Surveys in Geophysics

, Volume 29, Issue 6, pp 471–497

Review of Methodologies for Offshore Wind Resource Assessment in European Seas

Authors

    • Institute of Atmospheric Sciences and ClimateISAC-CNR c/o CRATI, Zona Industriale Lamezia Terme
    • Wind Energy Division, Risø National Laboratory for Sustainable EnergyTechnical University of Denmark
  • R. J. Barthelmie
    • Atmospheric Science Program, Department of GeographyIndiana University
    • Institute for Energy Systems, School of Engineering and ElectronicsThe University of Edinburgh
  • S. C. Pryor
    • Atmospheric Science Program, Department of GeographyIndiana University
Original Paper

DOI: 10.1007/s10712-008-9050-2

Cite this article as:
Sempreviva, A.M., Barthelmie, R.J. & Pryor, S.C. Surv Geophys (2008) 29: 471. doi:10.1007/s10712-008-9050-2

Abstract

The wind resource offshore is generally larger than at geographically nearby onshore sites, which can offset the higher installation, operation and maintenance costs associated with offshore wind parks. Successful offshore wind energy development relies to some extent on accurate prediction of wind resources, but since installing and operating a meteorological mast in situ is expensive, prospective sites must be carefully evaluated. Accordingly, one can conceptualize the wind resource assessment process as a two-phase activity: (i) an evaluation of wind resources at the regional scale to locate promising wind farm sites and (ii) a site specific evaluation of wind climatology and vertical profiles of wind and atmospheric turbulence, in addition to an assessment of historical and possibly future changes due to climate non-stationarity. Phase (i) of the process can involve use of in situ observations of opportunity derived from ships, lighthouses and buoys in conjunction with model tools and remote sensing products. The reliability of such data sources has been extensively investigated in different national and European projects especially in Northern Europe, and the results are summarized herein. Phase (ii) of the project often still requires in situ observations (which may or may not be supplemented with ground-based remote sensing technologies) and application of tools to provide a climatological context for the resulting measurements. Current methodologies for undertaking these aspects of the resource assessment are reviewed.

Keywords

Wind energyOffshoreResources assessmentEuropean seasWind mappingWind climatologyAtmospheric modellingIn situ observationsRemote sensing

1 Introduction

Installed wind energy capacity around the world has been growing, with Europe leading the global market. The International Energy Agency (IEA 2008) stated that, at the end of 2007, the wind energy capacity installed world-wide was 94,123 MW. More than half of this capacity (57,136 MW) was installed in the EU-27 (EU with 27 countries), representing a ten-fold increase since 1996 (European Wind Energy Association (EWEA) 2008). Currently, offshore installations are a comparatively small part of the wind turbine market. However, since 1999, large wind farms with wind turbines up to 5 MW have been erected offshore especially in the North Sea and Baltic Sea, i.e. Denmark, UK, Netherlands, Sweden, Ireland and Germany (Mechali et al. 2006; Barthelmie 2007) summing up to more than 1,000 MW wind power capacity installed by the end of 2007 (Fig. 1, EWEA 2008). However, with the expanding demand for renewable energy there is a need for larger wind farms located in regions of high wind resource, which has prompted a move towards increased emphasis on harnessing offshore wind resources. Moving offshore presents particular challenges to the wind energy industry but also many advantages (IEA 2005; Barthelmie and Pryor 2006; Koutiva et al. 2006; Negra et al. 2006; Wilhelmsson et al. 2006; Firestone and Kempton 2007). The higher wind resources at offshore sites and greater persistence of winds in power generating classes (Pryor and Barthelmie 2001), coupled with avoidance of many land use conflicts, mean offshore wind is set to develop in a significant way; and the potential offshore market is the main driver for large turbine technology development (Barthelmie 2007). Accordingly, large installations are planned off the Baltic Sea and North Sea coasts of Denmark and Germany (Akhmatov and Knudsen 2007; Akhmatov et al. 2007) and the UK (www.bwea.co.uk; French et al. 2005) and there is also increased interest in developing offshore wind farms in the Mediterranean region. Figure 2 gives an overview of the installed and planned offshore wind farms in Europe. Figure 3 shows projections of installed offshore wind capacity in Europe out to 2020 from the European Wind Energy Association for two scenarios. The first scenario is shown as “minimal” and indicates the likely increase in installed capacity without new policy initiatives, while the “policy impetus” scenario indicates the possible increase in installed capacity achievable if new policy initiatives are enacted.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig1_HTML.gif
Fig. 1

Offshore wind market development in Europe (1991–2007) (Courtesy of European Wind Energy Association (EWEA 2007))

https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig2_HTML.gif
Fig. 2

Location of existing (black) and planned (light blue) wind farms in North Europe (Top) and in South Europe (Bottom). Courtesy of EWEA (EWEA 2007)

https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig3_HTML.gif
Fig. 3

Offshore wind development expressed as cumulating capacity for 5-year periods up to 2020. Two scenarios are presented a “minimal” scenario (white), and a “policy impetus” scenario (orange), where the latter implies development and application of substantial policy measures to facilitate developments. Courtesy of EWEA (EWEA 2007)

1.1 Fundamentals of Wind Energy

The amount of energy a wind turbine can convert to electricity depends on the cube (the third power) of the wind speed. Power (P) is energy transfer per unit of time and the electrical power is usually measured in kilowatts (kW) and depends on air density (ρ), wind speed (U) and on the turbine rotor radius (r). Power may be measured at any point in time, whereas energy has to be measured over a certain time interval, i.e. a second, an hour, or a year.

The installed wind power (capacity) is determined by the number of turbines and the wind turbine rated power that indicates the energy produced per hour of operation when the turbine runs at its maximum performance. At a reasonable wind site, a typical wind turbine will produce electricity 70–85% of the time, and capacity factors (i.e. the annual production divided by the potential electricity production) can exceed 40% at some offshore sites (Barthelmie 2007).

The electrical power output produced at different wind speeds by a given turbine is described by a Turbine power curve (Fig. 4). Below a wind speed threshold no power is produced, above that threshold (the cut-in wind speed) power production increases strongly with wind speed. Above a second wind speed threshold the power production is independent of wind speed, and above a third threshold (typically 25 m s−1) power production ceases when the turbine is shut-down to avoid structural damage. Such conditions are relatively seldom observed even in high wind speed locations.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig4_HTML.gif
Fig. 4

Power curve as a function of wind speed from a turbine VESTAS V66-1650, 66 diameter rotor and rated power of 1650 Kw

To estimate how much energy a wind turbine produces in one year, the annual frequency distribution of wind speeds (Fig. 5) along with the frequency distribution of flow in directional sectors (typically of the order of 12, 30º sectors) (wind rose) (Fig. 5) must be known. The former directly dictates the wind power, while the latter is important for forecasting purposes and to calculate the loss of power caused by the reduction of wind speed due to the wakes generates behind turbines. This information is called the wind climatology. While a 30-year period is often used to define “climate normals” in atmospheric science, shorter periods (of perhaps no more than five years) are frequently used in the wind energy community. The wind climatology must be complete to be able to portray accurately the yearly cycle of wind speed (intra-annual variability) and long enough to keep account of inter-annual variations. Turbine designers need the wind climatology to optimise turbine design, so as to minimise generation costs. Wind farm investors and operators need this information to estimate their income from electricity generation.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig5_HTML.gif
Fig. 5

Experimental wind rose (left) and wind speed histogram (right) with and fitted Weibull distribution with the relevant A and k parameters (This example is taken from a representative offshore site in Denmark)

The probability density distribution for wind speed is typically well approximated by a two-parameter Weibull distribution function:
$$ f\left( u \right) = \left( {k /A} \right)\left( {U /A} \right)^{{K - 1}} \exp \left[ { - \left( {U /A} \right)^{k} } \right] $$
(1)
where A is the scale parameter, which is related to the mean of the distribution and k is the dimensionless shape parameter, which describes the spread (dispersion) of the distribution. For k > 1 and the same value of A a larger k indicates a narrower wind speed distribution, (Fig. 6). In locations with low average wind speed and a large amount of calms i.e. wind speed below 1 m s−1, k might be less than one and the Weibull distribution will become an exponential distribution.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig6_HTML.gif
Fig. 6

Variation of the Weibull distribution parameters as a function of the shape k and scale A parameters. The figure on the left depicts a low wind speed frequency distribution with a large percentage of calms, while the other two frames show higher wind speed regimes with larger wind energy potential

Although the annual electricity production varies enormously depending on the wind climate, production of electricity from wind can be cost-effective under a wide array of circumstances depending also on a number of factors other than wind speed such as:
  • The cost of alternatives like diesel in stand-alone systems;

  • The distance to the nearest suitable grid connection;

  • The structure of support mechanisms and financing; and

  • Installation costs. Including:

• Foundations;

• A transformer to convert the low voltage current from the turbine to current for the local electrical grid;

• Telephone connection for remote control and surveillance of the turbine; and

• Cabling costs.

A comprehensive description of wind energy fundamentals is beyond the scope of this paper. For further information, we suggest a guided tour on the home page of the Danish Wind Energy Association at http://www.windpower.org/en/tour.htm where a nice crash course for children http://www.windpower.org/en/kids/index.htm is also implemented. General information is also provided in Manwell et al. (2002). Information on the wind energy market can also be found on the home page of the EWEA at www.EWEA.org.

1.2 Wind Speeds in the Atmospheric Boundary Layer

The lower part of the atmosphere is directly influenced by the Earth’s surface and responds to surface forcing (thermal and mechanical) with a time scale of about an hour or less. This part of the atmosphere is defined as the Atmospheric Boundary Layer (ABL) with a depth depending on the strength of the forcing but typically between 100 m (during night) and 3000 m (during sunny and/or windy days). The atmospheric Surface Layer (SL) is accordingly defined as the part of the ABL where fluxes from the surface vary within 10% of their surface values (Stull 1988), and it is in this layer that the majority of existing turbines are located. The SL is continuously turbulent over its whole depth and, specifically, the wind increases with height following a logarithmic-linear relationship defined by:
$$ U\left( z \right) = \frac{{u_{ * } }}{{z_{0} }}\left[ {\hbox{In}\frac{z}{{z_{0} }} - \Uppsi_{m} \left( {\frac{z}{L}} \right)} \right] $$
(2)
where U(z) is the mean wind speed at height z, u* is the friction velocity related to the stress at the surface, κ is the von-Karman constant, zo is the roughness length (over the sea ~0.0001 m) defined as the height above the surface at which the mean wind becomes zero when extrapolating the logarithmic wind-speed profile downward to the surface. Ψm(z/L) is a function that takes into account the effects of the atmospheric stability through the stability index z/L where z is the height above the ground and L is the so called Obukhov length defined as
$$ L = - \frac{{u_{ * }^{3} }}{{\overline{{\frac{g\kappa }{T}\overline{{w^{\prime} T_{v}^{\prime} }} }} }} $$
(3)
where \( \overline{T} \) is the mean temperature, g = 9.8 m s−2 is the acceleration of gravity, and \( \overline{{w^{\prime} T_{v}^{\prime} }} \) is the virtual kinematic vertical heat flux.
In Fig. 7, we show different vertical profiles of mean wind speed under stable (z/L > 0), neutral (z/L~0), and unstable conditions (z/L < 0).
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig7_HTML.gif
Fig. 7

Variation of the vertical wind speed profile (where the U at height z, is normalized by the wind speed at a height of 10 m (i.e. U10) as a function of the stability parameter z/L. The left frame shows unstable conditions, the middle frame a neutral profile and the right frame a profile characteristic of stable conditions

Wind speed above the surface depends then on the landscape, on thermal and mechanical surface characteristics, and on local climate conditions.

The stability of the atmosphere depends on the temperature T of an air parcel with respect to the ambient temperature T1. If a parcel of air near the Earth’s surface is heated (i.e. during sunny days) whether or not it continues to rise will depend upon how the temperature difference ΔT = TT1 changes with altitude. The rising parcel of air will lose heat because it expands as atmospheric pressure falls, with a consequent drop of temperature. If ΔT remains positive with increasing altitude, the parcel of air will continue to rise and the atmosphere in this circumstance is said to be unstable. If ΔT becomes negative with increasing altitude, the air parcel will lose its buoyancy, and sink back to its original position; in this case the atmosphere is said to be stable (i.e. during night when the Earth surface release its heat to the above atmosphere). If T1 is the same as the temperature of the air parcel, the atmosphere is said to be neutral (i.e. during cloudy and windy days/nights where mechanical turbulence generated by the surface is the main driving forcing). While onshore thermal stability generally shows a daily cycle, offshore atmospheric stability depends on the temperature difference between sea and the air blowing over it (Barthelmie et al. 1996). Unlike the majority of inland sites, offshore locations frequently exhibit the highest wind speeds at night. These ‘inverted diurnal cycles’ are likely derived from differences in atmospheric stability, advection and thermally induced circulations such as sea breezes (Barthelmie et al. 1996; Coelingh et al. 1998; Lavagnini et al. 2003).

The relative paucity of wind speed observations offshore makes wind energy resource assessment even more challenging than for onshore locations. Motta et al. (2005) present information regarding the occurrence of different stability conditions on various time scales for offshore and coastal sites and demonstrate the dependence on wind speed and direction (i.e. fetch). An important issue for wind energy is the extrapolation of the experimental mean wind speed at higher level since measurements are rarely available at turbine hub-heights (Motta et al. 2003; Lange et al. 2004; Tambke et al. 2005; Barthelmie and Pryor 2006). The influence of thermal stratification on vertical profiles of wind speed is thought to be larger than over land due to lower mechanically generated turbulence (Barthelmie 1999). Thus, there is a greater propensity for non-neutral stability and thus non-zero values of Ψm(z/L) in Eq. (2). Van Wijk et al. (1986) and Motta et al. (2005) found inclusion of stability correction improves the prediction of the wind profile. However, recent evidence from wind speed profiles measured above 50 m suggest that the use of similarity theory may not be adequate for offshore wind speed profiles above 50 m (Tambke et al. 2005; Gryning et al. 2007).

A further issue of importance in dictating offshore wind resources is that many, if not all, of the existing offshore wind farms are located in “coastal areas” (i.e. the zone extending from the coastline where the wind speed and turbulence profiles are not in equilibrium with the underlying sea surface). Results suggest that the distance from the coastline over which wind speed vertical profiles are not at equilibrium with the sea surface extends up to 20 km and possibly 70–100 km from the coast (Barthelmie et al. 2007b; Hunter et al. 2007). Wind climates in offshore coastal areas deviate somewhat from those found over adjacent land areas with the most obvious difference being that wind speeds are generally higher than in onshore coastal zone, principally due to the lower surface roughness, which also results in lower ambient turbulence. In terms of resource assessment, changing surface roughness with wind speed (Charnock 1955) and tidal variations do not exert a strong influence of offshore wind climates (Barthelmie 2001a) unless tidal flats are exposed at low water. Because the wind speed profile in coastal area has not fully adjusted to the sea surface, there is difficulty in extrapolating near-surface measurements to turbine hub heights and there is also the possibility for high wind shear.

1.3 Focus of this Paper

Resource assessment covers a range of scales in space and time and plays a major role in determining whether exploitation of the wind resource is financially viable. It can be conceptualized as a two step process:
  1. (i)

    Regional wind resources assessment. This step is needed to evaluate whether the resource in a region is of sufficient magnitude to be worthy of a more detailed and costly assessment. Regional wind climatology may be determined using both existing measurements and modelling approaches to produce offshore wind resource maps that can be used by developers to identify a number of potential sites in prospective areas.

     
  2. (ii)

    Site specific analyses of the resource magnitude in candidate areas. In this step, a more detailed analysis of specific areas that exhibit high potential is performed. The goal is to determine the magnitude of the available resource and provide an initial assessment of possible sites and the most cost-effective wind farm size. To start this process, most wind farm developers will conduct on- site assessment for a minimum of one complete year. Here, we stress that one-year measurement period is not enough to conduct an accurate wind resources assessment; however, this short-term analysis will be related to a wind climate from long-term measurements at nearby sites or from model outputs in order to identify whether the measured wind speed is representative on the timescales the wind farm is expected to be producing electricity (20–30 years).

     
The sketch in Fig. 8 depicts stage (i) and (ii). In this review paper, we present an overview of methods for regional wind resource assessment and estimating site-specific power production potential using examples drawn from Europe.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig8_HTML.gif
Fig. 8

Overview schematic of the energy assessment procedure as applied to wind energy developments

2 Mapping Wind Resources at the Regional Scale

2.1 A Brief Review of European Projects

A number of projects have previously attempted to map wind resources both on- and off-shore. Perhaps the best known of these—the European Economic Community (EEC) “European Wind Atlas” project, which ended in 1988, led to the development of the “Wind Atlas Analysis and Application Program” (WAsP) (Troen and Petersen 1989; Mortensen et al. 2005) (Sect. 2.2.4). This program estimates regional and local surface wind statistics on the basis of conventional weather observations, and orographic and topographic information. It is one of a variety of methodologies that are now applied for predicting long-term wind speeds in offshore areas at the regional and site level (Sect. 3).

Specific offshore wind mapping projects funded by the European Commission include:
  1. (i)

    An extension of the European Wind Atlas developed by interpolation of onshore measurements (Petersen 1992). The European Wind Atlas employs meteorological data from a selection of monitoring stations, and shows the distribution of wind speeds at the regional to continental scale. It has been used extensively by developers and governments to estimate the size of the wind resource and regional variations. For offshore resources, coastal sites were the main source of data for the atlas, although some offshore measurements from lightships were also employed.

     
  2. (ii)

    An assessment of the offshore wind potential in the EU which was based on extrapolation of data from the voluntary observer fleet (ships) using WAsP (Garrad et al. 1993).

     
  3. (iii)

    Predicting Offshore Wind Energy Resources (POWER) which was based on downscaling of geostrophic wind speeds (at 0.5 ° × 0.5 °) (Benjamin and Miller 1990) from a gridded sea-level pressure data set using WAsP. Wind speeds were estimated for heights of between 10 and 130 m height for all European waters and were generally in good agreement with resource estimates derived using WAsP in combination with data from land based stations (Petersen 1992). Results from this project are referred to as GeoWAsP. As part of the project, the Coastal Discontinuity Model (CDM) was also developed; it uses air and sea temperature, together with the geostrophic wind speed to calculate the stability parameter z/L to correct the wind speed profile and determine the internal boundary layer height giving a prediction of the wind speed at each grid location (Watson et al. 2000).

     
  4. (iv)

    Efficient Development of Offshore Wind Farms (ENDOW) (Barthelmie et al. 2004) included a component of mesoscale modelling of wind resources in the Baltic Sea and a comparison with predictions from WAsP based on the same dataset as the model (Bergstroem 2001).

     

We detail the methods used in these and other projects in the following sections.

2.2 Modelling

Wind resources at any location vary on a range of time scales, and hence any resource assessment should address issues of climate variability and change. However, due to scarcity of complete datasets offshore, comparisons have generally been performed with the hypothesis that local wind regimes have not changed during the last 10–20 years. The validity of this assumption is likely to be regionally variable and is increasingly being questioned (Pryor et al. 2005b). Even in the absence of climate non-stationarity, purpose built measurement sites typically have data periods of 1–3 years and hence are not representative of wind climates over the 20–30 year lifetime of the wind farms. A further confounding influence is that homogeneous wind speed time series are rarely available for long periods because many monitoring locations have undergone change in land use and instrumentation (Pryor et al. 2008). A number of techniques have been developed to provide a stable assessment of historical resource and possible temporal trends.

2.2.1 Analysis and Re-Analysis Data Sets

Analysis datasets are derived by assimilating (integrating) measurements taken in a network of measurement sites all over the world including meteorological and ship observations and satellite-derived wind speeds for offshore areas into a forecast state-of- the-art model; the result is then used to produce a physically consistent (frequently global) gridded 3-D data set.

Re-analysis datasets are produced by applying a state-of-the-art model to historical datasets providing a long-term homogeneous time series. Multiple reanalysis data sets have been developed—the two most widely used to date are:
  • The NCEP-NCAR (National Centers for Environmental Prediction-National Center of Atmospheric Research) dataset from 1948 to date (Kalnay et al. 1996) with a spatial resolution of 1.875° by 1.875° and

  • The European Centre for Medium-range Weather Forecasts (ECMWF) ERA-15 (1979–1994) (Gibson et al. (1997) and ERA-40 (1957–2001) datasets (Uppala et al. 2005) with a spatial resolution of 2.5° by 2.5°.

More recently higher resolution reanalysis datasets have been developed (e.g. the North American Regional Reanalysis (Mesinger et al. 2006), which has a spatial resolution of approximately 0.3° by 0.3°, and the Japanese 25-year Reanalysis (JRA-25) (Onogi et al. 2005) which has global coverage and a resolution of approximately 1° by 1°. While the resolution of reanalysis data sets is (currently) too coarse for direct applications in wind energy resource estimation, reanalysis datasets have been extensively used to:
  • Provide lateral boundary conditions for mesoscale and regional climate modelling (Sect. 2.2.2);

  • Wind mapping; and

  • Quantify temporal trends in near-surface wind-speeds to evaluate short- term wind speed evaluation in a climatological context, in research and commercial projects.

With respect to the last item, analyses of the NCEP-NCAR reanalysis data set over the Baltic resolved a trend towards higher wind speeds and energy density between 1953 and 1999 (Pryor and Barthelmie 2003), and subsequent analyses have demonstrated the difficulty in deriving a stable assessment of the climatological ‘mean’ wind resource over Europe due to recent climate evolution (Pryor et al. 2005b).

2.2.2 General Circulation Models (GCM)

There is increased interest in understanding how global climate change might influence the magnitude of wind resources (Pryor et al. 2005c). The primary tools for developing such projections are General Circulation (or Global Climate) Models (GCM) and regional climate downscaling (i.e. use of GCM outputs in higher spatial and temporal resolution models). The main difference between re-analysis data sets and results from a General Circulation Model is in their initialization procedure and the subsequent assimilation of observations. In the re-analysis data sets the output is forced (nudged) towards a substantial world-wide network of observations, while a GCM is typically initialized using observational data and then run forward in time without data assimilation. However, nudging has been implemented in GCM models (i.e. the global climate model of the Laboratory of Dynamical Meteorology of the French CNRS, LMDZOR (Coindreau et al. 2007)), allowing relaxation of some variables (temperature, wind and humidity) toward analysis fields. Although the temporal and spatial resolution of GCM output is typically too coarse for direct application, wind speeds can be downscaled using either regional climate models (Pryor et al. 2005a) or statistical approaches (Pryor et al. 2005d).

2.2.3 Mesoscale Models

Reanalysis datasets and GCM models have insufficient resolution to allow direct use in wind resource assessment (due in part both to the lower resolution of the output and also to the coarse representation of orography and the land-sea mask). However, they can be used to provide the boundary conditions to mesoscale models capable of deriving wind climatologies at high resolution at the regional scale; mesoscale models resolve local and regional circulation patterns and the atmospheric boundary layer, and can be applied over domains of several hundreds of kilometres squared covered with a grid mesh with a resolutions of a few kilometres (Bergstroem 2001; Badger et al. 2006; Lavagnini et al. 2006; Jimenez et al. 2007). In one such study, the ECMWF reanalysis data were used to provide boundary conditions for the parallel version of the hydrostatic model BOLAM (BOlogna Limited Area Model) QBOLAM (Buzzi et al. 1994) to generate wind statistics over the Mediterranean Sea over a 2-year period at 10 km grid resolution (Lavagnini et al. 2006).

In other studies, re-analysis data have been used in combination with mesoscale models i.e. the Karlsruhe Atmospheric Mesoscale Model (KAMM) (Adrian and Fiedler 1991; Badger et al. 2006), the PSU/NCAR MM5 model from the Pennsylvania State University/NCAR (Jimenez et al. 2007) and with non-hydrostatic models such as the MIUU (Meteorological Institute University of Uppsala) (Bergstroem 2001).

The main issue to date has been the computing resources required to run the models, although this is ceasing to be a major limitation due to increasing computing power and can be addressed to some degree using a compositing approach, rather than simulating every single day in a specific time window. The climatological compositing method (Frank et al. 2001a; Bergstroem 2002) usually relies on defining climatologically representative scenarios of large scale forcing (frequently using reanalysis data). These scenarios are identified based on flow direction, stability and other criteria and are used to limit the expenditure of computing resources while maintaining full representation of the large-scale atmospheric variability. The mesoscale model is applied within each scenario of the large-scale forcing and the results from each scenario can then be composited (based on the frequency with which each scenario is observed) to generate average conditions at the annual or longer timescales.

Early examples of such model studies were made using KAMM. There were some issues reconciling the model results with observations and other models, possibly due to the model resolution. This was addressed by combining KAMM with WAsP (Frank et al. 2001a). In recent years, more mapping studies have been based on mesoscale models, which capture the physics of the atmosphere rather than linearised models like WAsP. For the Baltic Sea a comparison between WAsP and a mesoscale model showed good agreement for prediction of wind speeds at 50- m height away from the coast (±3%) but larger differences (10–20%) in coastal areas, which has been ascribed to stability variations, which are not accounted for in WAsP (Bergstroem and Barthelmie 2002).

It is difficult to evaluate the simulations of mesoscale model results with point measurements but results from satellite-derived wind speeds can also be used (Joergensen et al. 2001; Accadia et al. 2007), as discussed further in Sect. 2.3.

2.2.4 Microscale Models: WAsP and Diagnostic Models

WAsP is a linearised model, which builds regional and site-specific local wind climatologies based on in situ wind speed measurements from nearby sites where long-term data are available. WAsP estimates a regional wind climate from measurements by removing the local effects from the wind climatology available in a reference site (predictor) i.e. shelter from near-by obstacles; effects of changes in roughness; effects of the orography on scales less than 10 km; and thermally driven flow. The main advantage of using this definition is that it is virtually scale independent, i.e. small features in the landscape (hills, valleys, forests, lakes, etc.) do not affect the regional wind climate. Once the regional wind climatology is estimated, WAsP reverses the process and extrapolates down to the surface re-introducing the local effects around the location of interest (predictand). The whole process is described in Fig. 9. The primary advantages of the WAsP methodology are the relatively straightforward model inputs required and the computational speed. It is, of course, highly sensitive to the quality, quantity and accuracy of input data (Jimenez et al. 2007). A requirement for using WAsP is that the predictor and the predictand sites lie within the same regional wind climate area (~50 km depending on the homogeneity of the surroundings). In this frame, WAsP exhibits some limitations in terms of application to offshore regions because it does not specifically model thermal winds such as the sea breeze; furthermore, the lack of specific stability climates tend to force adjustment in the coastal zone to the first 5–10 km after which the wind speed reaches an equilibrium (Barthelmie 2001b). Since the width of the coastal zone varies with large-scale (synoptic) forcing and geographic location, it was inferred that WAsP predicts too rapid recovery of the wind speeds to the offshore value and hence overestimates winds very close to the coastline (Barthelmie 2001b) by a few percent in the first 2–3 km moving offshore from the coast.
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig9_HTML.gif
Fig. 9

Schematic of the use of WAsP in micro-siting

Another approach, used by Cassola et al. (2006) to compile an Italian Offshore Wind Atlas (IOWA), is to consider a diagnostic 3D mass-consistent model i.e. the WINDS code (Ratto et al. 1994). They produced maps of the wind fields above the sea at different heights in the Italian waters using a procedure that combines statistical analyses of wind speed aloft with numerical modelling of wind flows. The first step of this methodology is the statistical analysis of 10 years of wind speed and direction data at 5000 m above sea level from the ECMWF re-analysis. The Italian offshore territory is then divided in a number of partially overlapping “geographic areas” of the order of 200 × 200 km with a grid step of about 1 km. The simulated wind fields are compared and corrected using 150 coastal area stations and compared with the results from Lavagnini et al. (2006). The IOWA mean wind speed values in each Italian sea are always higher (of about 0.5 m s−1) than the values found in Lavagnini et al. (2006).

2.3 Data

2.3.1 Measurements of Opportunity

There are relatively few in situ observations of wind speeds in offshore areas and prior to 1990, wind speeds were seldom measured offshore for wind energy applications but rather measurements were made for meteorological services (Bumke and Hasse 1989; Schmidt and Puttker 1991), on ships (Quayle 1980; Graham 1982), on oil and gas platforms (Vermeulen et al. 1985), or from buoys (Gilhousen 1987; Dorman and Winant 1995). These types of measurement, though subject to some limitations—such as record duration, inhomogenities due to instrumentation change and time of measurement, flow distortion and variable (but frequently low) observation height—have been used in various projects to map the wind resources of the European seas, frequently in combination with the WAsP program. Such projects include assessment of the wind resource in the North and Norwegian Seas based on light-house and ship observations (Boerreson 1987; Korevaar 1990).

2.3.2 Satellite Borne Remote Sensing Observations

Recently, use of spatial data derived from instrumentation deployed on orbiting satellites has been demonstrated to be of utility in assessing near surface wind speeds and hence wind resource (Christiansen et al. 2006; Hasager et al. 2007).

The data sources most thoroughly explored to date for wind energy applications are derived from polar orbiting satellites equipped with scatterometers that operate by transmitting pulses within the microwave energy spectrum range (Fig. 10). Examples include the NASA/JPL’s SeaWinds Scatterometer on QuikSCAT satellite (Suzuki et al. 2007) and synthetic aperture radars (SAR) i.e. AMIs (Active Microwave Instrument) onboard the European Space Agency (ESA) ERS 1/2, (Koch and Feser 2006) and the more recent ASAR (Advanced Synthetic Aperture Radar) onboard ESA Envisat (ENVIronmental SATellite). The most important advantage of such data sets is the spatial coverage of information (Hasager et al. 2004); however, limitations, in terms of the accuracy, data truncation and availability of sufficient images for characterising wind resources, currently restrict the application of this data source to the feasibility stage (i) of the wind assessment process (Barthelmie and Pryor 2003; Pryor et al. 2004).
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig10_HTML.gif
Fig. 10

Wind scatterometer geometry. Picture: ©European Space Agency (ESA)

Use of scatterometers in near-surface wind speed retrievals is based on quantifying the reflection (backscatter) of a radar signal by the ocean surface. The operating microwave energy spectrum is either at ~5 GHz (C-band) or ~13 GHz (Ku-band). The European ERS-2 SAR works at C-band whereas the American/Japanese scatterometers (SASS, NSCAT, QuikSCAT and Midori-2) work at Ku-band. Microwaves have the clear advantage that they penetrate clouds and are not dependent on solar illumination of the remotely sensed objects. The backscatter increases with the presence of capillary waves which in turn are a product of the wind speed above the sea surface (the surface wind stress) (Lehner et al. 1998). Backscatter data are fed into a Geophysical Model Function (GMF) that relates the signal to the wind vector through empirical coefficients found from collocated in situ wind data and backscatter values, or a combination of in situ wind data, meteorological model data and backscatter values (Stoffelen and Anderson 1997; Hersbach et al. 2007). However, the algorithms used to infer wind speeds require that an accurate wind direction field is available; such information is critical to the accuracy of the derived wind speed fields (Young et al. 2007) and can be derived from numerical weather prediction models, or the same scatterometer through analysis of wind induced streaks (surface patterns) that are visible in images at scales above 200. To extract the orientation of these streaks, well aligned with the mean surface wind direction, an algorithm based on the signal local gradients from the antenna’s system is used.

QuikSCAT (July 1999- to date) has an 1800 km wide measurement swath on the Earth’s surface resulting in twice per day coverage over a given geographic region with a descending and an ascending orbit (Fig. 11). QuikSCAT covers approximately 90-percent of Earth’s oceans every day. Wind retrievals are done on a spatial scale of 0.25° × 0.25° latitude/longitude providing wind-speed measurements of 3–20 ms−1, with an accuracy of 2 m s−1 and 20° for wind speed and direction respectively. However, root mean square differences between quality controlled research ships and QuikSCAT are approximately ±1 ms−1 in wind speed and ±15° in direction (Bourassa et al. 2003). A disadvantage is that the QuikSCAT resolution does not capture much of the wind field spatial variability in coastal regions. However, wind data from QuikSCAT can be downloaded for free in a format re-sampled from the original format “swath mode” to a geographical coordinate system “gridded mode” and wind data are available approximately twice per day (http://winds.jpl.nasa.gov/missions/quikSCAT/index.cfm).
https://static-content.springer.com/image/art%3A10.1007%2Fs10712-008-9050-2/MediaObjects/10712_2008_9050_Fig11_HTML.gif
Fig. 11

QuikSCAT wind speed map of 24th April 2008, global map, ascending passes (Top) descending passes (Bottom). Note, the wind speeds are depicted in knots (1 knot ~0.5 m s−1). Courtesy: NASA/JPL-Caltech

SAR data from ESA ERS-2 are available for 1995 to date. Assuming wind direction information is available, SAR wind speeds derived using the most completely validated GMF, CMOD4 (Stoffelen and Anderson 1997) now updated to CMOD5, can be used to map wind speeds between 2 m s−1 and 24 m s−1 with an accuracy of ±2 m s−1 (Kerbaol et al. 1998) at typical grid resolutions used for wind energy mapping (of approx. 500 m by 500 m) made as a compromise between noise reduction and high image resolution (Hasager et al. 2005; Christiansen et al. 2006).

Generally, comparison of wind speeds at a nominal height of 10-m above the sea surface from differing remote sensing technologies have shown good agreement. As just one example (Monaldo et al. 2004) found a bias of 0.35 m s−1 and a standard deviation of around 2 m s−1 in a SAR wind speed comparison in the Gulf of Alaska. The resulting measurements are capable of yielding wind speed fields with sub-kilometre resolution starting from about 3 km offshore. Major disadvantages with the use of images from SARs in wind energy resource mapping are that they are obtained three to eight times monthly and wind speed data are provided for free only in the original coordinate system “swath mode” while “grid mode” wind data have to be purchased. However, for research purposes or for special events, dataset packages in grid mode can be downloaded for free upon agreement with the ‘owner’ agency.

Various studies on wind resources assessment using SAR and scatterometer data have been conducted in various projects funded by National or International Agencies (e.g. the European Commission or European Space Agency). In the EU FP5 (5th Framework Programme) “WEMSAR” Project (Furevik et al. 2003; Hasager et al. 2005), SAR wind speed images for various atmospheric situations were retrieved at several European test sites, i.e. the west coast of Norway, the Horns Rev offshore wind farm location in Denmark, and La Maddalena Island off the northern part of Sardinia, and compared to offshore wind resources from WAsP combined with KAMM.

The purpose of WEMSAR was to provide a tool for offshore wind resource assessment. The WEMSAR tool consists of two modules: a Module 1 for SAR wind retrieval and statistical analyses was developed for wind retrieval from ERS SAR images; in the second Module 2, the output files from the wind mapping module are read into the statistical module where all satellite wind fields are treated together to provide wind climate input to the WAsP micrositing model.

At Horns Rev, comparison of wind speeds derived from SAR, QuikSCAT and in situ data from an offshore meteorological mast indicated that SAR-derived wind speeds tended to underestimate in situ observations, even when the in situ observations were used to provide the wind direction (Hasager et al. 2006). However, it was found a relatively high correlation coefficient of 0.91 between the SAR and in situ measurements with a standard error in the mean wind speed of <1.3 ms−1 and a bias of <0.5 m s−1 (Hasager et al. 2004; Christiansen et al. 2006). In the Mediterranean Sea, comparison with data from a 10 m offshore tower located close to the La Maddalena island in the narrow strait between Sardinia and Corsica, indicated that wind directions from SAR were in fairly good agreement with in situ measurements whereas the SAR wind speed values underestimate up to 40% in situ measurements in all selected cases containing data in three wind speed regimes: low (3–9 ms−1), medium (9–13 ms−1) and high (13–18 ms−1). The difference was explained by the position of the tower location: too close to the cost and in between two large islands (Joergensen et al. 2001).

In the next section, we present some results from comparisons between QuikSCAT and models in climatological terms. We again point out that the principal advantage in the use of the satellite data is in the spatial and temporal coverage that they provide over large areas. Also, the length of the time series from QuikSCAT that can be downloaded for free from NASA now exceeds 8 years allowing calculation of robust wind statistics. Mean wind speed variations on inter-annual and intra-annual time scales can be estimated and compared using the inter-annual and intra-annual indices (in which the annual mean wind speed or energy density is normalized by climatological normals) this minimising the influence of wind speed biases.

2.4 Comparison and Integration of Models and Measurements

Output from mesoscale models has been extensively compared with experimental data from buoys, islands, ships and satellite in various seas around Europe and, as expected, the difference between measured and modelled wind speeds is typically highest in coastal areas and enclosed seas where orographic channelling and flows induced by thermal effects (i.e. breeze systems) may dominate (Bergstroem 2001; Badger et al. 2006; Lavagnini et al. 2006; Kara et al. 2007). However, evaluating the performance of numerical models in the offshore environment is hampered by the relative scarcity of in situ observations and difficulties in comparing grid cells averaged conditions with point observations. For these and other reasons, mesoscale model realizations have also been compared with output from other models (or reanalysis datasets) and remotely sensed wind fields (Sempreviva et al. 2004, 2006, 2007; Cavaleri 2005; Lavagnini et al. 2006; Badger et al. 2006; Accadia et al. 2007).

The conditions over the German Bight during 2004 were simulated using the MM5 model nested within the NCEP-NCAR reanalysis dataset and compared with observations from two offshore masts, one coastal site and three stations on islands (Jimenez et al. 2007). MM5 shows promising results with approximately a 4% bias in the mean wind speed at the two offshore masts. However, the modelled vertical profile of wind speed was shown to deviate substantially from the measured mean wind speed profile on the FINO 100-m meteorological mast. Results from MM5 were also compared with those from the WAsP model, and as found in previous analyses, the largest differences between the two models were found at distances of 5 to 50 km from the coast. Since reliable offshore measurements were only available at one distance from the coast, the wind speed gradient from coastal to offshore locations could not be investigated.

Sempreviva et al. (2006) presented a qualitative comparison of wind climatology over the whole Mediterranean in terms of spatial variation of wind roses, mean wind speed, seasonal and monthly variation over a grid of 25 × 25 km resolution using LMDZOR (year 2000); analyses from ECMWF (1999–2005); the GeoWAsP model (1984–1997); six years of wind data from QuikSCAT (1999–2005); and monthly mean speed values resulting from experimental offshore sites in the Italian waters (two islands, a platform and a buoy) and five Greek buoys from the POSEIDON measuring programme (Soukissian et al. 1999, 2002). QuikSCAT data were biased since only wind speeds larger than 2 m s−1 were considered reliable and retrieved in the time series. Though the temporal window of only one year for LMDZOR was too short for a reliable wind climatology and not all model time windows were overlapping, a generally qualitative fair agreement was found. Differences in annual wind speed differences amongst models were up to 6% in open seas. Discrepancies were higher for the enclosed seas. In the Adriatic Sea the difference between LMDZOR and GeoWAsP was 36%, and in the Aegean Sea wind speeds were 32% higher from ECMWF than in LMDZOR.

In Sempreviva et al. (2007), the comparison was performed based on 7 years of QuikSCAT data (July1999–June 2005) and three years of LMDZOR data (July 1999–2002). There were two purposes for this study:
  • To produce the most reliable spatial and temporal wind climatology over the Mediterranean for evaluating wind resources at the regional scale and calculating the variability on intra- and inter-annual time scales to assess the historical context of site specific measurements; and

  • To test the capability of a GCM to model historic wind conditions

Monthly mean wind speed values were estimated at twenty six selected offshore locations using both QuikSCAT and LMDZOR datasets and the regression coefficients (R2) at each site were estimated for the overlapping time window of three years. Values of R2 range between 0.71 and 0.88 and were generally largest for open sea locations, and lower R2 values were reported for the Aegean Sea possibly due to the presence of islands, which disturb the QuikSCAT signal and which are not taken into account in LMDZOR due to the model resolution. Yearly wind indexes (i.e. the yearly mean wind speed normalised by the mean wind speed over the 7-year period) and the monthly wind indexes (i.e. monthly mean values normalised by the mean yearly wind speed estimated over the 7-year period) estimated at each site, exhibit a high degree of spatial consistency within the same geographical area.

Once more in the Mediterranean Sea, simulations over a 2-year period at a 10 km resolution were derived using the QBOLAM hydrostatic model, and compared with observations from seven islands and nine buoys (Lavagnini et al. 2006). Results indicated discrepancies of up to 50% in the mean wind speed, although this may partly reflect disparities in the temporal window used in the numerical modelling and observations, and the influence of inter-annual variability as found by Pryor et al. (2006).

In Accadia et al. (2007), good agreement was found between 10-m wind speeds from QuikSCAT over a 2-year period and the QBOLAM mesoscale model over the Mediterranean Sea although ‘in open-sea regions the model underestimates wind strength from about 0.5 m s−1 in spring and summer to 1.0 m s−1 in winter, as evidenced by the existing biases against scatterometer data’.

3 Site-Specific Prediction

Once candidate sites for wind farm development have been identified by regional scale analyses, in situ measurements at potential offshore wind farm sites will typically be conducted to provide a detailed assessment of the wind resource by potential investors. As discussed in Sect. 1.3, in situ measurements are by nature short-term (typically of the order of one year) because of the pressure to build the wind farm for financial or planning reasons once the development process has begun. This induces a need for an assessment of the past variability on intra- and inter-annual time scales, and hence an analysis of how representative the measurement year of the long-term climatologically representative wind climate. Issues pertaining to in situ measurements in an offshore environment and linking to climatologically representative wind resource assessments are described below.

3.1 Purpose Built Meteorological Masts and Ground-Based Remote Sensing

Measurement programmes have been undertaken at many prospective offshore wind farm sites using purpose built meteorological masts. Examples include those in Denmark (Barthelmie et al. 2005), Germany (Neumann et al. 2004), Sweden (Ganader et al. 2001) and the UK although the latter are not well described in the literature due to commercial confidentiality. Danish offshore wind farms had extensive research programmes, which are described in Frandsen et al. (1996), Niklasson (1997), and Barthelmie et al. (2007a). Some of the measured data at coastal and offshore sites are available in the: “Database of Wind Characteristics” (http://www.winddata.com/) which was developed under the auspices of the International Energy Agency (Larsen and Hansen 2001). Additionally, results from monitoring of the first larger offshore wind farms are becoming available (e.g. Neckelmann and Petersen 2000; Mechali et al. 2006).

Specific measurements for offshore wind farms have been mostly undertaken below a height of 100 m leading to issues of extrapolating vertical wind speed and turbulence profiles to turbine hub-heights, although recent applications of remote sensing tools such as SODARs (Barthelmie et al. 2003, 2006; Coelingh et al. 2003), and more recently LIDARs (Antoniou et al. 2006) have allowed improved characterisation above standard measurement heights. A SODAR (SOund Detection And Ranging) operates by emitting short acoustic pulses into the atmosphere, a small fraction of which is scattered back to the receivers (Barthelmie et al. 2003). The frequency of the return signal is shifted and using this Doppler Effect the wind speed can be estimated at multiple heights. The most common issues in using SODARs for measurement of wind speed profiles are attenuation of the signal in high wind speeds and rain which reduce the returns from greater heights and the need for calibration of the absolute value of the wind speed with a nearby mast e.g. at 30 m. LIDARs (LIght Detection And Ranging) operate using a similar principle to SODAR. The LIDAR has a lens to provide a scan of the atmosphere at each height from which information can be built up about the three wind components (Bingöl et al. 2008a, b; Smith et al. 2006; Sjöholm et al. 2008).

3.2 Methodology for Relating On-Site Data to Longer-Term Records

As mentioned above, in situ measurements are typically conducted over a limited temporal window and hence there is a need to place those measurements in a climatological context. In this section we present an overview of the different methods used to estimate the wind climatology offshore. Assuming in situ data are available at the potential development site over a short-time period, possible sources of longer time data for extrapolation in the temporal domain include nearby land sites and the reanalysis data sets described above. In a recent study, site-specific predictions of the mean wind speed at five offshore locations around the UK were compared. The empirical method of Hsu (1988), which assumes a fixed linear relationship between wind speeds at 10 m over land and sea, generated a mean wind speed that had a root mean square error (RMSE) relative to the observations of between 20% and 40%, while the reanalysis datasets exhibited similar or slightly higher RMSE (McQueen and Watson 2006). This level of accuracy is clearly insufficient for robust wind resource estimation and so more sophisticated tools based on extrapolation from longer term measurements are typically employed, with the caveat that in general to produce a meaningful relationship between wind speeds at two sites they must be in reasonable proximity ~50 km depending on the orography in order to invoke the assumption that they are subject to the same large scale forcing. Once this criterion is fulfilled one (or more) of the following methods may be applied:

WAsP program. To generate a representative site wind climatology, long-term meteorological time series from a nearby site can be input to the WAsP Program (Sect. 2.2.4) to predict site wind speeds. Results should be compared with the short-term on-site records and with WAsP on-site predictions for the shorter time series.

The measure-correlate-predict (MCP) method (Bunn and Watson 1996; Rogers et al. 2005). This method typically assumes a linear relationship between wind speed at paired sites where one site with a long-term record acts as predictor and the wind speed at short-term measurement sites as the predictand. Once a regression equation has been conditioned based on the measurement overlap period, the regression parameters can then be used to derive an extended data record for the site of interest. MCP is generally applied using one regression analysis for each wind direction sector. An issue using MCP based on data from land sites is that most applications use linear regression which cannot account for observed differences in the wind speed distribution between the land site and the offshore sites.

The Weibull correction method (Barthelmie et al. 2008). This approach is based on the concept of modifying the Weibull parameters of the short-term data series to characterise a longer data sampling period. It compares sector-based wind speed distributions at the on- and the off-shore sites considering the on-shore long-term time series as representative of the area.
  1. (1)

    Weibull scale (A) and shape (k) factors are determined for each of 12 wind sectors and for the mean values in each point of the grid, for data sets from overlapping periods

     
  2. (2)
    The differences between the two datasets are expressed in terms of the following ratios
    $$ \frac{{{{A}}\left( {{\text{site}}\,{\text{offshore}}} \right)}}{{{{A}}\left( {{\text{site}}\,{\text{onshore}}} \right)}}\,{\text{and}}\,\frac{{{\text{k}}\left( {{\text{site}}\,{\text{offshore}}} \right)}}{{{\text{k}}\left( {{\text{site}}\,{\text{onshore}}} \right)}}\, $$
    (4)
    called correction factors
     
  3. (3)

    These correction factors are applied to the Weibull factors estimated for the short-term data sets.

     

A comparison of WAsP calculations with the data of offshore masts at Danish wind farm locations indicated estimates of the long-term average wind resource were in good agreement with the measurements with slight over-prediction of the wind speed by WAsP for short offshore fetches (Barthelmie et al. 1999; Lange and Hoejstrup 2001). Lavagnini et al. (2003) presents similar analyses for the North Adriatic area based on seven years of hourly data collected on an oceanographic platform 15 km offshore of Venice and long-term data were available at four coastal stations. Results indicate that WAsP underestimates the wind at the platform in the sea sectors and it overestimates in the land sectors. The MCP method was not found applicable in this area since correlations amongst station time series (either by wind direction sector or in total) were not statistically significant. The Weibull correction method reproduced the frequency in all sectors except two, both in case of onshore flow where the wind speed is overestimated.

4 Concluding Remarks

Herein, we present an overview of measurements and modelling techniques that have been used to assess offshore wind resources in Europe over the last twenty years, and an assessment of their comparability and accuracy. The availability of large computing resources and satellite observations has facilitated development of greatly enhanced spatial and temporal coverage of wind speeds relative to the first wind atlas, and are now capable of providing relatively robust regional assessments, that can also be linked to ‘climatological normals’ using reanalysis and other data sets. Many sites still require the establishment of a meteorological mast to overcome difficulties of uncertainties in the methods especially relating to small-scale spatial variations in the near-coastal zone and to provide detailed wind speed and turbulence profiles at wind turbine hub-heights. New ground-based remote sensing techniques such as LIDARs and SODARs present additional, sometimes supporting, measurements for specific site studies, and are allowing researchers to develop improved understanding of the marine atmosphere above the heights typical measured by meteorological masts.

Acknowledgements

This work was partially funded by the European Commission project “Prediction Of Waves, Wakes and Offshore Wind” (POW’WOW) (019898(SES6)). Rebecca Barthelmie acknowledges support from the Scottish Funding Council and the Edinburgh Research Partnership. Sara Pryor acknowledges funding support from the National Science Foundation (grant # 0618364). The authors wish to thank ESA—European Space Agency—for supplying Fig. 10.

Copyright information

© Springer Science+Business Media B.V. 2009