Renewable Energy 102 (2017) 433e444

Contents lists available at ScienceDirect

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

Offshore winds and wind energy production estimates derived from

ASCAT, OSCAT, numerical weather prediction models and buoys e A
comparative study for the Iberian Peninsula Atlantic coast
D. Carvalho a, *, 1, A. Rocha a, M. Go

mez-Gesteira b, C. Silva Santos c
a
rio de Santiago, 3810-193, Aveiro, Portugal
CESAM - Department of Physics, University of Aveiro, Campus Universita
b
EPHYSLAB - Environmental Physics Laboratory, Facultad de Ciencias, Universidad de Vigo, 32004, Ourense, Spain
c
MEGAJOULE Inovaça~o, Lda., TECMAIA, Rua Eng. Frederico Ulrich, 2650, Moreira da Maia, Portugal

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

Article history: Different offshore wind datasets were compared with buoys measurements in the Iberian Peninsula
Received 23 May 2016 Atlantic coast, aiming to assess the best alternatives to ocean surface measured winds for offshore wind
Received in revised form energy assessment and other applications.
23 September 2016
Results: show that although ASCAT (Advanced SCATterometer) high-resolution product showed the
Accepted 30 October 2016
Available online 3 November 2016
lowest wind speed temporal variability and wind power flux estimations errors, a WRF (Weather
Research and Forecast) model high-resolution simulation can be considered as the best alternative for
offshore wind energy applications. WRF's simulation showed wind power flux estimations errors very
Keywords:
Offshore wind energy
similar to ASCAT, the best representation of the wind speed mean state and Weibull probability density
Ocean wind functions, and provide offshore wind data at typical turbines hub heights. CCMP (Cross-Calibrated Multi-
Scatterometers Platform Ocean Surface Wind Vectors) can also be seen as a valid and readily available alternative of
Iberian peninsula offshore wind data.
Reanalyses The findings presented here can be of great value for offshore wind energy applications that focus on
WRF ocean areas where measured wind data are not available, or are insufficient for the desired purposes. An
informed choice of the most accurate offshore wind databases will provide more realistic wind energy
production estimates that will positively impact the preliminary planning stages of offshore wind energy
exploration projects.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction energy farms is expected [8]. The increasing interest in wind energy
derived from offshore (ocean) sites is mainly due to their higher
Offshore wind energy is presently one of the renewable energy energy potential production when compared to onshore (inland)
sources with highest growth potential and, considering the ever- wind farms, as a consequence of the steadier and higher wind
growing need to reduce fossil fuels dependency and greenhouse speeds present in ocean areas. Although the installation and
gases emissions, it is expected that offshore wind energy will be a exploration of offshore wind energy farms is presently more
cornerstone in the future “green energy” market [13,20,23]. Thus, expensive and technologically challenging than onshore ones, it is
in the near future an increase in the proliferation of offshore wind becoming increasingly difficult to find in Europe new attractive and
suitable areas for the implementation of onshore wind farms [30].
One of the main factors that still hampers the installation of new
offshore wind farms is the lack of offshore wind measured data, as a
* Corresponding author. result of the technical challenges and high investment costs asso-
E-mail addresses: david.carvalho@ua.pt, david.carvalho@nasa.gov (D. Carvalho),

mez-Gesteira), carlos.
ciated with wind measuring campaigns over ocean areas. Although
alfredo.rocha@ua.pt (A. Rocha), mggesteira@uvigo.es (M. Go
santos@megajoule.pt (C. Silva Santos). in-situ measured offshore wind data are available (moored and
1
Present address: Goddard Earth Sciences Technology and Research (GESTAR), drifting buoys, ships, vessels, etc.), this data shows a highly variable
Universities Space Research Association, Columbia, MD, USA; Global Modeling and availability both in space and time, making it not representative of
Assimilation Office (GMAO), NASA Goddard Space Flight Centre, Greenbelt, MD, the local wind regimes [1]. For offshore wind resource assessment,
USA.

http://dx.doi.org/10.1016/j.renene.2016.10.063
0960-1481/© 2016 Elsevier Ltd. All rights reserved.
434 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

highly accurate offshore wind is paramount: given that the ener- sources of offshore winds. Albeit the work of Carvalho et al. [9]
getic production is proportional to the wind speed cubed, appar- performed a comprehensive comparison of a wide range of alter-
ently small errors in the wind speed data will originate prohibitive native sources of offshore wind data with one year of in-situ ocean
deviations in the expected offshore wind energy production [12,17]. surface wind measurements (scatterometers, reanalyses/analyses,
At the same time, a realistic assessment of wave energy production blended datasets and NWP modelled winds), until the present
heavily relies on an accurate knowledge of the local offshore wind moment it is lacking a joint comparison that also includes the
fields since these are the main forcing agents of wave formation newer scatterometers (ASCAT and OSCAT), performing an inte-
and behaviour [4,32]. Furthermore, the knowledge and availability grated assessment and objectively determining which one of these
of offshore wind data are paramount for a wide range of academic datasets is the best alternative to in-situ offshore wind measure-
and business applications such as the study of ocean circulation ments, both in terms of wind statistics and offshore wind energy
(where offshore winds are one of the main forcing), down/up- estimates. Nevertheless, several studies can be found that compare
welling episodes, ocean surface mixed-layer variability, swell sys- ASCAT and/or OSCAT wind data with buoy measurements: Rani
tems, simulation and forecasts of ocean waves which require et al. [28] compared OSCAT, ASCAT and NWP wind forecasts with
accurate offshore surface wind data [6], etc. buoy measurements for the 2011 monsoon season (June to
Presently, several alternatives to in-situ offshore wind mea- September), concluding that ASCAT winds show higher accuracy
surements are available, ranging from remotely sensed (retrieved than OSCAT winds. Kumar et al. [21] compared 9 months of OSCAT,
by instruments on board satellites that orbit the Earth) to winds ASCAT and QuikSCAT winds with buoy measurements and pre-
simulated by numerical weather prediction (NWP) models and sented results that suggest that ASCAT winds are the ones closest to
products that combine in-situ, remotely sensed and NWP simu- the measurements and OSCAT winds show the highest errors,
lated data (reanalyses/analyses and blended products). Each one of higher than QuikSCAT winds. Edson [15] confirmed these findings,
these different types of offshore wind products has their strengths concluding that OSCAT has not proven to be as reliable as either
and weaknesses. Satellite wind data are available at a near-global QuikSCAT or ASCAT.
scale and in near real-time, but its wind data are derived from The findings presented in this study constitute a solid,
the measurement of other parameters (ocean surface backscatter, comprehensive and integrated comparison of practically all sources
cloud and water vapour features tracking, etc.). Moreover, this type of offshore winds, which can be of great value for offshore wind
of instruments usually shows low spatial and/or temporal resolu- energy assessment studies in areas where no offshore wind mea-
tions and large data gaps (rain contamination, instruments mal- surements are available (or the available data is insufficient and/or
function, etc.) and usually suffer from strong limitations in inadequate for the desired purposes). Furthermore, this study is
retrieving the wind vectors near the coast, prime candidate areas to focused on the Iberian Peninsula, an area with an attractive wind
install offshore wind farms. NWP models can provide wind data energetic potential proven by the fact that it is presently one of the
with high spatial and temporal resolutions for any area of the globe areas with the highest percentage of installed onshore wind power
in a relative fast manner. However, NWP modelled winds show (per capita). Combined with its large coastal line, this area has a
deviations when compared to wind measurements mainly due to high potential for future offshore wind energy exploration.
their inability in accurately representing the local terrain charac-
teristics (topography and roughness) and in resolving medium/ 2. Data and methodology
small scale meteorological phenomena that can significantly
impact the accuracy of their wind simulations and respective wind 2.1. Scatterometer offshore winds
energy production estimates. Reanalyses and analyses combine
NWP modelled data with atmospheric, oceanic and other mea- ASCAT-A was launched in late 2006 on board the first version of
surements, but they usually have coarse spatial resolutions the Meteorological Operational satellite (MetOp-A), through a joint
(50e250 km) that do not allow an accurate characterization of the collaboration between the European Space Agency (ESA) and the
local wind regimes and a realistic assessment of the local offshore European Organization for the Exploitation of Meteorological Sat-
wind energy production potential. Offshore wind blended products ellites (EUMETSAT). Details about ASCAT-derived products can be
usually consider as background a first-guess (analysis) of the wind found at http://www.knmi.nl/scatterometer/publications. The
vector and assimilate multiple sources of observed offshore wind MetOp-A satellite is currently in operation in a sun-synchronous
data (satellites, ships, buoys, etc.). circular orbit. ASCAT-A provides two spatial resolutions, 25 km
This study aims to compare wind data and offshore wind energy and 12.5 km over the global ocean, and also a 12.5 km resolution
production estimates derived from different sources of offshore product optimized for coastal areas. This coastal dataset differs
winds, namely the newer generation scatterometers Advanced from the standard 12.5 and 25 km datasets in the use of a spatial
SCATterometer (ASCAT) on board the satellites MetOP-A (ASCAT-A) box filter (rather than the Hamming filter) to generate a spatial
and MetOP-B (ASCAT-B) of the European Space Agency, and the average of the Sigma-0 retrievals from the Level 1B dataset. This
Indian Space Research Organization's Oceansat-2 SCATterometer enhanced coastal retrieval allows for the computation of winds as
(OSCAT) with buoy-measured offshore winds to assess which close to 15 km from the coast, while the standard 12.5 km product
alternative offshore wind databases are able to better represent the shows a 35 km land mask.
local wind regimes and, thus, can be considered as the best alter- ASCAT-B is the follow-on ASCAT-A, being successfully launched
native to in-situ near surface ocean measured winds particularly for into polar orbit in September 2012 on board the MetOp-B satellite.
offshore wind energy assessment applications. One year of MetOp-B is in a co-planar orbit and nearly half an orbit out of phase
satellite-derived, NWP modelled and reanalysis/analysis winds are with MetOp-A. ASCAT-A and ASCAT-B consist of the same tech-
compared with offshore wind measurements collected by eight nology, but ASCAT-B only has two products available: the standard
buoys moored offshore the Iberian Peninsula Atlantic coast. Several 25 km resolution and the 12.5 km resolution product optimized for
studies can be found in the published literature that compare coastal areas. All ASCAT products used in this study (ASCAT-A
offshore wind energy production estimates using one (or more) of 25 km, 12.5 km and coastal products; ASCAT-B 25 km and coastal
the aforementioned alternative sources of offshore wind data (eg., products) are from EUMETSAT Ocean and Sea Ice Satellite Appli-
[3,19,29,31]. However, no study was found that compared wind cation Facility (OSI SAF), provided through the Royal Netherlands
data or offshore wind energy production estimates from all these Meteorological Institute (KNMI) and downloaded from NASA's Jet
 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444 435

Propulsion Laboratory Physical Oceanography Distributed Active 2.4. High-resolution mesoscale offshore winds simulation
Archive Centre (JPL-PODAAC) site (https://podaac.jpl.nasa.gov).
OSCAT was launched on board the OceanSat-2 satellite by the One-year of near surface offshore winds were simulated for the
Indian Space Research Organization (ISRO) in September 2009. The buoy sites using the mesoscale model Weather Research and
swath retrievals are provided at 12.5 km pixel resolution, following Forecasts (WRF). The design of the simulation followed the opti-
a sun-synchronous near circular orbit and the local time on mization studies performed by Carvalho et al. [6,10e12], for the
ascending pass is at 12:00 AM (UTC - Coordinated Universal Time). area under study. The spatial resolution of this simulation is of
The OSCAT L2B product used in this study was produced by JPL in 5 km. Table 1 summarizes the main characteristics of all alternative
cooperation with ISRO and was downloaded from NASA's JPL- offshore wind databases used in this study.
PODAAC website. Further information about OSCAT can be found
at http://www.knmi.nl/scatterometer/publications. All satellite- 2.5. Offshore measured wind data
derived offshore wind data were subjected to a thorough quality
control, where all records flagged with poor quality (such as data The observed offshore wind dataset used in this study was
measured in the presence of rain) or not monitored were collected by eight buoys moored off the Portuguese west coast,
disregarded. Galician northern and western coasts and the Gulf of C
adiz. The
Portuguese buoys are operated by the Portuguese Hydrographical
2.2. Blended offshore wind products Institute, an agency of the Portuguese Navy, while the Spanish
buoys are operated and maintained by the Puertos del Estado
The National Climatic Data Centre Blended Sea Winds (NCDC- Spanish Agency. Summarized information regarding these buoys is
BSW) are a gridded dataset of offshore near surface wind vectors, presented in Table 2, and Fig. 1 depicts their locations.
covering the whole globe with a spatial resolution of 0.25
lat/lon The buoys wind measurement instruments are placed at 3e4 m
and providing 6 records per day. The wind speed data consist in a a.s.l., while satellite-derived winds consist in equivalent neutral
blend of multiple satellite sources: the SSM/I F08, F10, F11, F13, F14, winds at 10 m a.s.l. and all the alternative offshore wind data
F15 and F17 missions, QuikSCAT, the Tropical Rainfall Measuring sources under study refer their wind data to (stability-dependent)
Mission Microwave Imager (TRMM-TMI), and the Advanced Mi- 10 m a.s.l. Given that this work aims to assess which of the alter-
crowave Scanning Radiometer Earth Observing System (AMSR-E). native sources of offshore wind data are closer to real winds (sta-
Wind direction information is taken and interpolated from NCEP- bility-dependent), data from buoys and NWP models were not
R2 and ECMWF operational analysis to the blended wind speed converted to equivalent neutrally stable winds. However, it is
grid. This gap-free offshore winds database is available from 1987 necessary to extrapolate the buoy data to 10 m a.s.l. Such extrap-
to the present time in near real-time. More details about this olation should be done with methods that account for the atmo-
product are available in Zhang et al. [34]. spheric stability, like the Monin-Obukhov theory [25] or the
The Cross-Calibrated Multi-Platform Ocean Surface Wind Vec- method proposed in Liu and Tang [22], which require measure-
tors (CCMP) is also a gap-free global database of near surface ments regarding temperature, heat fluxes and friction velocity.
offshore winds, available with a horizontal resolution of 0.25
lat/ Since such data are not available for the buoys considered here, the
lon. Being a blended dataset, CCMP considers as first-guess field the logarithmic wind profile was used to extrapolate the measured
European Centre for Medium-Range Weather Forecasts (ECMWF) winds. The logarithmic method only requires wind speed and di-
ERA-40 reanalysis and the ECMWF operational analysis, into which rection measurements, but assumes a neutral atmosphere. As
several satellite-derived offshore wind observations and conven- proven by several authors [5,14,18,24], differences between
tional in-situ observations of offshore winds (ships, buoys, etc.) are stability-dependent and neutrally stable winds are low over the
assimilated. A description of this product is available in Atlas et al. global ocean, rarely exceeding 0.5 m s1. Besides this, the extrap-
[2]. olation is done for a small DH (from 3 to 4e10 m a.s.l.) over a surface
with low roughness, suggesting that eventual differences between
2.3. Reanalysis and analysis products measured and extrapolated winds are small. Moreover, by using a
complete year of wind data, it is reasonable to assume that the
The reanalyses/analyses considered in this study are: the average atmospheric stratification is close to neutral.
ECMWF Interim reanalysis (ERA-Interim); the National Centre for Data from the aforementioned alternative sources of offshore
Environmental Prediction Climate Forecast System Reanalysis wind data were compared with these measured offshore winds
(NCEP-CFSR); NASA's Modern Era Retrospective Analysis for considering a spatial (nearest neighbour at a maximum distance of
Research and Applications (NASA-MERRA); the National Centre for each database resolution) and temporal (within 1-h window)
Environmental Prediction Global Forecast System (NCEP-GFS) and collocation. The complete years (January to December) of 2011 and
the National Centre for Environmental Prediction Final Analysis 2013 were used.
(NCEP-FNL). The main differences between these reanalysis and
analysis products are mainly related to their spatial resolution, 2.6. Statistical evaluation of the wind data
amount of assimilated observations, design and quality of their data
assimilation system. While older reanalyses such as NCEP-R2 have To compare the different sources of offshore wind data with the
lower spatial resolutions (2.5
in latitude and longitude for NCEP- measurements, the Root Mean Squared Error (RMSE), the bias,
R2) and assimilate limited amount of meteorological measure- Standard Deviation of the Error (STDE) and correlation coefficients
ments (especially satellite-derived data), newer reanalyses offer (R2) for the wind speed and direction were computed considering
higher spatial resolutions (typically between 0.75 and 0.5
lat/lon) the simultaneous and valid wind speed and direction data records
and include much more assimilated data. When compared to among all databases. ASCAT and OSCAT products offer two records/
reanalyses, analyses products have the advantage of a very fast day (approx. 10 a.m./p.m. for all ASCAT products and 12 a.m./p.m.
(almost in real-time) availability, but the disadvantage of assimi- for OSCAT, UTC time), while the buoys, WRF simulation, NASA-
lating limited amount of meteorological measurements (basically MERRA and NCEP-CFSR have hourly records and CCMP, NCDC-
only the operational observations included in the Global Tele- BSW, ERA-Interim, NCEP-FNL and NCEP-GFS offer four records per
communication System). day (00, 06, 12 and 18 h UTC).
436 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

Table 1
Main characteristics of the considered offshore wind datasets.

Dataset Type of dataset Spatial resolution Temporal Processing level Included data sources Time coverage
resolution

WRF NWP 5 km Hourly e Forcing data from reanalyses e

ASCAT-A 25 km Satellite (Swath) 25 km 10 a.m./10 p.m. L2B ASCAT-MetOP A 2007ePresent
ASCAT-A 12.5 km Satellite (Swath) 12.5 km 10 a.m./10 p.m. L2B ASCAT-MetOP A 2009ePresent
ASCAT-A Coastal Satellite (Swath) 12.5 km 10 a.m./10 p.m. L2B ASCAT-MetOP A 2010ePresent
ASCAT-B 25 km Satellite (Swath) 25 km 10 a.m./10 p.m. L2B ASCAT-MetOP B 2012ePresent
ASCAT-B Coastal Satellite (Swath) 12.5 km 10 a.m./10 p.m. L2B ASCAT-MetOP B 2012ePresent
OSCAT Satellite (Swath) 12.5 km 12 a.m./12 p.m. L2B OSCAT 2010e2014
NCDC-BSW Blended 0.25
lat/lon 4 times/day L3 QuikSCAT, SSM/I, TMI, AMSR-E 1987ePresent
CCMP Blended 0.25
lat/lon 4 times/day L4 QuikSCAT, SSM/I, SSMIS, AMSR-E, WindSat, ECMWF Analyses 1987e2011
NCEP-R2 Reanalysis 2.5
lat/lon 4 times/day e NWP and observations 1979ePresent
ERA-Interim Reanalysis 0.75
lat/lon 4 times/day e NWP and observations 1979ePresent
NCEP-CFSR Reanalysis 0.5
lat/lon Hourly e NWP and observations 1979ePresent
NASA-MERRA Reanalysis 0.5
late2/3 lon Hourly e NWP and observations 1979ePresent
NCEP-FNL Analysis 1
lat/lon 4 times/day e NWP and observations 1999ePresent
NCEP-GFS Analysis 0.5
lat/lon 4 times/day e NWP and observations 2004ePresent

Table 2
Main characteristics of the considered buoys. Thus, in order to perform a comparison between the simulta-
neous and valid records of these products, a step-by-step meth-
Buoy Latitude Longitude Distance to the coast
odology was followed: first, a comparison was carried out among
Pen~ as 43
450 N 6
90 3600 W ~20 km all ASCAT products (ASCAT-A 25 km, 12.5 km and coastal products;
Bares 44
30 5400 N 7
370 500 W ~32 km
Villano 43
300 N 9
120 3600 W ~30 km
ASCAT-B 25 km and coastal products) to determine the best ASCAT
Silleiro 42
70 4800 N 9
230 2400 W ~40 km offshore wind database. For this, data from these five ASCAT
Leixo~es 41
10.2' N 9
34.80 W ~65 km products for the year 2013 were compared with measured
Nazare
1 39
31.20 N 9
38.40 W ~45 km offshore winds. After, the best ASCAT product was compared with
Nazare
2 39
33.60 N 9
120 600 W ~10 km
simultaneous (9 a.m./p.m.) records from NCEP-CFSR, NASA-MERRA

diz
Ca 36
280 3700 N 36
280 3700 W ~55 km
and WRF simulation (the only databases that have records coin-
cident in time with ASCAT). Given that OSCAT data (12 a.m./p.m.)
and ASCAT data (9 a.m./p.m.) do not share records coincident in
time, this database was compared separately. Since OSCAT records
coincide in time with data from all the other alternative offshore
wind databases (excluding ASCAT), OSCAT data were compared
with all the remaining databases that present simultaneous
records.
Following this statistical analysis of the simultaneous and valid
records among the offshore wind products, Weibull probability
density functions (PDFs) were used to characterize the local wind
regimes in terms of wind speed distribution frequency, assessing
which database offers the Weibull PDF closest to the one derived
from measured data. The Weibull shape and scale parameters
were computed using the maximum likelihood estimates method
described in Fisher [16]. The most probable and mean wind speed,
derived from the Weibull wind speed distribution, were also used
as comparative metrics. All the valid data records of each database
were used to compute the Weibull PDFs, allowing the assessment
of the hypothetical advantage of a given wind data source having
higher temporal sampling. Finally, offshore wind energy produc-
tion estimates (in the form of the available wind power flux) at
each site derived from measurements and from the alternative
offshore wind data sources was compared, considering wind
speeds between 3.5 and 25 m s1 (following the typical wind
turbine cut-in and cut-off speeds). The wind power flux (also
called wind energy density) is derived from the wind speed
following equation (1), where U is the wind speed and r is the air
density (the standard value of 1.225 kg m3 was assumed).

1 3
Pflux ¼ rU (1)
2
Fig. 1. Buoys locations.
 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444 437

3. Results and discussion compared to all the alternative sources of offshore winds that have
coincident records with OSCAT ascending and descending passes
3.1. Statistical results (12 a.m./p.m.) and with the buoy measurements. With the obvious
exception of ASCAT, all the other alternative offshore wind data-
Table 3 depicts the statistical comparison between the buoys bases have wind records coincident with OSCAT ascending and
and ASCAT products in terms of the mean values of each error descending passes. Thus, Table 5 shows the comparison between
metric for all buoys. This comparison considered data from all OSCAT, WRF, CCMP, NCDC Blended, ERA-Interim, NCEP-CFSR,
ASCAT products and the buoys for the year 2013, which encom- NASA-MERRA, NCEP-GFS, NCEP-FNL and NCEP-RII. Data from the
passes ASCAT-A and ASCAT-B missions. The bias means were year 2011 was used for all wind products and the mean biases were
calculated considering the absolute values of each individual bias to computed using the individual biases absolute values.
assess their magnitude, avoiding mutual cancellations. For each Table 5 shows that WRF and CCMP are the databases with the
error metric mean, the lowest error is bold and underlined for lowest errors. Thus, OSCAT is not able to show better results than
guidance. these databases. Considering Tables 4 and 5, ASCAT coastal product
Table 3 reveals that ASCAT coastal products (A and B) clearly and WRF show the lowest overall errors in representing the winds
standout from the remaining ASCAT databases, showing the lowest in the offshore sites under study. It is also noticeable a tendency for
errors in terms of RMSE, Bias and STDE and the highest correlations all databases to show worse performance in sites located closer to
for the wind speed and direction. The differences between the A the coast. Although this tendency is more pronounced for satellite-
and B versions of the coastal product are residual, what was ex- derived data (ASCAT and OSCAT), due to the well known the limi-
pected since they share the same instrumental configuration and tations of satellite-derived data near the coast due to land masking
design. Both low-resolution ASCAT products (25 km) show the contamination, NWP modelled winds also show higher errors in
highest errors, being the differences among them negligible. The sites located closer to the coast, albeit less markedly than satellite-
12.5 km resolution ASCAT-A product shows an intermediate per- derived winds. These issues are related to the difficulty of NWP
formance. Verhoef and Stoffelen [33] reported that ASCAT high- models to accurately represent winds in areas with strong land-sea
resolution products show a much better agreement with buoy gradients and discontinuities.
measurements than the 25-km products, containing small-scale
phenomena that are less pronounced in the 25-km wind fields.
Although not shown in Table 3, in all buoys (except in the Leixo ~es 3.2. Error dependence on measured wind speed
buoy) the individual biases for the wind speed are positive,
reflecting a tendency for these satellite-derived data to over- Scatterometers limitations in representing low wind speeds
estimate the offshore winds intensity in this region. (due to signal backscatter induced by rain drops and backscatter
Next, the best ASCAT product will be compared with the alter- lower threshold) and high wind speeds (due to atmospheric
native offshore wind databases that have records coincident with attenuation of the signal and backscatter upper threshold) are well
ASCAT ascending and descending passes (9e10 a.m./p.m.). Data known and documented. Thus, it becomes pertinent to analyse if
from ASCAT-A version of the coastal product for the year 2011 was the wind speeds from the alternative wind data sources show any
selected due to the higher availability of valid records (less data variation or dependence with the measured speed. Table 6 repli-
gaps) for this year. Table 4 depicts these results, now presenting the cates the calculations of Table 4 (ASCAT and the databases with
results for each individual buoy together with the mean values. coincident records with ASCAT ascending and descending passes),
Again, the mean biases were computed using the absolute values of and Table 7 the results of Table 5 (OSCAT and the databases with
the individual biases and the lowest mean errors are bold and coincident records with ASCAT ascending and descending passes),
underlined for guidance. but now with the wind speed RMSE and Bias computed for four
Table 4 shows that the ASCAT coastal product shows the best different bins: measured winds below 4 m s1; between 4 and
overall (mean) performance in what is related to the wind speed 8 m s1; between 8 and 12 m s1 and above 12 m s1. Tables 6 and 7
temporal variability (RMSE, STDE and correlation coefficient), show only the mean values for all buoys (the mean biases were
showing also the lowest bias for the wind direction. The WRF computed using the individual biases absolute values).
simulation shows the lowest errors for the wind speed bias and According to Tables 6 and 7, all offshore wind products show
wind direction RMSE, STDE and correlation coefficient. ASCAT worse performance in representing low and high wind speeds,
overestimates the wind speed in all buoys (positive biases) except while the errors are clearly lower for intermediate wind speeds
~es, while the other offshore wind databases show an overall
in Leixo (4e12 m s1). Contrarily to what could be expected from scatter-
tendency to underestimate the wind speed at these sites (negative ometer wind data, ASCAT coastal product shows the worst perfor-
biases), particularly the NASA-MERRA reanalysis. Payan [26] pre- mance only for low wind speeds, improving their accuracy with
viously found that ASCAT has a tendency to overestimate wind increasing wind speeds. Oppositely, OSCAT shows the typical
speeds in the Northern hemisphere and, oppositely, to underesti- scatterometer limitations in accurately representing low and strong
mate them in the Southern hemisphere. Next, OSCAT winds are wind speeds. All databases show positive and large biases for weak
wind speeds. As the wind speed increases, these biases become

Table 3
Statistics of the comparison between all ASCAT offshore wind products and buoy wind measurements. Wind data refer to the year 2013 and to the 10 a.m./pm hours (UTC).

Buoy Database RMSE Bias STDE R2

1
1
1

Speed (m.s ) Direction ( ) Speed (m.s ) Direction ( ) Speed (m.s ) Direction ( ) Speed () Direction ()

Mean ASCAT-A 25 km 1.55 28.66 0.64 7.63 1.40 27.03 0.90 0.76
ASCAT-B 25 km 1.57 29.97 0.64 8.53 1.42 28.24 0.90 0.75
ASCAT-A 12.5 km 1.35 28.32 0.52 8.04 1.23 26.62 0.92 0.77
ASCAT-A 12.5 km coastal 1.26 28.49 0.42 7.92 1.18 26.92 0.93 0.76
ASCAT-B 12.5 km coastal 1.26 28.31 0.42 6.95 1.18 27.05 0.93 0.78
438 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

Table 4
Statistics of the comparison between ASCAT, the alternative sources of offshore wind data that have coincident records with ASCAT and buoy wind measurements. Hourly wind
data refer to the year 2011.

Buoy Database RMSE Bias STDE R2

Speed (m.s1) Direction (
) Speed (m.s1) Direction (
) Speed (m.s1) Direction (
) Speed () Direction ()
~ as
Pen ASCAT-A 12.5 km coastal 1.75 43.79 0.88 11.19 1.52 42.33 0.87 0.48
WRF 5 km 2.01 48.22 0.01 7.28 2.01 47.67 0.79 0.44
NCEP-CFSR 2.19 49.33 0.22 11.14 2.18 48.05 0.76 0.38
NASA-MERRA 2.28 50.37 0.17 13.66 2.27 48.49 0.70 0.42
Bares ASCAT-A 12.5 km coastal 1.01 28.25 0.35 2.40 0.95 28.14 0.96 0.80
WRF 5 km 1.32 27.35 0.08 4.13 1.32 27.04 0.91 0.77
NCEP-CFSR 1.38 27.69 0.03 2.26 1.38 27.60 0.91 0.77
NASA-MERRA 1.73 26.62 0.89 3.23 1.48 26.42 0.89 0.78
Villano ASCAT-A 12.5 km coastal 1.40 26.04 0.50 0.16 1.30 26.04 0.94 0.82
WRF 5 km 1.74 20.94 0.16 3.29 1.73 20.68 0.89 0.84
NCEP-CFSR 1.69 22.12 0.06 3.08 1.69 21.91 0.90 0.82
NASA-MERRA 1.98 21.91 0.65 2.71 1.87 21.74 0.87 0.83
Silleiro ASCAT-A 12.5 km coastal 1.49 37.51 0.53 1.79 1.40 37.47 0.93 0.67
WRF 5 km 1.84 31.17 0.32 1.88 1.81 31.12 0.88 0.70
NCEP-CFSR 1.92 31.28 0.22 5.07 1.91 30.86 0.87 0.70
NASA-MERRA 2.32 35.65 0.68 7.92 2.22 34.76 0.82 0.63
C

adiz ASCAT-A 12.5 km coastal 1.17 32.90 0.37 2.22 1.11 32.83 0.93 0.78
WRF 5 km 1.71 35.64 0.18 7.95 1.70 34.74 0.84 0.72
NCEP-CFSR 1.87 35.53 0.12 7.46 1.86 34.73 0.82 0.74
NASA-MERRA 2.25 44.76 0.79 6.58 2.11 44.27 0.73 0.59
~es
Leixo ASCAT-A 12.5 km coastal 1.29 42.84 0.16 1.24 1.28 42.82 0.94 0.65
WRF 5 km 1.86 37.60 0.60 7.63 1.77 36.82 0.88 0.72
NCEP-CFSR 2.03 39.47 0.39 6.06 1.99 39.00 0.86 0.69
NASA-MERRA 2.78 38.10 1.69 7.67 2.21 37.32 0.80 0.68

1
Nazare ASCAT-A 12.5 km coastal 1.47 25.11 0.25 2.54 1.45 24.98 0.90 0.82
WRF 5 km 1.65 21.99 0.09 0.38 1.65 21.99 0.86 0.86
NCEP-CFSR 1.86 22.13 0.03 0.76 1.86 22.11 0.83 0.85
NASA-MERRA 2.60 29.45 1.63 2.85 2.02 29.32 0.79 0.80

2
Nazare ASCAT-A 12.5 km coastal 1.74 39.10 0.74 5.64 1.57 38.69 0.84 0.58
WRF 5 km 1.69 35.37 0.03 0.96 1.69 35.35 0.82 0.66
NCEP-CFSR 2.27 34.58 1.47 4.43 1.73 34.29 0.80 0.67
NASA-MERRA 2.05 38.58 0.59 2.68 1.96 38.48 0.73 0.55
Mean ASCAT-A 12.5 km coastal 1.42 34.44 0.47 3.40 1.32 34.16 0.91 0.70
WRF 5 km 1.73 32.29 0.18 4.19 1.71 31.93 0.86 0.71
NCEP-CFSR 1.90 32.77 0.32 5.03 1.82 32.32 0.84 0.70
NASA-MERRA 2.25 35.68 0.89 5.91 2.02 35.10 0.79 0.66

lower in magnitude and their signal passes from positive to nega- higher temporal resolution can be reflected in the results, unlike
tive. Thus, in the presence of high wind speeds, the biases are the statistical analysis performed until now that only considered
negative but large in magnitude again. Given this, it can be only the coincident records. Data from the complete year of 2011
concluded that all offshore wind products show a tendency to from all sources of offshore winds was used for the computation of
overestimate weak winds and underestimate strong winds, Weibull PDFs, its parameters and offshore wind energy production
whereas in the presence of intermediate winds this over/under- estimates.
estimation tendency is considerably lower. The only exception is Fig. 2 shows that the WRF and CCMP winds present the best
OSCAT, which systematically shows positive biases showing a results, showing wind speed PDFs closer to the measured ones
behaviour similar to QuikSCAT winds [7]. Wind speed biases can be when considering all buoys. Oppositely, offshore winds from NASA-
quite large for Ku-band scatterometers such as QuikSCAT and MERRA and NCEP-RII reanalyses seem to be the ones with the worst
OSCAT, but significantly lower for C-band scatterometers like overall performance. While in Bares, Villano, C
adiz and Leixo ~es
ASCAT, and are mostly limited to low winds. Unlike Ku-band, C- practically all products show relatively close PDFs (although some
band scatterometers are less prone to be affected by rain, being able databases show distinct PDFs shapes and locations between
to perform quite good wind retrievals in the presence of rain at high themselves and when compared to the measured PDFs), in Pen ~ as,
winds [27]. Silleiro, Nazare
1 and 2 the PDFs are quite different between
themselves. Clearly, the site where all databases, in general, and the
scatterometers, in particular, showed higher errors in describing
3.3. Weibull PDFs and offshore wind energy production estimates the local wind speed distribution is Nazare
2. The fact that this buoy
is moored at a close distance from the shore (10 km) in a land/ocean
The wind speed Weibull PDFs (Fig. 2) for all offshore wind da- transition zone well inside the scatterometers land masking area
tabases (including the measured winds) are compared to assess can explain this bad performance, since scatterometers show lim-
their ability to represent the wind speed distributions at the buoy itations in retrieving accurate wind measurements in such loca-
sites. This analysis is paramount for offshore wind energy assess- tions due to land contamination issues. As for NWP modelled data,
ment applications, which traditionally quantify the average wind their resolution is usually not fine enough to accurately resolve
energetic potential of a given area through wind speed Weibull land-sea discontinuities and medium/small scale atmospheric cir-
PDFs. Each database PDF was computed using all of its valid data culations induced by the terrain.
records, and not only the coincident records. This way, the hypo- Table 8 depicts the mean values (computed using absolute
thetical advantage of a given offshore wind product to have a
 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444 439

Table 5
Statistics of the comparison between OSCAT, the alternative sources of offshore wind data that have coincident records with OSCAT and buoy wind measurements. Wind data
refer to the year 2011 and to the 12 a.m./pm hours (UTC).

Buoy Database RMSE Bias STDE R2

Speed (m.s1) Direction (
) Speed (m.s1) Direction (
) Speed (m.s1) Direction (
) Speed () Direction ()
~ as
Pen OSCAT 12.5 km 2.54 59.42 1.17 9.91 2.25 58.59 0.76 0.34
WRF 5 km 1.74 54.13 0.07 9.07 1.73 53.36 0.86 0.41
CCMP 2.04 53.19 0.41 13.74 2.00 51.39 0.81 0.41
NCDC Blended 3.10 65.38 1.49 18.07 2.72 62.84 0.67 0.28
ERA-Interim 2.56 58.09 0.71 16.42 2.46 55.73 0.71 0.38
NCEP-CFSR 1.98 55.85 0.44 12.49 1.93 54.44 0.83 0.38
NASA-MERRA 2.14 54.72 0.35 15.50 2.11 52.47 0.78 0.42
NCEP-GFS 2.23 51.92 0.91 16.54 2.04 49.21 0.84 0.46
NCEP-RII 4.67 66.22 2.65 17.91 3.84 63.75 0.47 0.30
NCEP-FNL 2.23 52.02 0.91 15.29 2.04 49.72 0.84 0.46
Bares OSCAT 12.5 km 2.39 47.10 0.50 0.06 2.34 47.10 0.75 0.53
WRF 5 km 1.51 31.91 0.05 1.55 1.51 31.87 0.90 0.70
CCMP 1.42 25.72 0.39 0.79 1.36 25.71 0.92 0.81
NCDC Blended 2.17 38.25 0.61 5.12 2.08 37.91 0.80 0.66
ERA-Interim 1.68 29.53 0.13 2.56 1.68 29.42 0.88 0.76
NCEP-CFSR 1.68 37.55 0.09 4.95 1.67 37.23 0.88 0.66
NASA-MERRA 1.89 35.91 0.59 5.10 1.79 35.54 0.85 0.69
NCEP-GFS 1.70 33.95 0.25 4.02 1.68 33.71 0.88 0.69
NCEP-RII 3.48 48.82 1.54 0.90 3.12 48.82 0.68 0.51
NCEP-FNL 1.71 31.46 0.23 3.16 1.69 31.30 0.88 0.73
Villano OSCAT 12.5 km 1.56 33.64 0.81 2.11 1.33 33.58 0.94 0.69
WRF 5 km 1.68 26.87 0.16 4.69 1.67 26.45 0.91 0.79
CCMP 1.92 28.23 0.49 5.35 1.86 27.72 0.89 0.77
NCDC Blended 2.50 35.06 0.42 0.90 2.46 35.05 0.78 0.68
ERA-Interim 1.91 27.13 0.04 5.72 1.90 26.52 0.88 0.79
NCEP-CFSR 1.83 29.95 0.01 4.67 1.83 29.59 0.89 0.74
NASA-MERRA 1.96 24.91 0.54 4.53 1.88 24.49 0.89 0.80
NCEP-GFS 2.05 25.11 0.27 2.43 2.03 25.00 0.86 0.81
NCEP-RII 3.39 43.12 0.37 2.52 3.37 43.05 0.64 0.57
NCEP-FNL 2.31 28.02 0.46 1.93 2.26 27.95 0.83 0.77
Silleiro OSCAT 12.5 km 1.91 46.02 1.02 2.58 1.61 45.95 0.90 0.42
WRF 5 km 1.76 39.05 0.48 0.49 1.70 39.05 0.89 0.57
CCMP 1.80 35.27 0.10 0.19 1.80 35.27 0.87 0.65
NCDC Blended 2.93 48.56 1.39 10.95 2.58 47.31 0.73 0.52
ERA-Interim 1.86 38.51 0.57 10.54 1.77 37.04 0.88 0.61
NCEP-CFSR 2.03 43.03 0.41 2.26 1.99 42.97 0.84 0.54
NASA-MERRA 2.13 38.68 0.39 3.93 2.09 38.48 0.83 0.59
NCEP-GFS 2.09 36.66 0.56 2.78 2.01 36.56 0.84 0.60
NCEP-RII 3.76 56.00 1.52 7.37 3.44 55.51 0.59 0.41
NCEP-FNL 3.23 43.48 1.97 5.27 2.56 43.15 0.73 0.49
C

adiz OSCAT 12.5 km 2.75 50.00 1.04 8.20 2.55 49.32 0.67 0.53
WRF 5 km 1.85 40.37 0.14 6.93 1.85 39.77 0.80 0.69
CCMP 2.11 41.05 0.23 5.40 2.10 40.69 0.74 0.68
NCDC Blended 2.50 51.72 0.88 7.28 2.34 51.21 0.69 0.51
ERA-Interim 2.20 42.28 0.64 6.81 2.11 41.72 0.71 0.65
NCEP-CFSR 2.08 44.18 0.08 2.95 2.08 44.09 0.76 0.62
NASA-MERRA 2.39 49.49 0.32 6.01 2.37 49.12 0.65 0.53
NCEP-GFS 2.09 42.61 0.50 3.25 2.03 42.49 0.78 0.64
NCEP-RII 3.63 67.96 0.19 0.88 3.62 67.96 0.21 0.31
NCEP-FNL 2.84 42.24 1.12 9.59 2.60 41.14 0.69 0.65
~es
Leixo OSCAT 12.5 km 2.25 48.13 0.05 5.08 2.25 47.87 0.82 0.48
WRF 5 km 1.69 33.69 0.82 10.07 1.48 32.15 0.91 0.73
CCMP 1.59 26.38 0.74 3.94 1.41 26.08 0.92 0.80
NCDC Blended 2.37 39.13 0.13 10.91 2.37 37.58 0.76 0.68
ERA-Interim 1.83 30.50 0.87 8.40 1.62 29.32 0.90 0.75
NCEP-CFSR 1.68 32.85 0.52 7.49 1.59 31.99 0.90 0.73
NASA-MERRA 2.74 35.08 1.94 7.38 1.93 34.30 0.84 0.71
NCEP-GFS 1.99 36.38 0.76 4.56 1.84 36.09 0.88 0.69
NCEP-RII 2.92 50.66 0.09 7.07 2.92 50.16 0.72 0.49
NCEP-FNL 1.57 29.42 0.30 6.91 1.54 28.59 0.91 0.79

1
Nazare OSCAT 12.5 km 1.91 44.37 0.77 1.25 1.75 44.36 0.84 0.52
WRF 5 km 1.57 37.78 0.03 1.47 1.57 37.75 0.87 0.65
CCMP 1.63 37.20 0.10 7.38 1.63 36.46 0.86 0.67
NCDC Blended 2.41 42.03 0.81 0.42 2.27 42.03 0.74 0.58
ERA-Interim 1.84 40.03 0.02 4.34 1.84 39.79 0.82 0.62
NCEP-CFSR 1.96 33.63 0.19 0.53 1.95 33.62 0.80 0.67
NASA-MERRA 2.46 39.16 1.30 5.56 2.09 38.77 0.75 0.63
NCEP-GFS 1.96 36.93 0.55 3.76 1.88 36.74 0.81 0.64
NCEP-RII 3.22 48.51 1.15 0.31 3.00 48.51 0.59 0.48
NCEP-FNL 1.98 41.27 0.48 2.06 1.92 41.22 0.82 0.58

2
Nazare OSCAT 12.5 km 2.36 66.02 1.59 52.75 1.74 39.70 0.97 0.34
(continued on next page)
440 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

Table 5 (continued )

Buoy Database RMSE Bias STDE R2

1 1 1
Speed (m.s ) Direction (
) Speed (m.s ) Direction (
) Speed (m.s ) Direction (
) Speed () Direction ()

WRF 5 km 2.21 19.80 0.23 6.65 2.20 18.64 0.51 0.84

CCMP 1.64 22.87 0.28 7.25 1.61 21.70 0.75 0.84
NCDC Blended 4.66 25.81 2.79 7.25 3.73 24.77 0.38 0.78
ERA-Interim 3.07 20.71 1.43 7.50 2.72 19.31 0.54 0.84
NCEP-CFSR 1.70 43.57 0.67 7.25 1.56 42.96 0.77 0.42
NASA-MERRA 1.53 17.61 0.63 0.75 1.39 17.60 0.82 0.90
NCEP-GFS 3.14 24.62 1.93 14.00 2.47 20.25 0.46 0.82
NCEP-RII 2.83 40.55 2.53 14.25 1.27 37.96 0.90 0.34
NCEP-FNL 1.51 21.64 0.21 2.50 1.50 21.50 0.86 0.78
Mean OSCAT 12.5 km 2.21 49.34 0.87 10.24 1.98 45.81 0.83 0.48
WRF 5 km 1.75 35.45 0.25 5.12 1.71 34.88 0.83 0.67
CCMP 1.77 33.74 0.34 5.50 1.72 33.13 0.84 0.70
NCDC Blended 2.83 43.24 1.06 7.61 2.57 42.34 0.69 0.58
ERA-Interim 2.12 35.85 0.55 7.79 2.01 34.86 0.79 0.67
NCEP-CFSR 1.87 40.08 0.30 5.32 1.83 39.61 0.83 0.60
NASA-MERRA 2.15 36.94 0.76 6.10 1.96 36.35 0.80 0.66
NCEP-GFS 2.16 36.02 0.72 6.42 2.00 35.00 0.79 0.67
NCEP-RII 3.49 52.73 1.26 6.40 3.07 51.96 0.60 0.42
NCEP-FNL 2.17 36.19 0.71 5.84 2.02 35.57 0.82 0.66

Table 6
Wind speed RMSE and Bias per buoy wind speed bin between ASCAT, the alternative sources of offshore wind data that have coincident records with ASCAT and buoy wind
measurements. Wind data refer to the year 2011 and to the 10 a.m./pm hours (UTC).

Buoy Database <4 m s1 4-8 m s1 8-12 m s1 >12 m s1
1 1 1 1 1 1
RMSE (m.s ) Bias (m.s ) RMSE (m.s ) Bias (m.s ) RMSE (m.s ) Bias (m.s ) RMSE (m.s1) Bias (m.s1)

Mean ASCAT-A 12.5 km coastal 2.14 1.34 1.13 0.33 1.11 0.02 1.02 ¡0.07
WRF 5 km 2.05 1.13 1.61 ¡0.09 1.58 0.51 1.66 0.80
NCEP-CFSR 2.06 0.98 1.79 0.37 1.78 0.65 2.35 0.72
NASA-MERRA 2.20 1.00 1.87 0.84 2.31 1.76 3.22 2.83

Table 7
Wind speed RMSE and Bias per buoy wind speed bin between OSCAT, the alternative sources of offshore wind data that have coincident records with OSCAT and buoy wind
measurements. Wind data refer to the year 2011 and to the 12 a.m./pm hours (UTC).

Buoy Database <4 m s1 4-8 m s1 8-12 m s1 >12 m s1

RMSE (m.s1) Bias (m.s1) RMSE (m.s1) Bias (m.s1) RMSE (m.s1) Bias (m.s1) RMSE (m.s1) Bias (m.s1)

Mean OSCAT 12.5 km 2.82 1.83 1.85 0.52 1.72 0.01 2.18 0.09
WRF 5 km 1.88 0.94 1.55 0.14 1.66 0.64 1.61 0.65
CCMP 1.99 0.98 1.58 ¡0.07 1.65 0.79 1.97 1.21
NCDC Blended 3.16 2.21 2.25 0.74 2.01 0.25 2.87 0.81
ERA-Interim 2.24 1.14 1.75 0.15 1.94 0.87 2.05 1.12
NCEP-CFSR 2.10 1.11 1.72 0.21 1.89 0.51 1.96 0.37
NASA-MERRA 2.13 1.04 1.88 0.76 2.45 1.93 2.98 2.64
NCEP-GFS 2.10 1.01 1.84 0.22 2.06 0.49 2.26 0.28
NCEP-RII 3.89 2.83 3.15 0.93 3.51 0.45 4.04 0.64
NCEP-FNL 2.24 1.19 2.09 0.10 2.36 0.47 2.66 0.34

values) for all buoys of the Weibull scale (A) and shape (k) pa- was confirmed that all databases (but particularly the scatter-
rameters, wind power flux (Pflux), the mean (Um) and most probable ometers) showed higher errors at sites closer to the coast.
wind speed (Uprob) errors in terms of its deviations from the values Overall, ASCAT (in particular the high-resolution coastal version
derived from measurements. N represents the (average) number of with 12.5 km of horizontal resolution) showed the lowest error
valid wind speed records of each offshore wind product, and all the statistics related to the representation of the temporal variability of
mean errors were computed with absolute values to avoid mutual the wind speed (lowest RMSE, STDE and higher R2) and wind power
cancellation. flux estimations closest to the ones derived from measurements.
Table 8 confirms that the WRF simulation is the offshore wind WRF consistently showed the lowest wind speed errors following
database that shows the best overall results for these metrics, with ASCAT in the aforementioned metrics. However, since WRF wind
the Weibull distribution parameters closest to the ones derived power flux estimation errors are very close to ASCAT ones (8.0% vs.
from measured data although NCEP-GFS and NCEP-FNL depict the 6.2%) and it was the wind database with the best representation of
shape parameter (k) with the lowest mean errors. Albeit CCMP does the wind speed mean state (lowest wind speed bias, mean wind
not show the lowest errors in none of these metrics, it closely fol- speed closest to the measured one and the Weibull PDF's closest to
lows WRF and can be considered as the second best alternative. the ones derived from the measurements), it can be considered that
ASCAT coastal product yields the best wind power flux estimates, a WRF high-resolution simulation is the best alternative offshore
but very closely followed by WRF. Although not shown in Table 8, it wind database for offshore wind energy assessment applications.
 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444 441

Fig. 2. Weibull PDFs for all offshore wind products.

Nevertheless, ASCAT instrument represents a clear progress in QuikSCAT and to its contemporary OSCAT, showing better or
the scatterometry field when compared to its predecessor equivalent error metrics when compared to the other alternative
442 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

Table 8
Weibull parameters, wind power flux, mean and most probable wind speed errors for all databases. Wind data refer to the year 2011.

Buoy Database A (m.s-1) K (m.s-1) Um (m.s-1) Uprob (m.s-1) Pflux (W.m-2) N

(error in %) (error in %) (error in %) (error in %) (error in %)

Mean ASCAT-A12.5 km coastal 14.9% 33.5% 14.5% 41.5% 6.2% 357

ASCAT-A 12.5 km 18.0% 38.4% 17.4% 48.2% 9.1% 303
OSCAT 12.5 km 15.1% 16.7% 14.2% 31.5% 29.7% 324
ASCAT-A 25 km 21.5% 44.9% 21.2% 55.3% 14.4% 320
WRF 5 km 4.7% 11.0% 4.2% 14.9% 8.0% 52561
CCMP 5.5% 12.8% 5.1% 15.3% 13.9% 1460
NCDC Blended 14.6% 24.0% 14.0% 36.0% 18.5% 1191
ERA-Interim 10.1% 13.4% 9.6% 22.0% 17.6% 1460
NCEP-CFSR 6.0% 7.2% 5.7% 12.7% 10.9% 8759
NASA-MERRA 10.8% 17.5% 11.0% 17.1% 34.8% 8760
NCEP-GFS 8.4% 6.7% 8.2% 15.2% 16.3% 1372
NCEP-RII 20.2% 9.6% 19.5% 31.7% 44.2% 1460
NCEP-FNL 14.8% 6.7% 14.5% 17.4% 31.2% 1460

sources of offshore wind data tested in this study. Carvalho et al. [9] wind power flux map computed with WRF wind data, mainly off
showed that QuikSCAT wind error statistics and wind power flux the Portuguese Atlantic coastline. These differences between WRF
estimations were worse than the majority of the alternative and CCMP representation of the mean wind power flux can be
offshore wind databases testes in that study (which besides ASCAT related to the fact that Tables 5 and 8 showed that CCMP tends to
and OSCAT coincide with the ones tested in the present study), overestimate the measured wind speed more than WRF. Although
while the results presented in Table 5 showed that OSCAT instru- the differences between WRF and CCMP in terms of wind speed
ment has not proven to be as reliable as ASCAT, showing worse overestimation are small, the wind power flux varies with the wind
performances when compared to other offshore wind products. speed cubed, which will significantly amplify these differences in
Besides its validity as an alternative wind database for offshore terms of wind speed.
wind energy applications, ASCAT can also be seen as the best It is also noticeable the higher resolution of WRF simulation,
alternative source of offshore wind data for other oceanographic making it able to depict in greater detail smaller localized areas
and meteorological purposes where the temporal accuracy (ability where the wind power flux varies. CCMP wind power flux map
to accurately represent the wind speed temporal variability) is the shows a somewhat uniform mean wind power flux off the Portu-
main priority. However, given its low temporal resolution guese Atlantic coast (approx. 100 km offshore), and the wind power
(maximum of 2 records per day), ASCAT should not be considered flux gradient is also more uniform between the coast and 100 km
for applications that require higher temporal sampling, as is the offshore than WRF wind power flux map. In the latter, it is visible
case of studies where the representation of the wind diurnal cycle that although the wind power flux along he Portuguese Atlantic
is important, for example. Given its good results (closely following coast is not very high until 100e150 km off the coast, there are two
WRF metrics), CCMP can also be seen as a valid and readily available localized areas where the wind power flux is higher right until the
alternative to WRF and ASCAT: WRF modeling tasks require coast: near Cape Roca (near Lisbon) and Cape St. Vincent (near
considerable computational resources, time and expertise to obtain Sagres, in the southwest tip of Portuguese mainland). Moreover,
quality results; satellite-derived swath data has low temporal res- WRF wind energy resource map shows that the Strait of Gibraltar
olution and needs substantial quality control, and its swath format has higher mean wind power flux. Although the higher wind en-
makes it not directly usable to assess offshore wind energy po- ergetic resource near Cape St. Vincent is also detectable in CCMP
tential using the traditional methods (wind energy resource map- wind power flux map, the ones near Cape Roca and in the Strait of
ping, constant and stable periodicity of wind data at a given site, Gibraltar are not.
etc.). Besides CCMP lower spatial and temporal resolution when
The latter constitutes another advantage of gridded datasets compared to WRF capabilities, the latter allows the derivation of
such as the WRF simulation and CCMP, which allow a direct wind data (speed and direction) and, subsequently, wind energy
computation of offshore wind energy resource maps and to derive estimations at any desired height above ground or sea surface level.
constant, stable and gap-free wind data time series. Section 3.4 Since typical wind turbines hub heights are of around 100e120 m
shows offshore wind energy resource maps using WRF and CCMP above ground/sea surface level, a direct derivation of wind energy
datasets for the area under study. estimates at the wind turbine hub height becomes very useful,
avoiding the need to perform inter- or extrapolations that consti-
tute additional sources of error and uncertainty. Fig. 4 depicts the
3.4. Offshore wind energy resource mapping
wind power flux for the area under study at a height of 120 m above
ground/sea surface level, directly derived from WRF simulation 3-
Offshore wind energy resource maps were computed using WRF
dimensional wind grid.
and CCMP wind data for the area under study. Following what was
Fig. 4 shows that the wind power flux at this height is sub-
done in section 3.3 (Table 8), the wind energy production potential
stantially higher than the one seen for 10 m above ground/sea
was assessed through the wind power flux (W.m2) at 10 m above
surface level, reaching values of 500 W m2 off the northwest end
ground/sea surface level.
of the Iberian Peninsula and in the Strait of Gibraltar. Higher than
Fig. 3 shows that, for both wind energy resource maps, the
average wind power flux annual means are still seen near Capes
offshore areas with higher wind power flux for the area under
Roca and St. Vincent, reaching values around 400e450 W m2.
study are located off the northwest end of the Iberian Peninsula,
Thus, besides WRF's proven ability to produce reliable wind data
reaching annual mean values in the range of 300e350 W m2. It is
when compared to other alternative offshore wind databases, its
visible that the wind energy resource map derived from CCMP
capability to deliver wind data at typical wind turbines hub heights
wind data show higher values of mean wind power flux than the
 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444 443

Fig. 3. Offshore wind resource maps for the area under study using WRF (left) and CCMP (right) wind data e 10 m above ground/sea surface level.

constitutes a decisive advantage. mean state (lowest wind speed bias, mean wind speed closest to
the measured one and the Weibull PDF's closest to the ones derived
from the measurements). In addition to its ability to produce reli-
4. Conclusions able wind data, WRF allows a direct computation of offshore wind
energy resource maps and wind energy production estimates at
This study compared different alternative sources of offshore typical wind turbines hub heights, an advantage of paramount
wind data, namely the new-generation scatterometers ASCAT and importance that avoids the need to perform inter- or extrapolations
OSCAT, with in-situ measured offshore winds to assess which that add uncertainty to the wind data. Due to its good performance,
alternative offshore wind databases are able to better represent the CCMP can also be seen as a valid and readily available alternative to
local wind regimes and provide offshore wind energy production WRF and ASCAT, particularly in cases where time and/or resources
estimates closer to the ones derived from the measurements at the to consider the use of the WRF model or ASCAT swath data are
selected buoy locations. limited, or not adequate for the desired purposes (need for gridded
The results presented in this study show that although ASCAT data, higher spatial resolution and/or temporal sampling).
(particularly the high-resolution coastal version) showed the Nevertheless, it should be borne in mind that ASCAT instrument
lowest error statistics related to the representation of the temporal represents a clear progress in the scatterometry field when
variability of the wind speed (lowest RMSE, STDE and higher R2) compared to its predecessor QuikSCAT and the contemporary
and wind power flux estimations closest to the ones derived from OSCAT, and it can be seen as the best alternative source of offshore
measurements, a WRF high-resolution simulation can be consid- wind data for oceanographic and/or meteorological applications
ered as the best alternative offshore wind database for offshore where an appropriate representation of the wind temporal vari-
wind energy assessment applications. This choice is based on the ability is the main priority, as long as its temporal resolution is
fact that WRF's simulation performance in the estimation of the enough for such applications. ASCAT showed improved ability to
local wind power fluxes was very similar to ASCAT one, and it was retrieve wind speeds under high speeds regimes and under rain
the wind database with the best representation of the wind speed conditions when compared to OSCAT and QuikSCAT scatterometers
(considering past studies), consequence of its C-band operation
frequencies (in opposition to the traditional Ku-band instruments
such as OSCAT and QuikSCAT), although it still shows limitations in
accurately retrieving winds at low wind speeds regimes.
An issue common to all offshore wind products tested in this
study is the degradation of their performance in sites located closer
to the coast. This tendency is more pronounced for satellite-derived
data (ASCAT and OSCAT), due to the well-known land masking
limitations of scatterometers, although NWP modelled winds also
showed higher errors in sites located closer to the coast albeit less
markedly than satellite-derived winds.
For the offshore wind energy sector, the findings of this study
are expected to have a positive impact since wind energy resource
maps/atlases and estimation of wind energy production estimates
are made, in the absence of measured data, using mainly satellite
and/or NWP offshore wind data. Thus, an informed choice of the
most accurate offshore wind databases will most definitely provide
much more realistic wind energy production estimates that will
positively impact the preliminary planning stages of offshore wind
energy exploration projects, which can determine their economic
Fig. 4. Offshore wind resource maps for the area under study - WRF simulation at viability (or not). Besides the direct interest of the offshore wind
120 m above ground/sea surface level.
444 D. Carvalho et al. / Renewable Energy 102 (2017) 433e444

energy agents, the findings presented here can be of great value for planetary boundary layer parameterizations for onshore and offshore areas in
the Iberian Peninsula, Appl. Energy 135 (2014d) 234e246.
wave energy, climate, oceanic, meteorological and wave energy
mez-Gesteira, C. Silva Santos, Potential impacts of
[13] D. Carvalho, A. Rocha, M. Go
resource assessment applications that focus on ocean areas where climate change on European wind energy resource under the CMIP5 future
locally acquired wind data are either not available or is insufficient climate projections, Renew. Energy 101 (2017) 29e40.
and, therefore, alternative sources of offshore wind data have to be [14] D.B. Chelton, M.H. Freilich, Scatterometer-based assessment of 10-m wind
analyses from the operational ECMWF and NCEP numerical weather predic-
considered. tion models, Mon. Weather Rev. 133 (2005) 409e429.
[15] R.T. Edson, Comparisons and evaluations between the Oceansat-2 (OSCAT)
Acknowledgements and ASCAT scatterometers over tropical cyclones, in: 31st Conference on
Hurricanes and Tropical Meteorology, March 30-April 04 2014, San Diego, CA,
2014.
This work was partially supported by Xunta de Galicia under the [16] R.A. Fisher, Statistical Methods and Scientific Inference, Oliver and Boyd,
project “Programa de Consolidacio
n e Estructuracio

n de Unidades Edinburgh, 1956.
[17] A.M. Foley, P.G. Leahy, A. Marvuglia, E.J. McKeogh, Current methods and ad-
de Investigacio
n Competitivas: Grupos de Referencia Competitiva”
vances in forecasting of wind power generation, Renew. Energy 37 (1) (2012)
(GRC2013-001) co-funded by the European Regional Development 1e8.
Fund (FEDER) The authors would like to express their gratitude to [18] A.B. Kara, A.J. Wallcraft, M.A. Bourassa, Air-sea stability effects on the 10 m
winds over the global ocean: evaluations of air-sea flux algorithms,
all climate, meteorological and oceanographic institutions referred J. Geophys. Res. 113 (2008) 4009e4014.
in the text, for providing the data used in this work. This study was [19] I. Karagali, M. Badger, A.N. Hahmann, A. Pen ~ a, C.B. Hasager, A.M. Sempreviva,
supported by FEDER funds through the “Programa Operacional Spatial and temporal variability of winds in the northern european seas,
Renew. Energy 57 (2013) 200e210.
Factores de Competitividade e COMPETE” and by Portuguese na-
[20] H.-G. Kim, H.-J. Hwang, S.-H. Lee, H.-W. Lee, Evaluation of SAR wind retrieval
tional funds through FCT e Fundaça ~o para a Cie
^ncia e a Tecnologia, algorithms in offshore areas of the Korean Peninsula, Renew. Energy 65
within the framework of Project “Urban Atmospheric Quality, (2014) 161e168.
Climate Change and Resilience.” EXCL/AAG-MAA/0383/2012. [21] R. Kumar, A. Chakraborty, A. Parekh, R. Sikhakolli, B.S. Gohil, A.S.K. Kumar,
Evaluation of oceansat-2-derived ocean surface winds using observations
from global buoys and other scatterometers, IEEE Trans. geosci. remote Sens.
References 51 (5) (2013) 2571e2576.
[22] W.T. Liu, W. Tang, Equivalent Neutral Wind, JPL Publication, 96e17, Jet Pro-
[1] I. Alvarez, M.M. Gomez-Gesteira, M. deCastro, D. Carvalho, Comparison of pulsion Laboratory, California Institute of Technology, Pasadena, California,
different wind products and buoy wind data with seasonality and interannual 1996.
climate variability in the southern Bay of Biscay (2000-2009), Deep Sea Res. [23] C. Mattar, D. Borvar
an, Offshore wind power simulation by using WRF in the
Part II Top. Stud. Oceanogr. 106 (2014) 38e48. central coast of Chile, Renew. Energy 94 (2016) 22e31.
[2] R. Atlas, R.N. Hoffman, J. Ardizzone, S.M. Leidner, J.C. Jusem, Development of a [24] C.A. Mears, D.K. Smith, F.J. Wentz, Comparison of special sensor Microwave
New Cross-calibrated, Multiplatform (CCMP) Ocean Surface Wind Product. imager and buoy-measured wind speeds from 1987 to 1997, J. Geophys. Res.
AMS 13th Conference on Integrated Observing and Assimilation Systems for 106 (11) (2001) 719e729.
Atmosphere, Oceans, and Land Surface (IOAS-aols), Phoenix, Arizona, 2009. [25] A.S. Monin, A.M. Obukhov, Osnovnyezakonomernostiturbulentnogopereme-
[3] I. Balog, P.M. Ruti, I. Tobin, V. Armenio, R. Vautard, A numerical approach for shivanija v prizemnom sloe atmosfery (basic laws of turbulent mixing in the
planning offshore wind farms from regional to local scales over the Medi- atmosphere near the ground), AN SSSR, Tr. Geofiz. 24 (151) (1954) 163e187.
terranean, Renew. Energy 85 (2016) 395e405. [26] Payan, Improvements in the Use of Scatterometer Winds in the Operational
[4] A.R. Bento, P. Martinho, C.G. Soares, Numerical modelling of the wave energy NWP System at Meteo-France.10th International Winds Workshop, Tokyo,
in Galway Bay, Renew. Energy 78 (2015) 457e466. Japan, 2010.
[5] M.A. Bourassa, D.M. Legler, J.J. O'Brian, S.R. Smith, SeaWinds validation with [27] M. Portabella, A. Stoffelen, W. Lin, A. Turiel, A. Verhoef, J. Verspeek,
research vessels, J. Geophys. Res. 108 (3019) (2003) 16. J. Ballabrera-Poy, Rain effects on ASCAT retrieved winds: towards an
[6] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, WRF model ocean surface wind improved quality control, IEEE Trans. Geosci. Remote Sens. 50 (2012)
simulation forced by different reanalysis: comparison with observed data 2495e2506.
along the Iberian Peninsula coast, Ocean. Model. 56 (2012) 31e42. [28] S.I. Rani, M. Gupta, P. Sharma, V.S. Prasad, Intercomparison of Oceansat-2 and
[7] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, I. Alvarez, C. Silva Santos, Com- ASCAT winds with in-situ buoy observations and short-term numerical
parison between CCMP, QuikSCAT and buoy winds along the Iberian Peninsula forecasts, Atmos. Ocean 52 (1) (2014) 92e102.
coast, Remote Sens. Environ. 137 (2013a) 173e183. [29] E. Sharp, P. Dodds, M. Barrett, C. Spataru, Evaluating the accuracy of CFSR
[8] D. Carvalho, A. Rocha, C.S. Santos, R. Pereira, Wind resource modelling in reanalysis hourly wind speed forecasts for the UK, using in situ measurements
complex terrain using different mesoscale-microscale coupling techniques, and geographical information, Renew. Energy 77 (2015) 527e538.
Appl. Energy 108 (2013b) 493e504. [30] T.H. Soukissian, Use of multi-parameter distributions for offshore wind speed
[9] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, C. Silva Santos, Comparison of modeling: the Johnson SB distribution, Appl. Energy 111 (2013) 982e1000.
reanalyzed, analyzed, satellite-retrieved and NWP modelled winds with buoy [31] T.H. Soukissian, A. Papadopoulos, Effects of different wind data sources in
data along the Iberian Peninsula coast, Remote Sens. Environ. 152 (2014a) offshore wind power assessment, Renew. Energy 77 (2015) 101e114.
480e492. [32] J.E. Stopa, J.-F. Filipot, N. Li, K.F. Cheung, Y.-L. Chen, L. Vega, Wave energy
[10] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, C. Silva Santos, WRF wind simu- resources along the Hawaiian Island chain, Renew. Energy 55 (2013)
lation and wind energy production estimates forced by different reanalyses: 305e321.
comparison with observed data for Portugal, Appl. Energy 117 (2014b) [33] A. Verhoef, A. Stoffelen, Validation of ASCAT 12.5-km Winds, Version 1.2. SAF/
116e126. OSI/CDOP/KNMI/TEC/RP/147, EUMETSAT Technical Report, Available at:, 2009
[11] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, C. Silva Santos, Offshore wind http://www.knmi.nl/scatterometer/publications/.
energy resource simulation forced by different reanalyses: comparison with [34] H.-M. Zhang, R.W. Reynolds, J.J. Bates, Blended and Gridded High Resolution
observed data in the Iberian Peninsula, Appl. Energy 134 (2014c) 57e64. Global Sea Surface Wind Speed and Climatology from Multiple Satellites:
[12] D. Carvalho, A. Rocha, M. Go
mez-Gesteira, C. Silva Santos, Sensitivity of the 1987-Present”. American Meteorological Society 2006 Annual Meeting, Paper
WRF model wind simulation and wind energy production estimates to #P2.23, Atlanta, GA, January 29-February 2, 2006.