simulations of Alpine foehn,” Meteorol. Atmos. Phys. 99, 105 (2008).
[6] Hu M., J. W. Nielsen-Gammon, and F. Zhang, “Evaluation of three Planetary Boundary Layer
schemes in the WRF model,” J. Appl. Meteorol. Climatol. 49, 1831 (2010).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 146/167
Book of Proceedings
[7] Shin H. and S. Y. Hong, “Intercomparison of planetary boundary-layer parameterizations in the
WRF model for a single day from CASES-99,” Boundary-Layer Meteorol. 139, 261 (2011).
[8] Storm and S. Basu, “The WRF model forecast-derived low-level wind shear climatology over the
United States Great Plains,” Energies 3, 258 (2010).
[9] Draxl, A. Hahmann, A. Pea, J. Nissen, and G. Giebel, “Validation of boundary-layer winds from
WRF mesoscale forecasts with application to wind energy forecasting,” in 19th Symposium on
Boundary Layer and Turbulence (Keystone, USA, 2010).
[10] Suselj K and A. Sood, “Improving the Mellor-Yamada-Janic Parametrization for wind conditions in
the marine boundary layer,” Boundary-Layer Meteorol. 136, 301 (2010).
[11] Neumann T, K. Nolopp, and K. Herklotz, “First operating experience with the FINO1 research
platform in the North Sea,” DEWI Mag. 24, 27 (2004).
[12] Cañadillas, A. Westerhellweg, and T. Neumann, “Testing the performance of a ground-based wind
LiDAR system: One year intercomparison at the offshore platform FINO1,” DEWI Mag. 38, 58
(2011).
[13] Vickers and L. Mahrt, “Quality control and flux sampling problems for tower and aircraft data,” J.
Atmos. Oceanic Technol. 14, 512 (1997).
[14] Cañadillas B, “A study of the marine boundary layer by LES-modelling and experimental
observations with focus on offshore wind energy applications,” Ph.D. dissertation (University of
Hannover, 2010).
[15] Lee, J. Finnigan, and K. T. Paw, “Coordinate systems and flux bias error,” in Handbook of
Micrometeorology: A Guide for Surface Flux Measurement and Analysis, edited by X. Lee, W. J.
Massman, and B. E. Law (Kluwer Academic, The Netherlands, 2004).
[16] Foken, F. Wimmer, M. Mauder, C. Thomas, and C. Liebethal, “Some aspects of the energy
balance closure problem,” Atmos. Chem. Phys. Discuss. 6, 3381 (2006).
[17] Sun M, Z. L. Zhu, X. F. Wen, G. F. Yuan, and G. R. Yu, “The impact of averaging period on eddy
fluxes observed at ChinaFLUX sites,” Agric. Forest Meteorol. 137, 188 (2006).
[18] Xiao X, H. C. Zuo, Q. D. Yang, S. J. Wang, L. J. Wang, J. W. Chen, B. L. Chen, and B. D. Zhang,
“On the factors influencing surface-layer energy balance closure and their seasonal variability over
semiarid loess plateau of Northwest China,” Hydrol. Earth Syst. Sci. Discuss. 8, 555 (2011).
[19] Oncley P, J. A. Businger, C. A. Friehe, J. C. LaRue, E. C. Itsweire, and S. S. Chang, “Surface
layer profiles and turbulence measurements over uniform land under near-neutral conditions,” in
9th Symposium on Boundary Layers and Turbulence (1990), pp. 237–240.
[20] Drennan W. M. and L. K. Shay, “On the variability of the fluxes of momentum and sensible heat,”
Boundary-Layer Meteorology 119, 81 (2006).
[21] Hgstrm U., E. Sahl_ee, W. M. Drennan, K. K. Kahma, A. S. Smedman, C. Johansson, H.
Pettersson, A. Rutgersson, L. Tuomi, F. Zhang, and C. Johansson, “Momentum fluxes and wind
gradients in the marine boundary layer- a multiplatform study,” Boreal Environ. Res. 13, 475
(2008).
[22] Nilsson E. O., A. Rutgersson, and P. P. Sullivan, “Flux attenuation due to sensor displacement
over sea,” J. Atmos. Oceanic Technol. 27, 856 (2010).
[23] Oh H. M., K. E. Kim, K. J. Ha, L. Mahrt, and J. S. Shim, “Quality control and tilt correction effects
on the turbulent fluxes observed at an ocean platform,” J. Appl. Meteorol. Climatol. 50, 700 (2011).
[24] Charnock H., “Wind stress on a water surface,” Q. J. R. Meteorol. Soc. 81, 639 (1955).
[25] Stull R. B., An Introduction to Boundary Layer Meteorology (Kluwer Academic, Dordrecht, 1988).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 147/167
Book of Proceedings
[26] van Wijk A. J. M., A. C. M. Beljaars, A. A. M. Holtslag, and W. C. Turkenburg, “Evaluation of
stability corrections in wind speed profiles over the North Sea,” J. Wind Eng. Ind. Aerodyn. 33, 551
(1990).
[27] Skamarock W., J. Klemp, J. Dudhia, D. Gill, D. Barker, M. Duda, X. Huang, W. Wang, and J.
Powers, A Description of the Advanced Research WRF Version 3, NCAR Tech. Note NCAR/TN475þSTR (2008).
[28] Hong S. Y., J. Dudhia, and S. H. Chen, “A revised approach to ice microphysical processes for the
bulk parameterization of clouds and precipitation,” Mon. Weather Rev. 132, 103 (2004).
[29] Mlawer E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, “Radiative transfer for
inhomogeneous atmosphere: RRTM, a validated correlated-k model for the long-wave,” J.
Geophys. Res. 120, 16663, doi:10.1029/97JD00237 (1997).
[30] Dudhia J., “Numerical study of convection observed during the winter monsoon experiment using
a mesoscale twodimensional model,” J. Atmos. Sci. 46, 3077 (1989).
[31] Chen F. and J. Dudhia, “Coupling an advanced land-surface/hydrology model with the Penn
State/NCAR MM5 modeling system. Part I: Model description and implementation,” Mon. Weather
Rev. 129, 569 (2001).
[32] Kain J. S. and J. M. Fritsch, “A one-dimensional entraining/detraining plume model and its
application in convective parameterization,” J. Atmos. Sci. 47, 2784 (1990).
[33] Hong S. Y., Y. Noh, and J. Dudhia, “A new vertical package with an explicit treatment of
entrainment processes,” Mon. Weather Rev. 134, 2318 (2006).
[34] Janic Z. I., Nonsingular Implementation of the Mellor-Yamada Level 2.5 Scheme in the NCEP
Meso Model (Office Note 437, National Centers for Environmental Prediction, Camp Springs, MD,
USA, 2001).
[35] Sukoransky S., B. Galperin, and A. V. Perov, “Application of a new spectral theory of stably
stratified tur-bulence to the atmospheric boundary layer over sea ice,” Boundary-Layer Meteorol.
117, 231 (2005).
[36] Nakanishi M., “Improvement of the Meyor-Yamada turbulence closure model based on LargeEddy Simulation data,” Boundary-Layer Meteorol. 99, 349 (2001).
[37] Nakanishi M. and H. Niino, “An improved Mellor-Yamada level-3 model with condensation physics:
Its design and verification,”Boundary-Layer Meteorol. 112, 1 (2004).
[38] Bougeault P. and P. Lacarrre, “Parameterization of orography-induced turbulence in a mesobetascale model,” Mon. Weather Rev. 117, 1872 (1989).
[39] Hong S. Y., “A new stable boundary-layer mixing scheme and its impact on the simulated East
Asian summer monsoon,”Q. J. R. Meteorol. Soc. 136, 1481 (2010).
[40] Monin A. S. and A. M. Obukhov, “Basic turbulence mixing laws in the atmospheric surface layer,”
Tr. Inst. Teor. Geofiz. Akad. SSSR 24, 163 (1954).
[41] Suranjana S. et al., “The NCEP climate forecast system reanalysis,” Bull. Am. Meteorol. Soc. 91,
1015 (2010).
[42] Reynolds R. W. and D. B. Chelton, “Comparisons of dayly sea surface temperature analysis for
2007-08,” J. Clim. 23, 3545 (2010).
[43] Skamarock W. C., “Evaluating mesoscale NWP models using kinetic energy spectra,” Mon.
Weather Rev. 132, 3019 (2004).
[44] Zhang D. and R. A. Anthes, “A high resolution model of the planetary boundary layer-sensitivity
tests and comparison with SESAME-79 data,” J. Appl. Meteorol. 21, 1594 (1982).
[45] Janic Z. I., “The step-mountain coordinate: Physics package,” Mon. Weather Rev. 118, 1429
(1990).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 148/167
Book of Proceedings
[46] Shin H. H. and S. Y. Hong, “Impacts of the lowest model level height on the performance of
planetary boundary layer parameterizations,” Mon. Weather Rev. 140, 664 (2012).
[47] Peña A. and A. Hahmann, “Atmospheric stability and turbulence fluxes at Horns Rev—An
intercomparison of sonic, bulk and WRF model data,” Wind Energy 15, 717 (2011).
[48] Steeneveld G. J., L. F. Tolk, A. F. Moene, O. K. Hartogenesis, W. Peters, and A. A. M. Holtslag,
“Confronting the WRF and RAMS mesoscale models with innovative observations in the
Netherlands: Evaluating the boundary layer heat buget,” J. Geophys. Res. 116, D23114,
doi:10.1029/2011JD016303 (2011).
[49] Mauder M., C. Libethal, M. Gckede, J. P. Leps, F. Beyrich, and T. Foken, “Processing and quality
control of flux data during LITFASS-2003,” Boundary-Layer Meteorol. 121, 67 (2006).
[50] Pagowski M., “Some comments on PBL parameterizations in WRF,” in Joint WRF-MM5 Users
Workshop, National Center for Atmospheric Research, Boulder, CO, 2004.
[51] Sukoriansky S., B. Galperin, and I. Staroselsky, “A quasinormal scale elimination model of
turbulent flows with stable stratification,” Phys. Fluids 17, 085107 (2005).
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 149/167
Book of Proceedings
Wind-Farm Parametrisations in Mesoscale Models
P.J.H. Volker a,1, J. Badgera, A.N. Hahmanna and K.S. Hansena
a
DTU Wind Energy, Roskilde, Denmark Germany
Presented at the International Conference on Aerodynamics of Offshore Wind Energy Systems and Wakes ICOWES 2013,
Copenhaguen, Denmark, June 2013.
Abstract
In this paper we compare three wind-farm parametrisations for mesoscale models against
measurement data from the Horns Rev I offshore wind-farm. The parametrisations vary from a simple
rotor drag method, to more sophisticated models. Additional to Volker [5] we investigated the
horizontal resolution dependency of the considered approaches.
1. Introduction
The offshore wind-farm technology has matured significantly in the past decade. The current largest
installed wind farm, the London array phase I, with a nominal capacity of 630MW, is almost four times
larger than the Horns Rev I (160MW) wind farm, which began to operate in 2005. The North Sea area,
which is at the moment the most lucrative region for offshore wind farms, is however limited. Therefore
it will become increasingly important to study the effect of large wind farms on the atmosphere from an
economical (wind farm efficiency, wind farm interaction), as well as from an ecological point of view.
We use in this study the mesoscale model for it’s ability to take atmospheric conditions into account,
which can influence the wake extension. Its drawback is the lack in resolution compared to
Computational Fluid Dynamics (CFD) models, implying that it is not possible to simulate single turbine
wakes. Instead a wind farm parametrisation should model the average effects of wind turbines inside a
mesoscale grid cell.
Three wind farm schemes will be analysed. The first approach, hereafter ROTOR-DRAG, adds only a
drag force to the flow, which is proportional to the turbine blade area intersection with the model grid
level. The second parametrisation, hereafter referred to as WRF-WF, is included in the Weather
Research and Forecast Model (WRF) [4], a publicly available open source model written in Fortran. It
applies a drag and a Turbulence Kinetic Energy (TKE) source term to the flow at each rotor
intersecting vertical grid level. The third wind farm scheme, Explicit Wake Parametrisation (EWP) [5],
has been developed at the DTU Wind Energy Department. Similarly to the other approaches, in the
EWP approach also a drag force is added to the momentum budget. However, it accounts for the
unresolved wake expansion. Here we assume that the unresolved wake expansion can be described
by a turbulence diffusion process [6]. The parametrisation has been implemented in the WRF model.
In a previous study [5] we analysed the WRF-WF and EWP approach against long term
measurements from the offshore wind farm Horns Rev I. We found that the additional TKE in the
WRF-WF approach caused an intensive mixing zone, leading to a too fast velocity deficit recovery
close to the wind farm. Furthermore, the scheme was found to be vertical resolution dependent,
1
Corresponding author: Patrick J. H. Volker
DTU Wind Energy, Technical University of Denmark, Department of Wind Energy, DTU Risø Campus, Frederiksborgvej 399,
Building 115, 4000 Roskilde, Denmark: Tel.: + 45 46775066, e-mail: [email protected]
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 150/167
Book of Proceedings
consequently the energy extracted from the flow will vary with vertical resolution. In this article we
extend the previous study by analysing the horizontal dependence of both parametrisations. This is
done by simulating the same wind farm size with a 1.12 km, 1.68km and 2.24km horizontal resolution.
We verify the wind farm parametrisations against long term in-situ measurements from the Horns Rev
I wind farm. The mesoscale model was run in the idealised case mode with 40 vertical layers. The
vertical resolution was in the planetary boundary layer (PBL) in the order of 10m (the five lowest mass
points were on 10, 30, 50 ,71 and 92 meters respectively).
2. Mesoscale Model
Mesoscale models are designed to forecast weather phenomena with typical length scales down to
around 10 km. Therefore, to limit computational costs, a relatively coarse horizontal resolution in the
order of kilometres is required. The model’s vertical resolution is in the Planetary Boundary Layer
(PBL) typically in the order of decametres to allow the vertical temperature and moisture structure to
be resolved sufficiently. Mesoscale models are intended to resolve, similarly to Reynolds Averaged
Navier-Stokes (RANS) models, only the mean flow, whereas the turbulence part of the spectrum is
completely modelled. The basic assumption is that there is a the scale separation between the
resolved mesoscale processes and the unresolved turbulent ones, since no explicit filtering is applied.
This means that the solution will not converge to the expected value with horizontal grid refinement,
since from a certain horizontal resolution onwards, double counting will take place. Mesoscale models
are generally non hydrostatic and fully compressible. This means that they contain a prognostic
equation for each of the three wind velocity components and a complete continuity equation.
Furthermore, they contain a prognostic equation for the temperature as well as for all moisture
components. Finally, the pressure is obtained via the equation of state. The time step used in the
prognostic equations is determined by the Courant number, which is a function of the advection
velocity and the horizontal grid size. The lower boundary values are, over land, provided by soil
(diffusion) models and over water they are generally obtained from reanalysis data.
2.1 Parametrisation of Wind Turbines
All unresolved processes need to be parametrised. Examples are local and non-local (convection)
turbulent transport, turbulent surface fluxes, moisture phase changes and radiation. Since D0 << x,
where D0 is the wind turbine diameter and x the horizontal grid spacing, the effect of wind turbines is
also unresolved. On the other hand, since D0 > z, where z is the vertical grid spacing, the vertical
turbine-induced wake structure can be described. Due to the coarse horizontal resolution we will
typically find several turbines in one mesoscale grid cell. It is therefore not possible to resolve single
turbine wakes and hence the interaction between single turbines is not accounted for in the model.
Instead the parametrisation should apply a grid cell averaged deceleration, accounting for the average
impact of all the grid cells containing turbines. After that point, the mesoscale model is intended to
simulate the wind farm wake. We aim of describing the grid cell average deceleration as accurately as
possible, since the wake expansion within a mesoscale grid-cell is expected to be considerable. The
energy extracted from a single turbine can be modelled by adding an additional force in the opposite
flow direction to the velocity balance equation. For a compressible fluid, neglecting viscous effects, the
most general form of the RANS equation reads
∂U i
∂U i ∂ui u j
+Uj
+
= Fi .
∂t
∂x j
∂x j
(1)
Here, we use capital letters for mean quantities and lower-case letters for the fluctuations. The index i
represents the x; y and z direction. All the forcing contributions are incorporated in the force (per
mass) term Fi on the right hand side, such as the pressure gradient force or Coriolis force. Also the
turbine induced drag force, FDi , is part of the forcing term Fi.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 151/167
Book of Proceedings
2.2. Rotor Drag
In this approach, a drag force is applied to the horizontal components of Eq. (1) at every turbine blade
intersecting model level k. The additional thrust force in the horizontal direction i reads
FDi ( k ) =
CT N A( k ) Ui ( k ) U (k )
2 ( ∆ x ) ∆ z (k )
2
,
(2)
where CT is the thrust coefficient, N the number of turbines per grid cell, Ak the turbine blade area
intersecting with model level k and Ui(k) the horizontal wind velocity component in the direction i = x; y.
The absolute horizontal velocity at level k is U ( k ) = U (k )2 + V (k )2 , where U(k) and V(k) are the
horizontal velocity components in the x and y direction, respectively. ∆x and ∆zk denote the horizontal
and vertical grid spacing at level k. In this way the sub-grid velocity deficit expansion in the vertical
direction is neglected, since the vertical velocity deficit is restricted to the turbine blade extension.
2.3. WRF-WF
The WRF-WF scheme is from version 3.2.1 included in the WRF model. This parametrisation applies
the drag force from Eq. (2), as well as an additional source term of TKE, to the flow. The total TKE
applied to the model level k reads
ui fDi =
Ce N ∑ k A(k ) U ( k )
2 ( ∆x ) ∆z (k )
2
3
.
(3)
Here we used Ce for the factor of proportionality, which is equal to CT - Cp, where CP is the power
coefficient. The additional TKE will lead to an increased turbulence (diffusion) coefficient for
momentum K m = q l S( m ) , where q = 2TKE is the turbulence velocity, l the turbulence length scale
and S(m) a stability correction for momentum. In this way the vertical sub-grid-scale wake expansion is
obtained implicitly.
2.4. Explicit Wake Parametrisation
In the Explicit Wake Parametrisation (EWP) [5], it has been assumed that the directed turbine-bladeinduced turbulence is small compared to the shear production in the wake flow. The downstream subgrid scale velocity deficit development is described explicitly by a turbulence diffusion process.
The velocity deficit is assumed to be approximately Gaussian shaped and it is described by a
characteristic length and velocity scale. From a one dimensional diffusion equation we obtain the
characteristic length scale,
 2K m
 U0
σ2 = 

2
x +σ0 ,

(4)
where σ0 is the initial length scale and U0 the hub height velocity. The one dimensional diffusion has a
Gaussian distribution as analytical solution. If this is used, we obtain for the velocity deficit
∆U ( z) = Us
1  z− h 
− 

2 l 
e
2
= U s f (z,σ ), consequently U = U0 − Us f ( z,σ ) ,
(5)
where h is the turbine hub height and Us the maximum velocity deficit. The function f(z, σ) ≡ f
expresses the velocity deficit distribution. The maximum velocity deficit Us can be obtained from the
total thrust equation. This gives
1
ρCT π R02U02 = ρ ∫ dy
2
∆y
WAUDIT
Contract#
238576
zmax
∫
U 0 (U0 − U ) dz ,
(6)
0
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 152/167
Book of Proceedings
where ρ is the atmospheric density, R0 the turbine blade radius, y the horizontal direction
perpendicular to the flow and ∆y the model grid spacing in the y direction. Combining Eq. (5) and Eq.
(6), integrating over the domain height zmax gives for the maximum velocity deficit
Us =
π CT R02U0
.
2 ∆y σ
(7)
Where σ is the grid-cell averaged length scale. Eq. (7), the right hand side of Eq. (6) and Eq. (5)
describe the velocity deficit completely.
3. Measurement Data
For the model evaluation we used the measurements from the Horns Rev I wind farm, see Fig. ??.
The Horns Rev I wind farm consists of 80 2MW wind turbines, each with a blade diameter of 80m and
a hub at 70m above sea level. The turbines are arranged in 10 columns from East to West and 8 rows
from South to North. The turbine spacing is 560m in the West to East direction. Two masts are
installed to the East of the wind farm. The masts M6 and M7 are located 2000m and 6000m
downstream, see Fig.1. The masts contain cup anemometers at 70m height A full description of the
wind farm layout is given in [1]. For the model validation only measurements from the year 2005 to
2009 under neutral atmospheric conditions have been selected. The wind direction interval is from
255º ≤ θ ≤ 285º and the wind speed interval from 7.5ms-1 to 8.5ms-1. The Northerly wind direction is
defined to be at 0º and the rotation is in the clockwise direction. The neutral conditions are defined for
Monin-Obukhov lengths |L| > 500 m.
Figure 1: Location of the Horns Rev I wind farm and the meteorological masts.
4. Model Set-Up
We used the WRF model in idealised case mode with open boundaries and no surface heat fluxes.
The model is initialised with a constant Geostrophic wind, which converges to U = 7.97ms-1 and V =
0.09ms-1 (θ = 269.4º) at hub-height. In this model configuration the Coriolis force acts on the velocity
perturbation from the initial condition. The dry atmosphere converges to a neutral temperature profile
with an inversion height at around 700 m. The cloud micro-physics, as well as the convection scheme,
were turned off. For all simulations the MYNN (1.5) PBL turbulence diffusion scheme [3] has been
used as required by the WRF-WF parametrisation. The model was set-up with 80 x 30 grid cells in the
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 153/167
Book of Proceedings
horizontal direction. In the vertical direction we used 40 layers. The lowest layers were at 10, 30, 50,
71 and 92m respectively. To study the horizontal dependence of the three wind farm parametrisations,
we ran the WRF model with a 1120m (R1120), 1680m (R1680) and 2240m (R2240) horizontal grid
spacing, respectively. The model set-up is summarised in Table 1.
Table 1: WRFV3.4 simulation set up
In the evaluation the velocities from a single model simulation are evaluated, against the wind
direction averaged wind speed of the met masts.
4.1. Wind-Farm Layout
The wind farm’s layout for the three simulations is depicted in Fig. 2. The number of turbines for all
simulations is equal to 80, distributed in 8 rows and 10 columns. The number of turbine-containing
grid-cells is 20, 12 and 6, for the R1120, R1680 and R2240 simulation, respectively.
Figure 2: In this figure the different wind farm layouts are plotted. From the left side to the right side we plotted the
wind farm layout for the R1120, R1680 and R2240 simulation.
We use the dotted lines for the grid cells with an unchanged turbine density. The turbine density per
grid-cell is constant in the R1120 simulation. However, it is not constant for the R1680 and R2240
simulations.
4.2. Wind-Farm Scheme Adjustments
WRF-WF scheme
The WRF-WF scheme uses by default an empirical power curve for obtaining the power coefficient CP.
The thrust coefficient is afterwards derived by the empirical relationship CT = min(7Cp/4; 0.9). To
guarantee a scheme independent thrust, we use the thrust coefficient from the Vestas V80 thrust
curve as in the EWP scheme. Thereafter, the inverse relationship from above is applied to obtain the
power coefficient, Cp = CT/1.75. In this way the ratio between the thrust and power coefficient remains
unchanged in the WRF-WF parametrisation.
EWP scheme
The initial length scale has been set to l0 = 1.5R0. Here we included viscous effects and vertical
meandering to the inviscid fluid solution l0 = R0.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 154/167
Book of Proceedings
5. Wind Farm Evaluation
In Sect. 5.1 we evaluate the model hub-height velocity field for the different approaches and horizontal
resolutions, against the wind direction averaged velocities obtained from cup anemometers mounted
at 70 m. In Sect. 5.2 we analyse quantitatively the horizontal resolution dependence of the modelled
wind farm wake velocity field.
In this study we use the near and far wind farm wake for a 10% and 5% velocity reduction,
respectively. This definition was made for terminology convenience only.
Figure 3: Normalised velocity recovery plot for the ROTOR-DRAG, WRF-WF and EWP scheme. The mast measurements
are represented by the diamonds. a) ROTOR-DRAG, b) WRF-WF and c) EWP approach. The error bars represent the
standard deviation σ of the measurements.
5.1. Near Wind-Farm Wake
In Fig. 3 we show the normalised hub-height velocity for the three approaches. We use Uh for the
downstream grid cell averaged hub height velocity and U0h for the wind farm upstream velocity at hub
height. The model upstream velocity is obtained from the reference run without wind farm, whereas
the measured up-stream velocity is from the upstream reference turbine. The bars indicate the wind
speed standard deviations. The normalised velocity within the wind farm has been obtained by
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 155/167
Book of Proceedings
averaging over all turbine containing grid-cells in the cross-stream direction, i.e. for the R1120, R1680
and R2240 simulation we averaged over 4, 3 and 2 rows respectively (Fig. 4). Figure 4 shows the
normalised velocity field for the ROTOR-DRAG, WRF-WF and the EWP approach. From Fig. 4a we
find that in the wind farm wake, the modelled normalised velocity has a low bias compared to the longterm averaged measurements, when a drag source is applied. Furthermore, we find a horizontal gridsize dependence. The hub-height velocity fields for the R1680 and R2240 simulations are similar.
Whereas the velocity field for the higher resolution run shows a slightly lower normalised velocity field
at the end of the wind farm, at a distance of around 5000m and a higher normalised velocity in the
near wind farm wake. The maximum difference in normalised velocity is 3% at the end of the wind
farm, at mast 7 the difference between the simulations is reduced to less than 1%. From Fig. 4b, we
find that for the WRF-WF approach the simulated normalised velocity at hub-height has a positive
bias, compared to the measurements. We notice also that the velocity field is less sensitive to a
variation in horizontal resolution. Similarly to the ROTOR-DRAG approach, the velocity fields for the
lower resolution runs are almost identical within the wind farm and in the near wake. The velocity field
with the higher horizontal resolution is again slightly lower within the wind farm and slightly higher
closely after mast M6. The differences are always less than 1%. Figure 4c shows for the EWP
approach a higher modelled normalised velocity at mast M6 and a slightly lower one at mast M7.
Regarding the horizontal resolution dependence we find within the wind farm that the velocity deficit
increases with in increase in the resolution. Between M6 and M7 the velocity fields become horizontal
resolution independent. The difference in normalised velocity between the R1120 and R2240
simulation is slightly higher at the end of the wind farm.
Figure 4: Velocity deficit recovery plot for the ROTOR-DRAG approach. For the R1120, R1680 and R2240 simulation
respectively.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 156/167
Book of Proceedings
5.2. Wind-Farm Wake
In Fig. 3a, Fig. 3b and Fig. 3c, we plotted the normalised velocities at hub-height for the ROTORDRAG, WRF-WF and EWP approaches, respectively. The velocity from the wind farm simulation is
again normalised by the velocity from the reference simulation. We show always from the top to the
bottom panel, the results from the R1120, R1680 and R2240 simulations.
5.3. ROTOR-DRAG Approach
From Fig. 4 we find that the near wake extends to around 7km downstream in all simulations. The far
wind farm wake extension becomes resolution dependent. This results from the decreasing wake
width in the cross-wind direction, which can not be resolved anymore with a lower model resolution.
WRF-WF Scheme
In Fig. 5, where the normalised velocity for the WRF-WF has been plotted, we see that the near wake
velocity reduction varies almost 40% (around 1.7km in the R1680 run and around 2.8km in the R2240
simulation). Although the maximum velocity reduction in the WRF-WF approach is around 15%, which
is around 10% less than with the ROTORDRAG approach, we find that the wind farm induced wake
extends further downstream for the WRF-WF approach. This implies that the velocity restoring
mechanisms differ for both approaches. Similarly to the results from the ROTOR-DRAG approach we
find that the wake width in the far wake is not resolved anymore in the lower resolution simulations.
Figure 5: Identical to Fig. 4, this time for the WRF-WF approach.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 157/167
Book of Proceedings
EWP Scheme
We find from Fig. 6 that the near wake extension varies by circa 6%, between the R1120 and R2240
simulation (around 8.0km in the R1120 compared to 7.5,km in the R2240).
The contour lines for a 2% velocity reduction are almost identical to those one of the WRF-WF
scheme. The wake orientation differs however, see [5].
Figure 6: Identical to Fig. 4, this time for the EWP scheme.
6. Discussion & Conclusion
We analyse three wind farm approaches (ROTOR-DRAG, WRF-WF and EWP) against long term
averaged measurements from Horns Rev I.
For the near wind farm wake, Fig. 3, we found that compared to mast measurements the velocity
reduction was overestimated in the ROTOR-DRAG approach and underestimated in the WRF-WF
approach. For the EWP approach we found an underestimation of the velocity reduction at mast M6,
2000m downstream from the wind farm and a slightly overestimated velocity reduction at mast M7,
6000m downstream. Furthermore, we found that the ROTOR-DRAG approach was the most sensitive
to the horizontal resolution variation within the wind farm. The near wake extension was however
almost horizontal resolution independent. The EWP scheme is shown to be sensitive to a horizontal
resolution variation within the wind farm (1% difference). In the wind farm wake, between M6 and M7,
the normalised velocity converged. The near wake extension varied around 6% between the R1120
and R2240 simulation. The horizontal resolution sensitivity within the wind farm was the smallest for
the WRF-WF. However, the near wind farm wake extension differed by around 40% between the
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 158/167
Book of Proceedings
R1680 and R2240 simulation. Regarding the far wake extension, we found that the wake width in the
horizontal cross-stream direction becomes too small to be resolved with a decreasing horizontal
resolution. This leads to a reduced wind farm wake extension in the R2240 simulation for all the
schemes.
Acknowledgements
Research has been funded in by the FP7-WAUDIT project, no 238576, and the EERA-DTOC FP7ENERGY-2011-1/no 282797, and we would like to acknowledge Vattenfall AB and DONG Energy A/S
for using data from the Horns Rev offshore wind-farm.
References
[1] Hansen, K. S. Presentation of Horns Rev offshore wind farm and the Vestas V80 Wind turbine.
Report: Eera-Dtoc, 2012.
[2] Tennekes, H., and Lumley J. L.. A First Course in Turbulence. The MIT Pess, 1972.
[3] Nakanishi, M., and Niino, H.. Development of an improved turbulence closure model for the
atmospheric boundary layer. J of the meteorol Soc of Jpn, 87:895-912, 2009.
[4] Skamarock, W., Klemp J., Dudhia J., Gill, D., Barker, D., Duda M., Huang X., Wang W., and
Powers, J,. A Description of the Advanced Research WRF Version 3. NCAR Technical note, 2008.
[5] Volker, P. J., Badger, J., Hahmann, A. N., Ott, S. Implementation and evaluation of a wind farm
parametrisation in a mesoscale model. Sumitted, 2013.
[6] Wyngaard, J. C. Atmosheric Turbulence. Cambridge Press, 2010.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 159/167
Book of Proceedings
Other publications from the WAUDIT network
Journal papers
[1] Badger J, Volker PJH, Prospathospoulos J, Sieros G, Ott S, Rethoré P-E, Hahmann AN, Hasager
C (2013) Wake modelling combining mesoscale and microscale models. Submitted to Renewable
Energy.
[2] Borden Z, Koblitz T, Meiburg E (2012) Turbulent mixing and wave radiation in non-Boussinesq
internal bores. Physics of Fluids 24(8): 82-106.
[3] Chávez R, Lozano S, Correia P, Sanz Rodrigo J, Probst O (2013) On the application of the
Principal Component Analysis for an efficient climate downscaling of surface wind fields. Energy
Procedia 40: 67-76.
[4] Conan B, van Beeck J, Aubrun S (2012) Sand Erosion Technique Applied to Wind Resource
Assessment. Journal of Wind Engineering and Industrial Aerodynamics 104: 322-329.
[5] Desmond C, Watson S (2012) A study of stability effects in forested terrain. Journal of Physics:
Conference Series and Proceedings of The Science of Making Torque from Wind 2012,
Oldenburg, 9-11 October 2012. In press for Journal of Physics
[6] Desmond C, Watson S, Aubrun S, Avila S, Hancock P, Sayer A (2013) A study on the inclusion of
forest canopy morphology data in numerical simulations for the purpose of wind resource
assessment. Journal of Wind Engineering and Industrial Aerodynamics, Under review.
[7] Duraisamy JV, Dupont E, Carissimo B (2012) Downscaling wind energy resource from mesoscale
to microscale model and data assimilating field measurements. Journal of Physics: Conference
Series and Proceedings of The Science of Making Torque from Wind 2012, Oldenburg, Germany,
October 9-11. Accepted for publication.
[8] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2011a) Scaling Of Turbulence In The
Atmospheric Surface-Layer: Which Anisotropy? Journal of Physics: Conference Series, 318(7),
072008. doi:10.1088/1742-6596/318/7/072008.
[9] Koblitz T, Bechmann A, Sogachev A, Sørensen N, Réthoré P-E (2013) CFD model of stratified
atmospheric boundary-layer flow. Wind Energy, accepted for publication.
[10] Muñoz-Esparza D, Canadillas B, Neumann T, van Beeck J (2012) Turbulent fluxes, stability and
shear in the offshore environment: Mesoscale modelling and field observations at FINO1. J.
Renewable Sustainable Energy 4, 063136, http://dx.doi.org/10.1063/1.4769201
[11] Santos-Alamillos FJ, Pozo-Vázquez D, Ruiz-Arias JA, Lara-Fanego V, Tovar-Pescador J (2013)
Analysis of WRF model wind estimate sensitivity to physics parameterization choice and terrain
representation in Andalusia (Southern Spain). Journal of Applied Meteorology and Climatology
52(7): 1592–1609.
[12] Sanz Rodrigo J, Borbón Guillén F, Gómez Arranz P, Courtney MS, Wagner R, Dupont E (2013)
Multi-Site Testing and Evaluation of Remote Sensing Instruments for Wind Energy Applications.
Renewable Energy 53: 200-210.
[13] Sanz Rodrigo J, Lozano Galiana S, Fernandes Correia PM, Cantero Nouqueret E, García Hevia
B, Stathopoulos C, Borbón F, Chávez Arroyo RA, Gancarski P, Koblitz T, Barranger N, Conan B
(2013) Benchmarking of wind resource assessment flow models. The Alaiz complex terrain test
case. To be submitted to Renewable & Sustainable Energy Reviews.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 160/167
Book of Proceedings
[14] Stathopoulos C, Kaperoni A, Galanis G, Kallos G (2013) Wind power prediction based on
numerical and statistical models. Journal of Wind Energy & Industrial Aerodynamics. 112 (2013)
25-38.
[15] Volker PJH, Badger J, Hahmann AH, Ott S (2013) Implementation and evaluation of a wind farm
parametrisation in a mesoscale model. Submitted to Boundary-Layer Meteorol.
[16] Volker PJH, Huang H-Y, Capps SB, Badger J, Hahmann AH, Larsén X (2013) Lower Marine
Atmosphere Response to a Large Offshore Wind Farm. To be submitted to J. Geophys. Res.
[17] WAUDIT (2013) WAUDIT Book of Proceedings. Edited by J. Sanz Rodrigo, CENER. To be
published
[18] Yeow TS, Cuerva A, Pérez-Alvarez J (2013) Reproducing the Bolund Experiment in Wind Tunnel.
Wind Energy, accepted for publication.
Technical Reports
[19] Borbón Guillén F, Courtney M, Dupont E, Lefranc Y, and Girard R (2012) Results of the
measurement campaign in complex terrain - Assessment. Safewind deliverable Dc-2.3. Technical
report, CENER, Spain, June 2012.
[20] Holmes H (2011) Quality Assurance of Wind Resource Assessment Models Part 1: Background
Information, WAUDIT “Standardization” Workshop, University of Hamburg, 19 May 2011.
[21] Holmes H (2011) Quality assurance of wind energy assessment models: Part I. Verification &
Validation Background. WAUDIT Guidance Report WP6-D24, May 2011.
[22] Santos-Muñoz D, Martin ML, Pascual A, Valero F (2013) Probabilistic wind speed forecast based
on multi-physic ensemble prediction systems. Join SRNWP Workshop Physical Parmetrizations
and Ensemble Predictions Systems. Madrid, Spain, June 2013.
[23] Sanz Rodrigo J (2010) State-of-the-Art of Wind Resource Assessment. Results of the WAUDIT
questionnaire. Deliverable D7 of the FP7-WAUDIT project, grant agreement number 238576, 24
pp, January 2010.
(2013) Benchmarking of wind resource assessment flow models. The Alaiz complex terrain test
case. Deliverable D26 of the FP7-WAUDIT project, grant agreement number 238576, 35 pp,
October 2013.
Conference/Workshop Reports
[25] Albuquerque I (2010) Forest Winds in complex terrain. 6th PhD Seminar on Wind Energy in
Europe, Trondheim, Norway, September 2010.
[26] Albuquerque I, Sanz Rodrigo J, Landberg L, Watson S (2011) Exploring Several Turbulent
Closure Methods for Simulating Forest Winds in Complex Terrain. EWEA-2011, Brussels,
Belgium, March 2011.
[27] Albuquerque I, Sanz Rodrigo J, Landberg L, Watson S (2011) Comparison between k-epsilon
and Baseline turbulence closure models simulating forest winds in complex terrain. PhD
Research Conference, Department of Electronic and Electrical Engineering, Loughborough
University, UK, May 2011.
[28] Barranger N (2010) Assessing wind energy potential using the high resolution meso-scale model
RAMS. 6th EAWE PhD Seminar, NTNU, Trondheim, Norway, September 2011.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 161/167
Book of Proceedings
[29] Barranger N (2011) Evaluation of turbulence scheme of the mesoscale model RAMS for high
resolution wind resource assessment. 7th EAWE PhD Seminar, TU Delft, Delft, Netherlands.
[30] Barranger N, Kallos G (2012) The use of an atmospheric model to solve wind forcing turbulent
flows over complex terrain for wind resource assessment. Energy meteorology session of the
EMS general assembly, Łódź.
[31] Barranger N, Stathopoulos C, Kallos G (2013) Dynamical downscaling in complex terrain using
the Mesoscale Model RAMS. EWEA-2013, Vienna.
[32] Barranger N, Stathopoulos C, Kallos G (2013) High resolution topography and land cover
databases for wind resource assessment using mesoscale models. EGU assembly 2013, Vienna.
[33] Barranger N, Stathopoulos C, Kallos G (2013) A power spectrum analysis of turbulent flows in
complex terrain: evaluation of RAMS-LES model against tall mast measurements. Energy
meteorology session of the EMS general assembly, Reading (UK).
[34] Berg J, Bechmann A, Courtney M, Koblitz T, Hristov Y (2012) In the wake of Bolund:
Benakanahalli - Stratification and complex terrain. European Wind Energy Conference,
Copenhagen, Denmark, April 2012.
[35] Borbón F (2011) Lidar (Light Detection and Ranging): Measurement uncertainty in complex
terrain, 6th PhD Seminar on Wind Energy in Europe, Trondheim (Norway), September 2011.
[36] Borbón F, et al. (2011) Investigation of sources for lidar uncertainty in flat and complex terrain.
EWEA-2011, Brussels (Belgium), March 2011.
[37] Borbón F, et al. (2011) Classification of lidar measurement errors in complex terrain. 13th ICWE
proceedings, Amsterdam (The Netherlands), July 2011.
[38] Borbón Guillén F (2011) Classification of Lidar measurement errors in complex terrain conditions.
EAWE 7th PhD Seminar on Wind Energy. Delft, The Netherlands, October 2011.
[39] Borden Z, Koblitz T, Meiburg E (2011) Non-Boussinesq internal bores: Bridging the gap between
the single layer, and boussinesq cases. Bulletin of the American Physical Society 56.
[40] Cañadillas B, Neumann T, Muñoz-Esparza D (2012) First insight in offshore wind profiles up to
250m under free and wind turbine wake flows. DEWI Magazine 41: 18 - 24.
[41] Cañadillas B, Muñoz-Esparza D, Neumann T (2011) Fluxes estimation and the derivation of the
atmospheric stability at the offshore mast FINO1. EWEA Offshore 2011, Amsterdam, The
Netherlands, December 2011.
[42] Cañadillas B, Neumann T, Muñoz-Esparza D (2011) First insight in offshore wind profiles up to
250m under free and wind turbine wake flows. EWEA Offshore 2011, Amsterdam, The
[43] Conan B, van Beeck J, Aubrun S, Devinant P (2011) Physical modelling of wind in complex
terrain: application to wind turbine siting. VKI PhD symposium proceedings, Brussels (Belgium),
March 2011.
[44] Conan B, Buckingham S, van Beeck J, Sanz J, Aunbrun S (2011) Feasibility of micro-siting in
montainous terrain by wind tunnel physical modelling. EWEA-2011 proceedings, Brussels
(Belgium), March 2011.
[45] Conan B, van Beeck J, Aunbrun S, Devinant P (2011) Sand erosion technique applied to wind
resource assessment. 13th ICWE, Amsterdam (The Netherlands), July 2011.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 162/167
Book of Proceedings
[46] Cuzzola F, Aubrun S, Leitl B (2012) Characterization of a wind turbine model. EWEA Special
Topic Conference ‘The Science of Making Torque from Wind’. Oldenburg, Germany, October
2012.
[47] Cuzzola F, Doerenkaemper M, Leitl B, Schatzmann M (2011) Design of an aerodynamically
scaled wind turbine. Physmod2011-International Workshop on Physical Modeling of Flow and
Dispersion Phenomena. Hamburg, Germany, August 2011.
[48] Cuzzola F, Leitl B, Schatzmann M (2010) Physical Modelling of a Wind Turbine. 6th PhD Seminar
on Wind Energy in Europe, Trondheim (Norway), September 2010.
[49] Cuzzola F, Leitl B, Schatzmann M (2010) Wind turbines in ABL-flow. A review on wind tunnel
studies, iTi Conference on Turbulence proceedings. Bertinoro, Italy.
[50] Desmond C, Watson S (2011) Wind flow around a single tree: A study on the sensitivity of CFD
models. 13th ICWE, Amsterdam (The Netherlands), July 2011.
[51] Desmond C, Watson S (2012) Turbulence modelling of flow around a single tree. 7th European
Academy of Wind Energy PhD seminar on Wind Energy in Europe, Delft, 25-26 October 2012.
[52] Desmond C, Sayer A, Watson S, Hancock P (2012) Forest canopy flows in non-neutral stability,
European Wind Energy Association Annual Conference, Copenhagen, 16-19 April 2012.
[53] Desmond C, Watson S, Aubrun S, Ávila S (2013) Improving simulations in forested terrain by
including detailed canopy morphology. European Wind Energy Association Annual Conference,
Vienna, 4-7 February 2013
[54] Desmond C, Watson S (2014) A methodology for the measurement and modelling of stability
effects. EWEA-2014, Barcelona, Spain, March 2014.
[55] Doerenkaemper M, Cuzzola F, Leitl B, Schatzmann M (2011) Measurements on the wake of an
aerodynamically scaled wind turbine. Physmod2011-International Workshop on Physical
Modeling of Flow and Dispersion Phenomena, 2011, Hamburg, Germany, August 2011.
[56] Duraisamy JV, Dupont E, Carissimo B (2013) Downscaling wind energy resource in complex
terrain from mesoscale to microscale model and data assimilating field measurements at few
locations. European Wind Energy Association, Vienna, Austria, February 2013.
[57] Duraisamy JV, Dupont E, Carissimo B (2012) Downscaling wind energy resource from mesoscale
to microscale model and data assimilating field measurements. Journal of Physics: Conference
Series and Proceedings of The Science of Making Torque from Wind 2012, Oldenburg, Germany,
October 9-11.
[58] Duraisamy JV, Dupont E., Carissimo B. (2012) Downscaling the wind energy resource in complex
terrain using coupled mesoscale and microscale CFD modeling. 7th European Academy of Wind
Energy PhD seminar on Wind Energy in Europe, Delft, 25-26 October 2012.
[59] Duraisamy JV, Dupont E, Carissimo B (2012) Towards a new methodology for wind resources
downscaling with CFD in complex terrain. EWEA-2012, Copenhagen, Denmark, April 2012.
[60] Duraisamy JV, Dupont E, Carissimo B (2011) Downscaling the wind energy resource in complex
terrain using coupled mesoscale and microscale CFD modeling. European Wind Energy
Association, Brussels, Belgium, March 2011.
[61] Fitton G (2010) Multifractal Analysis And Simulation Of Wind Energy Fluctuations. 6th PhD
Seminar on Wind Energy, Trondheim.
[62] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2011b) The Anisotropic Multifractal Model and
Wind Turbine Wakes. 7th PhD Seminar on Wind Energy in Europe, 115–118.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 163/167
Book of Proceedings
[63] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2012) Torque Fluctuations In The Framework
Of A Multifractal 23/9-Dimensional Turbulence Model. Presented at the EUROMECH Colloquium
528.
[64] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2013) Multifractal Statistical Methods And
Space-Time Scaling Laws For Turbulent Winds. Euromech 2012.
[65] Fitton G, Tchiguirinskaia I, Schertzer D, Lovejoy S (2013) Scaling Anisotropy And Extremes In
The Wake Of A Turbine. 21eme Congres Francais de Mecanique: Journal of Mechanics and
Industry.
[66] Gancarski P, Sanz Rodrigo J (2013) Windbench. WAUDIT Network meeting, Pamplona, Spain,
January 2013.
[67] Gancarski P, Sanz Rodrigo J (2013) Windbench. 9th EAWE PhD seminar in Wind Energy in
Europe, Visby, Sweden, September 2013.
[68] Holmes H , Sanz Rodrigo J, Cabezón D, Schatzmann M (2012) Model evaluation methodology
for wind resource assessment, in prep.
[69] Holmes H, Schatzmann M, Leitl B (2011) Model evaluation methodology for wind energy
applications. PHYSMOD 2011, Hamburg, Germany 22-24 Aug 2011.
[70] Holmes H, Schatzmann M, Leitl B (2011) Model evaluation methodology for wind energy
applications. École Polytechnique Fédérale de Lausanne Environmental Fluid Mechanics and
Wind Engineering Graduate Student Seminar, 12 September 2011.
[71] Holmes H, Schatzmann M, Leitl B, Sanz Rodrigo J (2011) Quality assurance of wind energy
assessment models. 13th ICWE proceedings, Amsterdam (The Netherlands), July 2011.
[72] Holmes H, Schatzmann M, Leitl B (2011) Quality assurance methods for models used in wind
energy assessment. EGU General Assembly, Vienna (Austria), April 2011.
[73] Keshtova F (2012) Full-scale measurements of aerodynamic induction in a rotor plane Three
dimensional synthetic wind field based on CTRW theory. The Science of Making Torque from
Wind, 9-11 October 2012, Oldenburg, Germany
[74] Keshtova F, Peinke J (2012) Intermittent spatial and temporal structure of wind fields.
EUROMECH Colloquium 528, 22-24 February 2012, Oldenburg, Germany.
[75] Keshtova F, Peinke J (2011) Intermittent spatial and temporal structure of wind fields. 7th EAWE
PhD Seminar on Wind Energy in Europe. Delft University of Technology, 27 - 28th October 2011,
Delft, The Netherlands.
[76] Kieran Fish A, Santos-Muñoz D, Valero F (2012) Comparison of wind speed forecasts from MM5
and WRF ARW ensemble prediction systems over the Iberian Peninsula. 12th WRF Users'
Workshop. Boulder (CO), USA, June 2012.
[77] Koblitz T, Borbón Guillén F, Sanz Rodrigo J (2011). Newsletter WAUDIT Project. CENER,
Sarriguren, Spain, July 2011.
[78] Koblitz T, Bechmann A, Berg J, Sogachev A, Sørensen N, Réthoré P-E (2012) Atmospheric
stability and complex terrain - Comparing measurements and CFD. The science of Making
Torque from Wind, Oldenburg, Germany, October 2012.
[79] Koblitz T, Bechmann A, Sogachev A, Sørensen N (2012) Atmospheric stability in CFD Representation of the diurnal cycle in the atmospheric boundary layer. European Wind Energy
Conference, Copenhagen, Denmark, April 2012.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 164/167
Book of Proceedings
[80] Koblitz T, Bechmann A, Sogachev A, Sørensen N (2012) Atmospheric stability and complex
terrain: Comparing measurements and CFD. 8th PhD Seminar on Wind Energy in Europe, Zurich,
Netherlands, September 2012.
[81] Koblitz T, Bechmann A, Sogachev A, Sørensen N (2011) Modification of CFD code to model the
atmospheric boundary layer. 13th ICWE, Amsterdam, Netherlands, July 2011.
[82] Koblitz T, Bechmann A, Sogachev A, Sørensen N (2011) Modification of CFD code to model the
atmospheric boundary layer. 7th PhD Seminar on Wind Energy in Europe, Delft, Netherlands,
October 2011.
[83] Koblitz TW, Bechmann A, Sogachev A, Sørensen NN (2011) Modification of CFD code to model
the atmospheric boundary layer. 13th ICWE, Amsterdam (The Netherlands), July 2011.
[84] Koblitz T, Bechmann A, Sørensen N (2010) The 2D lid-driven cavity–Validation of CFD code to
model non-Neutral Atmospheric Boundary Layer Conditions. 6th PhD Seminar on Wind Energy in
Europe, Trondheim, Norway, October 2010.
[85] Koblitz TW, Bechmann A., Sørensen N.N., 2010. The 2D lid-driven cavity – Validation of CFD
code to model non-Neutral Atmospheric Boundary Layer Conditions, 6th PhD Seminar on Wind
Energy in Europe proceedings, Trondheim (Norway), 157-160, September 2010.
[86] Muñoz-Esparza D, García-Sánchez C, Cañadillas B, van Beeck J (2012) Multiscale Large Eddy
Simulations of a convective offshore boundary layer: towards a mesoscale-LES coupling. 20th
Symposium on Boundary Layers and Turbulence, Boston, Massachusetts (USA), July 2012.
[87] Muñoz-Esparza D, Cañadillas B, van Beeck J (2012) Mesoscale modelling of Low Level Jets in
the North Sea. 20th Symposium on Boundary Layers and Turbulence, Boston, Massachusetts
(USA), July 2012.
[88] Muñoz-Esparza D, Cañadillas B (2012) Forecasting the diabatic offshore wind profile at FINO1
with the WRF mesoscale model. DEWI Magazine 40: 73 - 79.
[89] Muñoz-Esparza D, Cañadillas B, Neumann T, van Beeck J (2011) WRF mesoscale modelling
and LiDAR measurements of tall wind profiles at FINO1. EWEA Offshore 2011, Amsterdam, The
[90] Muñoz-Esparza D, Cañadillas B, van Beeck J (2011) WRF mesoscale modelling and
measurements of the diabatic wind offshore wind profile at FINO1. 7th PhD Seminar on Wind
Energy in Europe, October 2011 Delft, The Netherlands.
[91] Muñoz-Esparza D, van Beeck J, Cañadillas B (2011) Impact of turbulence modeling on the
performance of WRF model for offshore short-term wind energy applications. 13th International
Conference on Wind Engineering, Amsterdam, The Netherlands, December 2011.
[92] Muñoz-Esparza D, van Beeck J (2011) Forecasting of offshore boundary layer conditions at
FINO1 using high resolution WRF-PBL schemes for wind energy applications. EWEA 2011,
Brussels, Belgium, March 2011.
[93] Muñoz-Esparza D, Conan B, Croonenborghs E, Parente A, van Beeck J, Sanz J (2011)
Sensitivity to inlet conditions of wins resource assessment over complex terrain using three CFD
solvers and wind tunnel data, EWEC-11, Brussels (Belgium), March 2011.
[94] Muñoz-Esparza D, Conan B (2010) Physical and numerical modeling of flow over a real complex
terrain. 6th PhD Seminar on Wind Energy in Europe, Trondheim, Norway, September 2010.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 165/167
Book of Proceedings
(2014) Hierarchy of flow models and experimental support for wind resource assessment. The
WAUDIT-Alaiz complex terrain test case. EWEA-2014, Barcelona, Spain, March 2014.
[96] Sanz Rodrigo J, Desmond C, Landberg L (2014) The WAUDIT Marie Curie Initial Training
Network, a success story. EWEA-2014, Barcelona, Spain, March 2014.
[97] Sanz Rodrigo J (2011) WAUDIT Wind Resource Assessment Audit and Standardization.
Overview of the network. Introduction of the WAUDIT Special Session at the International
Conference on Wind Engineering ICWE13, Amsterdam, Netherlands, 13 July 2013.
[98] Sanz Rodrigo J (2010) WAUDIT ITN: Wind Resource Assessment Audit and Standardization.
Workshop on International University Education in Wind Energy at HUSUM Wind Energy 2012,
Husum, Germany,18 September 2012.
[99] Sanz Rodrigo J (2010) WAUDIT ITN: Wind Resource Assessment Audit and Standardization.
Marie Curie Monitoring Action and Information Days, Barcelona, Spain, 10 June 2010.
[100] Sanz Rodrigo J, van Kuik G, Barth S (2010) WAUDIT Wind Resource Assessment Audit and
Standardization: a Marie Curie Initial Training Network coordinated with the European Academy
of Wind Energy. EWEA-2010, Warsaw, Poland, April 2010.
[101] Santos-Alamillos FJ (2013) Analysis of the spatiotemporal balancing of the wind energy
resources in Andalusia (southern Spain). Center for Wind Energy Research (Forwind). Internal
Seminar, Center for Wind Energy Research (ForWind), University of Oldenburg, Germany, May
2013.
[102] Santos-Alamillos FJ, Von Bremen L, Heinemann D, Pozo-Vázquez D, Ruiz-Arias JA, TovarPescador J, Lara-Fanego V, Linares-Rodríguez A, Martínez-Valenzuela J, Arbizu-Barrena C,
Quesada-Ruiz S, Junk C (2013) Sensitivity analysis of the WRF model to different PBL
parameterization schemes for reproducing wind speed fluctuations associated with convective
phenomena in the North Sea. Poster. 13 EMS Annual Meeting, Reading, UK, September 2013.
[103] Stathopoulos C (2011) Application of Kalman Filter as a Dynamical Property of MOS, Waudit
School, Delft, Netherlands, October 2011.
[104] Stathopoulos C, Sanz J, Badger J, Cantero E, Fernades P, Galanis G, Lozano S (2012)
Evaluation of the numerical wind atlas downscaling methodology in complex terrain, European
Wind Energy Association Conference, Copenhagen, Denmark, April 2012.
[105] Stathopoulos C, Kaperoni A, Galanis G, Kallos G (2012) Dynamical-Statistical Techniques for
wind power prediction. European Geosciences Union conference, Vienna, Austria, April 2012.
[106] Stathopoulos C, Barranger N, Larsén XG, Sanz J (2012) Implementation of spectrum
analysis in mesoscale modeling for wind energy assessment studies, 12th European
Meteorological Society conference, Lodz, Poland, September 2012.
[107] Stathopoulos C, Sanz J (2013) Evaluation of a meso-micro scale application over complex
terrain for wind farm sitting. 9th Phd seminar of European Wind Energy Association, Conference
proceedings, Gotland, Sweden, September 2013.
[108] Ternisien T, Prospathopulos J, Chaviaropoulos PK (2010) New Model Development
Concerning Turbulence and Wakes. 6th PhD Seminar on Wind Energy in Europe,Trondheim,
Norway, October 2010.
[109] Ternisien T, Prospathopulos J, Chaviaropoulos PK (2011) Development of an EARS model for
the simulation of turbulent atmospheric flows. ICWE 13, Amsterdam, Netherlands, July 2011.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 166/167
Book of Proceedings
[110] Ternisien T, Prospathopulos J, Chaviaropoulos, PK (2011) Implementation and calibration for
atmospheric turbulence of an EARS model. 7th PhD Seminar on Wind Energy in Europe
proceedings, Delft University of Technology, Delft, Netherlands, October 2011.
[111] Ternisien T, Prospathopulos J, Chaviaropoulos PK (2012) Anisotropy resolving model for the
RANS simulation of offshore wind turbine wakes. The Science of Making Torque from Wind,
Oldenburg, Germany, October 2012.
[112] Ternisien T (2011) Interaction between boundary layer and wind turbine wake. Course LES in
Hydrodynamics and Offshore Wind Energy, University of Copenhagen, Denmark, August 2011.
[113] Vanadruel M, Sanz Rodrigo J (2011) Grid and Reynolds number sensitivity of k-epsilon
atmosphery boundary layer models in complex terrain: the Bolund test case. 6th PhD Seminar on
Wind Energy in Europe, Trondheim, Norway, October 2010.
[114] Vasiljevic N, Lea G, Courtney M, Mann J, Mikkelsen T (2013) The long-range WindScanner
system – how to synchronously intersect multiple laser beams. EWEA 2013, Vienna.
[115] Vasiljevic N, Lea G (2012) Software implementations in the long-range windscanner system.
8th PhD Seminar in Wind Energy, Zurich.
[116] Vasiljevic N, Courtney M; Wagner R, Mann J, Mikkelsen T (2011) A windscanner simulator.
EWEA 2011, Brussels.
[117] Vasiljevic N (2010) Lidar measurements: Wake inversigation. 6th PhD Seminar in Wind
Energy, Trondheim.
[118] Volker PJH, Badger J, Hahmann AH, Hansen KS (2013) Wind-Farm Parametrisations in
Mesoscale Models. ICOWES, 2013.
[119] Volker PJH, Badger J, Hahmann AH, Ott S (2012) Wake effects of large offshore wind farms
on the mesoscale atmosphere. EWEA-Conference, 2012.
[120] Volker PJH, Hall A, Capps SB, Huang H-Y, Sun F, Badger J, Hahmann AH (2012) Impact of
Wind Farms on the Marine Atmospheric Boundary Layer. AGU, 2012.
[121] Volker PJH, Badger J, Hahmann AH, Hansen KS (2011) Outline for a study of wake effects of
large offshore wind farms: a study of mesoscale atmosphere and ocean feedbacks. Wake
Conference, Visby, Schweden, 2011.
[122] Yeow TS, Cuerva A, Pérez J (2012) Pressure Measurements of the Detachment Bubble on
the Bolund Island. European Wind Energy Association, Copenhagen, April 2012.
[123] Yeow TS, Cuerva A, Conan B, Pérez J (2012) Wind Tunnel Analysis of the Detachment
Bubble on Bolund Island. The Science of Making Torque from Wind, Oldenburg, October 2012.
[124] Yeow TS, Cuerva A, Pérez J, Conan B, Buckingham S, van Beeck J (2011) Modelling of
atmospheric boundary layer: Generation of shear profile in wind tunnel, 6th PhD Seminar on
Wind Energy in Europe proceedings, Trondheim (Norway), 170-179, September 2010.
[125] Yeow TS, Cuerva A, Pérez J (2011) Pressure Measurements of the Detachment Bubble on
the Bolund Island. International Workshop on Physical Modeling of Flow and Dispersion
Phenomena, Hamburg, August 2011.
WAUDIT
Contract#
238576
Tel. (+34) 948 25 28 00
Fax. (+34) 948 27 07 74
Pag. 167/167

WAUDIT Book of Proceedings - Wind resource assessment, audit

Transcription

Similar documents

NIP TUCK: private domain extraction and public

Trail Map - Keystone

dr. blum invitation

Germs Bio-terrain?

electricitate_oferta noua_EN.cdr

Unleash your Holset

Vertical Motion

Brochure - Hummer H1

Adding Logos to Clothing

Chelton EFIS Brochure