NEW APPROACHES TO OFF-SHORE WIND ENERGY MANAGEMENT
EXPLOITING SATELLITE EO DATA
Marco Morelli (1), Andrea Masini (2), Sara Venafra (2) and Marco Alberto Carlo Potenza (1)
(1)
Department of Physics, University of Milano - Via Celoria 16, Milano (Italy) - Email: [email protected]
(2)
Flyby S.r.l. – Via Puini 97, Livorno (Italy)
ABSTRACT
Wind as an energy resource has been increasingly in
focus over the past decades, starting with the global oil
crisis in the 1970s. The possibility of expanding wind
power production to off-shore locations is attractive,
especially in sites where wind levels tend to be higher
and more constant.
Off-shore high-potential sites for wind energy plants are
currently being looked up by means of wind atlases,
which are essentially based on NWP (Numerical
Weather Prediction) archive data and that supply
information with low spatial resolution and very low
accuracy. Moreover, real-time monitoring of active offshore wind plants is being carried out using in-situ
installed anemometers, that are not very reliable
(especially on long time periods) and that should be
periodically substituted when malfunctions or damages
occur.
These activities could be greatly supported exploiting
archived and near real-time satellite imagery, that could
provide accurate, global coverage and high spatial
resolution information about both averaged and near
real-time off-shore windiness. In this work we present
new methodologies aimed to support both planning and
near-real-time monitoring of off-shore wind energy
plants using satellite SAR (Synthetic Aperture Radar)
imagery.
Such methodologies are currently being developed in
the scope of SATENERG, a research project funded by
ASI (Italian Space Agency). SAR wind data are derived
from radar backscattering using empirical geophysical
model functions, thus achieving greater accuracy and
greater resolution with respect to other wind
measurement methods.
In detail, we calculate wind speed from X-band and Cband satellite SAR data, such as Cosmo-SkyMed
(XMOD2) and ERS and ENVISAT (CMOD4)
respectively. Then, using also detailed models of each
part of the wind plant, we are able to calculate the AC
power yield expected behavior, which can be used to
support either the design of potential plants (using
historical series of satellite images) or the monitoring
and performance analysis of active plants (using nearreal-time satellite imagery). We have applied these
methods in several test cases and obtained successful
results in comparison with standard methodologies.
1.
INTRODUCTION
Wind estimation is very important to support the
planning and the monitoring of an off-shore wind farm.
The elaboration of SAR imagery to value sea surface
wind represents an active research field thanks to the
global coverage and the high resolution of the
instrument, not currently achievable by other
oceanographic remote sensing ones, such as altimeters
and scatterometers. It is known that the resolution of a
scatterometer is 25 or 50 Km while SAR wind
estimation is possible as far as 1 Km from the coast. The
accuracy in the wind speed extraction presents an
accuracy of about 10% with an RMS of 1-2 m/s, making
SAR data very attractive to monitor off-shore wind
plants production in near-real time.
Several algorithms have been developed in order to
value wind field on sea surface from SAR products. To
date, there are two methods: the first analyses the
spectral features of SAR data (1, 2) using an empirical
linear relation to value wind speed from imagery cut-off
frequency, while the second methodology is based on a
geophysical model function (GMF) for which the
normalized radar cross section (NRCS) or sigma-naught
is expressed as a function of wind field (intensity and
direction) and radar incidence angle (3). This last
methodology has been originally adopted to SAR data
in C-Band (4, 5, 6) with the development of CMOD4
(7), CMOD5 (8) and CMOD-IFR2 (9). The model
function in C-Band has been extended to X-Band SAR
imagery with the implementation of XMOD1 (10), and
improved by the development of XMOD2 (11).
Off-shore wind farms are usually situated in sites where
windiness is not well known because of the lack and the
difficulty of installing anemometers on the sea. Also the
use of atlas wind does not represent the best
methodology to adopt in order to detect the zones of
major energy production due to their low resolution and
accuracy. For this purpose we have used historical
satellite series of wind data to support the design of an
off-shore wind farm, locating the areas in which the
wind energy production estimation is major.
Moreover the possibility of monitoring real data
production of an off-shore wind farm by means of
satellite measures allows to detect eventual
malfunctions and/or low efficiencies, warranting a
better control of the wind plant to the manager of this.
Section 2 describes the data space and the methods
used. Results are illustrated in section 3, while in
section 4 conclusions and recommendations are
reported.
2.
The major intensity of the backscattering occurs when
 =0°

(Upwind) and when
=180° (Downwind) and
also in corresponding of their multiple integers, as
shown in Figure 1.
METHODS
The development of a GMF to relate wind field to SAR
incidence angle (larger than 20°) is based on the socalled Bragg scattering (12). For this theory the
intensity variation in SAR imagery is proportional to the
amplitude of sea surface waves. In detail, the waves for
which there is a major scattering are those whose
wavelength is
 Bragg   Radar /( 2 sin  )

where Radar is the radar wavelength and  is the
incidence angle between the radar incident beam and the
perpendicular at the incident surface. This means that
the backscattering from short surface waves, responsible
of local wind, is more stressed. The model used for the
estimation of sea wind speed is summarized in Equation
1, where the relation between the normalized radar cross
section
0
Figure 1: Normalized Radar Cross Section versus
Azimuth Angle at 19 GHz (3)
C(1) - C(18) are instead 18 coefficients whose
estimation (7, 10, 11) has allowed us to evaluate the
SAR wind speed that minimizes the following function,
by means of an inversion procedure:
and wind field is described:
 0  B0 1  B1 cos( )  B2 cos2 
F (U )  M U ,     0 
2
(1)
(7)
Where B0, B1 and B2 parameters are functions of wind

speed U and incidence angle

as shown in Equations
2-6, while
is the azimuth angle between the radar
look angle and the wind direction.
B0  10  U 
(2)
  C(1)  C(2)  C(3)
2
  C(4)  C(5)  C(6)
2
(3)

(4)

(5)

B1  C(7)  C(8)  C(9)  C(10)  C(11)  C(12) 2 U
2

B2  C(13)  C(14)  C(15) 2 C(16)  C(17)  C(18) 2 U
(6)
 
Where M U ,  is the radar cross section predicted by
the wind model used.
In X-Band we use SAR data acquired by COSMOSkyMed (Constellation of small Satellites for
Mediterranean Basin Observation) satellite constellation
(13), funded by Italian Space Agency (ASI) and
conceived as a dual-use system, civilian and defence,
for a wide range of applications, such as risk
management, scientific and commercial applications,
etc. The SAR in X-Band has the advantage to operate
independently of weather and day night, providing the
possibility to observe the area of interesting
continuously. In particular we elaborate SAR products
acquired in DGM (Detected Ground Multi-Look)
Stripmap Himage mode (see Figure 2) Level 1B with a
spatial coverage of about 40 x 40 Km, in VV
polarization, with an incidence angle between 18° and
45°, and a ground resolution of about 3 m.
Figure 2: COSMO-SkyMed acquisition modes
These data need a calibration step (14, 15), before their
processing, in order to obtain quantitative information
independent from the used sensor and the embraced
assembly line. In fact it is necessary to compensate the
range spreading loss, the antenna pattern gain and the
incidence angle variations of the received signal. By
means of Equation (8) the SAR COSMO-SkyMed
calibration has been carried out.
 0 (i, j ) 
Where
2 Re xp
img (i, j ) 2 Rref
sin( ref )
F 2K
 0 (i, j )
(8)
represents the NRCS in corresponding
to image position (i, j ) on the Range and Azimuth
respectively, img (i, j ) is the value of SAR pixel in
Digital Number (DN),
slant range,
 Re f
R exp
R Ref is the calibration reference
is the reference slant range exponent,
is the calibration reference angle, F is the scaling
factor and K is the calibration constant.
In our work we utilize the ESA software NEST
(http://envisat.esa.int/nest/) which is able to return the
SAR data just calibrated.
Then the sigma-naught has been averaged on 400 x 400
pixel sub-images which correspond to a coverage of 1 x
1 Km since the pixel spacing of the COSMO-SkyMed
Strip Image data is 2.5 m/pixel. From this mean value
we estimate wind speed through Equation (7) as
mentioned above (See Figure 3).
Figure 3: Flow-Chart used for the sea surface wind
speed estimation from SAR data
The RMS difference between XMOD2 and QuikSCAT
wind calculated on 53 SAR images is 0.8 m/s for low
winds (lower than 7 m/s), while the RMS between
XMOD2 and GFS (Global Forecast System) winds on a
data set of 121 SAR products is 2 m/s for winds
comprised between 7 and 25 m/s (11). Moreover a
comparison between XMOD2 and buoy data has been
carried out, providing an RMS of 1.3 m/s (11).
We stress that the wind direction in input at our
processor is provided externally by other satellite data.
This wind field is important not only to yield the input
wind direction but also to allow a checking/validation
step for the wind speed estimated by the model. It is
well-known that SAR gives information only about
wind speed known the direction, with respect to
scatterometer which provides both wind speed and
direction. This last can be deduced from SAR imagery
examining the features leaved by the wind on sea
surface. In detail Gerling (16) realized that wind
direction was estimated by means of Fourier transform
techniques detecting the features with a major scatter.
This deduction is more evident near the coasts of little
islands, supposing that there is an ambiguity of 180° due
to the impossibility to distinguish upwind from
downwind. Even if this approach is significant to this
day, sometimes wind features are not visible in SAR
images or there are features associated to other physical
phenomena at the same spatial resolution, such as ocean
drift systems.
For the case study we have used wind field provided by
NOAA
(National
Oceanic
and
Atmospheric
Administration) NCDC (National Climatic Data Center)
free available at the following web link:
http://www.ncdc.noaa.gov/thredds/dodsC/oceanwinds6h
r.html. These marine surface data are observed by
research and VOS ships, moored and drifting buoys, and
multiple satellites, then they are archived at NCDC as
national records. The individual satellite data were
downloaded from the website of the Remote Sensing
System (RSS) Inc. (http://www.remss.com/). The data
are then blended together with advanced methods to
increase resolution and reduce both bias and
sampling/analysis errors. The blending also serves as a
tool for “data reduction and” and “information
deduction”. The global gridded products are then served
to the communities via web based interactive servers,
with features of visualization, data manipulation, data
sub-setting and downloading in user preferred formats.
These wind data are 6-hourly and 0.25° global sea
winds, available from July 1987 till today. In Figure 4
there is an example of blended and gridded global wind
fields from the multiple satellites at January 15, 2012,
00:00 UTC, while in Figure 5 the climatological mean
annual cycles (January 1995-December 2004) averaged
over several latitude bands are illustrated.
several latitude bands
These sea winds are interpolated both on spatial and
temporal scale in order to obtain wind fields comparable
with the scale used by the RCS model. In this way we
are able to estimate wind speed from SAR data in nearreal time, fundamental for the monitoring of off-shore
wind farms.
About the support to the design and the planning of an
off-shore wind plant, we have used the sea global winds
above-mentioned. In particular we have built up a
database of wind fields from January 1, 2000 till
December 31, 2011 for a total of twelve years with the
aim to acquire information regarding the winds trend
which affects the energy production efficiency on
several areas with a global coverage. Historical series of
wind satellite data and their analysis allow to detect the
better zones from a wind energy production viewpoint
with a high level of accuracy.
There are different types of wind turbines, but generally
for planning an off-shore wind farm, wind turbines with
horizontal axis are most used. The energy production of
a wind plant is usually carried out with the analysis of
the power curve. It is convenient to define a model for
the electrical power output Pe in order to use it in
discussing any wind system. We assume that Pe varies
as uK between cut-in and rated wind speeds uc and uR
respectively, where k is the Weibull shape parameter.
The model wind turbine used for this project is the
following (17):
Figure 4: Example of NASA RSS satellite global sea
winds at January 15, 2012, 00:00 UTC
(9)
Where u is the wind speed at the hub height, PeR
represents the rated electrical power and uF is the
furling wind speed, that is the wind speed at which the
turbine is shut down to prevent structural damage. The
wind speed at hub height z is obtained by u = uref
ln(z/z0 ) / ln(zref /z0 ) where uref is the wind speed at
reference height zref (10 m above sea level in our case)
and z0 is the roughness length equal to 0.0002 for sea
surface.
The parameters a and b are given by Equations (10) and
(11):
Figure 5: Example of climatological mean annual
cycles (January 1995-December 2004) averaged over
(10)
(11)
A typical plot of electrical power output versus wind
speed is shown in Figure 6, for k=2.
Figure 7: Weibull density function f(u) for scale
parameter c=1
Figure 6: Model wind turbine output versus wind speed
At this point it is very important to know the average
power output Pe,ave of a wind energy system for the
purpose of determining the total energy production and
the total income. Pe,ave is the power produced at each
wind speed times the fraction of the time that wind
speed is experienced, integrated over all possible wind
speeds. The integral form of Pe,ave is:
(12)
In which f(u) is a probability density function of wind
speeds. The Weibull distribution of f(u) is described by
the following equation:
(13)
Where c and k are the Weibull parameters which
indicate respectively the scale parameter, indicative of
windiness of the examined site and the shape parameter
indicative of the output power curve versus wind speed,
as shown in Figure 7.
Substituting the Equations (9) and (13) in (12) and
reducing to the minimum number of terms, the result is:
(14)
With this expression we are able to estimate the average
power production of a turbine, known Weibull
parameters, whereas uR, uc and uF are provided by the
turbine manufacturer. The term inside the braces of
Equation 14 is called capacity factor CF, which
represents an important design item in addition to the
average power. Typical CF are included between 20%
and 40%.
There are several methods for the purpose to estimate
the Weibull parameters. One of these methodologies
starts with the integration of Equation (13) to obtain the
distribution function F(u) which can be linearized by
taking the logarithm. The result is described by the
following equation:
(15)
which is in the form of a straight line y=ax+b. Thus the
Weibull parameters are calculated. Generally the
Weibull prediction is within 20% of the actual value for
either wind speed range that is not bad for a data set
which is difficult to describe mathematically.
From historical series of wind data we are also able to
estimate the most frequent wind direction in
correspondence of which there is also a major output
wind power. In fact full power output would normally
be obtained when the wind direction is perpendicular to
the plane of rotation.
3.
RESULTS
Yearly energy production and its variation with annual
wind statistics must be well known to support the
planning of an off-shore wind farm. The historical
satellite wind data of RSS are used for this purpose,
while the estimation of wind speed from SAR data
allows us to monitor wind plants in near-real time,
analyzing their performance and detecting eventual
malfunctions and/or low efficiencies.
In this section we present the results of these
methodologies and a case study of application near the
coasts of Apulia. In the following figure we illustrate
the comparisons between wind speed retrievals
calculated by XMOD2 (blue arrows) and those provided
by NCDC NOAA (red arrows). In Figure 8 also the
wind speed differences between the two retrievals are
shown in red.
Figure 9: Histogram of wind speed differences between
XMOD2 and NCDC NOAA for a COSMO-SkyMed
imagery off the coasts of Apulia
This comparison is also agreed with the accuracy of
XMOD2 model which is within 10%.
In order to support the design of an off-shore wind farm
we have used the wind field database of NCDC NOAA
for twelve years from 2000 till 2011, as mentioned in
the previous section. In Figure 10 we show the wind
speed distribution for the all years off the Apulia coasts
with a longitude of 18° and a latitude of 42°. Using
Equation (15) we have calculated the Weibull
parameters k and c useful to retrieve the output energy
production power. The estimated Weibull function is
also plotted in Figure 10 with a red line.
Figure 8: XMOD2 wind speeds (blue) compared with
NCDC NOAA ones (red) and their differences
XMOD2 wind speed has been estimated on 1 x 1 Km
sub-images areas, using a COSMO-SkyMed Stripmap
Himage data acquired on June 2, 2009 at 04:00 UTC.
NCDC NOAA satellite data are instead interpolated
both spatially and temporally to obtain the same scale of
retrieval. The agree is fairly good with a mean
difference of 1.4 m/s and a standard deviation of about
0.6 m/s and with the 82% of measures with differences
until 2 m/s, as shown in Figure 9.
Figure 10: Wind speed distribution at longitude 18° and
latitude 42° from 2000 till 2011 and estimated Weibull
function over plotted
The Weibull fit has provided a square correlation
coefficient of about 0.987, which is very good.
Exploiting a Siemens SWT-6.0-154 offshore turbine
with a PeR of 6 MW, uC of 3 m/s, uR of 12 m/s and uF
of 25 m/s and with hub height of 90 m (See Figure 11) ,
the plot of the average power output (See Equation (14))
for all twelve years versus wind speed is illustrated in
Figure 12.
Figure 11: Siemens SWT-6.0-154 offshore turbine
energy production versus wins speed
Figure 14: Monthly mean wind speeds for all the years
and for the site at issue
Figure 12: Average power output versus wind speed
This scheme can be applied for each year: Figure 13
shows the retrieved energy production at the output of
the off-shore wind turbine for the all analysed years.
Every year there is an energy production of about 15
GWh from a single turbine.
The Capacity Factor CF for the case study is about 30%
and the number of equivalent hours is meanly 2635
h/year.
Analysing the wind direction distribution is also
fundamental to orient the plane of rotation of the turbine
perpendicularly at the most prevalent wind direction for
which there is also a major energy production. For the
case of study we have retrieved that the maximum
energy production is around 10 m/s, as shown in Figure
12. Overlapping the wind rose corresponding to the
wind direction distribution for all the years (blue) and
that relative to the wind direction distribution with the
maximum energy production (red) (See Figure 15), we
highlight that the turbine plane of rotation has to be
oriented at about 130° respect to the North, since the
most prevalent energy production is for wind direction
of about 40°.
Figure 13: Pe,ave from January 1, 2000 till December
31, 2011 for the examined site
In Figure 14 we illustrate the monthly mean wind
speeds for this site.
Figure 15: Wind direction distribution from 2000 till
2011 (blue) and for the maximum energy production
(red) off the Apulia coasts
Keeping in mind that the accuracy in wind speed
estimation is within 10%, the accuracy in average
output wind energy production will be within 20%, a
very good result.
4.
ACKNOWLEDGEMENTS
The authors are grateful to Dr. Ettore Lopinto, ASI
Project Manager of SATENERG, for his collaboration
and his constant support to our work and to the project.
6.
REFERENCES
1.
KERBAOL, V., CHAPRON, B., ELFOUHAILY,
T. AND GARELLO, R. 1996. Fetch and Wind
Dependence of SAR Azimuth Cutoff and Higher
Order Statistics in a Mistral Wind Case. In
International Geoscience and Remote Sensing
Symposium IGARSS’96, 21-26 May 1996, Lincoln,
Nebraska, USA.
2.
ZECCHETTO, S., NIRCHIO, F., DI TOMASO, S.
AND DE BIASIO, F. 2008. Similarities and
differences of SAR derived wind fields using two
different methods: the local gradient and the
continuous wavelet transform methods. In
Proceedings of 2nd International Workshop on
Advances in SAR Oceanography from ENVISAT
and ERS missions, 21-25 January 2008, Frascati,
Italy.
6.
ZECCHETTO, S., AND DE BIASIO, F. 2008. A
Wavelet Based Technique for Sea Wind Extraction
from SAR Images IEEE Trans. Of Geoscience and
Remote Sensing, 46, 10, 2983-2989.
7.
STOFFELEN, A. AND ANDERSON, D. 1997.
Scatterometer Data Interpretation: Measurement
Space and inversion. Journal of Atmospheric and
Oceanic Technology, 14(6), pp. 1298-1313.
8.
HERSBACH, H. 2002. An Improved Geophysical
Model Function for ERS C-Band Scatterometry.
Available
online
at
http://earth.esa.int/pub/SCATTEROMETER/ecmwf
_rep/cmod5.pdf
9.
Y. Quilfen, B. Chapron, T. Elfouhaily, K. Katsaros
and J. Tournadre, 1988. Observation of tropical
cyclones by high-resolution on scatterometry.
Journal of Geophysical Research, v. 103, p. 77677786
CONCLUSIONS
We have reviewed wind speed retrieval methodologies
from SAR imagery with the objective to support the
monitoring the performances of an offshore wind farm
in near-real time. Differently from other techniques such
as the use of altimeters, scatterometers, wind atlas,
anemometers, SAR winds fields are more exact in terms
of accuracy and spatial resolution. They also provide
measures independent from the instrument degradation
due to environmental factors. The results from SAR
data are in good agreement with other satellite wind
products and with ground truth data. We have also used
historical series of satellite wind data in order to support
the planning of an off-shore wind plant and to detect the
areas with a major energy production. We expect to
improve our models in terms of accuracy when wind
energy provided by a real offshore wind farm will be
available, with the aim to compare the output estimated
energy production with the real performance analysis,
detecting eventual malfunctions and/or damages and
alerting the manufacturer.
5.
5.
KERBAOL, V., CHAPRON, B. AND VACHON,
P.W. 1998. Analysis of ERS-1/2 SAR Wave Mode
Imagettes, Journal of Geophysical. Research.,
103(C4), pp. 7833-7846.
3.
ELACHI, C. (Ed.) 1987. Spaceborne radar remote
sensing: applications and techniques, pp. 202-204,
(New York: IEEE Press).
4.
VACHON, P.W. and DOBSON, F.W. 1996.
Validation of wind vector retrieval from ERS-1
SAR images over the ocean. The Global
Atmosphere and Ocean System, 5, pp. 177-187.
10. NIRCHIO, F. and VENAFRA, S. 2010.
Preliminary model for wind estimation from
Cosmo/SkyMed X band SAR data. In Proceedings
of International Geoscience and Remote Sensing
Symposium IGARSS 2010, 25 – 30 July 2010,
Honolulu, Hawaii, USA pp. 3462-3465.
11. NIRCHIO, F. AND VENAFRA, S. 2013. XMOD2
- An improved geophysical model function to
retrieve sea surface wind fields from CosmoSkyMed X-band data. European Journal of Remote
Sensing (submitted and accepted).
12. W. J. Plant, 1990. Bragg scattering of
electromagnetic waves from the air/sea interface.
Surface waves and fluxes. Geerneart and Plant,
Eds., Vol. 2, Kluwer Academic, 41-108.
13. “COSMO-SkyMed SAR Products Handbook” Doc.
No:
ASI-CSM-ENG-RS-092-A.
http://www.cosmo-skymed.it/en/products.htm
14. H. Laur, P. Bally, P. Meadows, J. Sanchez, B.
Schaettler, E. Lopinto and D. Esteban, 2004.
Derivation of the backscattering coefficient in ESA
ERS SAR PRI Products. ESA Technical note ESTN-RS-PM-HL09, Issue 2, Rev. 5f.
15. INVERSI, P. 2009. Cosmo/SkyMed image
calibration. Available online at: www.e-
geos.it/products/pdf/COSMO-SkyMedImage_Calibration.pdf
16. GERLING, T.W. (1986). Structure of the surface
wind field from the Seasat SAR, J. Geophys. Res.,
91(C2), 2308–2320
17. Powell, W. R.: “An Analytical Expression for the
Average Output Power of a Wind Machine,” Solar
Energy, Vol. 26, No. 1, 1981, pp. 77-80.