The Global Wind Atlas uses a downscaling process. A schematic of the methodology and verification methods is shown in Fig. 1↓
Figure 1 Schematic diagram showing the downscaling process, which is key to the Global Wind Atlas methodology. The verification methods are also shown in the same diagram.
We start with large scale wind climate data and end with micro scale wind climate data. The large scale wind climate data is provided by atmospheric reanalysis data, from meteorological centres around the world. These data are on a grid with spacing of about 50 km depending on the dataset. We perform a generalization process on this data. The result is a set of generalized wind climates which have the same spacing as the reanalysis data that was used to create them.
Next we take this set of generalized wind climates and apply them in microscale modelling system over the globe (apart from poles and far offshore oceans areas). The modelling process is made up of a calculation for the local wind climates every 250 m at three heights, 50, 100 and 200 m. So on a 250 m grid, there is a local wind climate estimation. Local wind climate characteristics are aggregated up to a 1 km grid. Datasets and tools for analysing statistics based on the 250 m grid values are available on the Global Wind Atlas website.
Figure 2 Wind power density calculated at 50 m above the surface for a 50 km x 50 km area at two different resolutions, (left) 2.5 km and (right) 100 m. The colour scale used in each map is the same. Dark blue is low values and bright green and yellow are high values. The mean power density for the area is given in black. The mean power density for the windiest 50% area is given in red.
Figure 2↑ shows the wind power density at 50 m for a 50 km x 50 km area modelled at two different resolutions, namely 2.5 km and 100 m. As the resolution increases features in the terrain become better resolved. Resolved hills and ridges give rise to increased wind speeds. As wind power density is a function of wind speed cubed, the impact of the resolved terrain features is significant.
For the 50 km x 50 km area the area mean wind power density is estimated to be around 320 W m-2 by the modelling at a resolution of 2.5 km. For the 100 m resolution modelling the mean power density is around 505 W m-2, i.e. an increase of 50% compared to the lower resolution estimates.
The comparison becomes more striking when the distribution of the wind power density is considered. Consider this: we split the 50 km x 50 km into two areas; the first area where the wind power density is below the median value and the second area where the wind power density is above the median value. Next we calculate the mean wind power density in the second higher wind area; the 2.5 km resolution modelling gives 380 W m-2, whereas the 100 m resolution modelling gives 640 W m-2, an increase of nearly 70%.
The impact gets stronger as we look at the even windier areas. As wind turbines will be deployed at the favourable sites, it is important to be able to capture the distribution of wind power density due to terrain features, and this is only possible by consideration of high resolution effects. As one concentrates on the most favourable areas, the wind power density goes up.
Figure 3 Graph showing the mean wind power density (y-axis) for the windiest fractile area (x-axis) of the 50 km x 50 km area in Fig. 2↑
, calculated from modelling using different resolutions, namely, 10, 5, 2.5, and 0.1 km. A guiding line is drawn at fractile 0.5, which gives power density values corresponding to those in Fig. 2↑
. Note that when modelled at high resolution we see that the windiest areas are significantly windier than the average for the whole area.
Figure 3↑ shows the calculated mean power density for the 50 x 50 km area when averaging over different amounts of the windiest areas for different modelling resolutions. It goes up most for the high resolution modelling. Even for rather simple terrain, such as in Denmark, the effect of resolution is important. A similar 50 km x 50 km area showed an increase of wind power density of around 25% for the windiest 5% of the area (i.e. the windiest 1/20th of the area).
The concept of a generalized wind climate is a key element of the wind atlas methodology developed at DTU Wind Energy. The European Wind Atlas  sets out the method fully. Since then, the generalization has been used in numerical wind atlas methodologies, where mesoscale modelling output is generalized before application in microscale modelling with WAsP. In this chapter we describe the generalization fundamentals and how the method used in the mesoscale modelling has been adapted to the reanalysis data.
The concept of generalization
The description of the topography and the land surface in the global models and in nature is very distinct. The two panels in Fig. 4.1↓ illustrate some of these differences. The left figure illustrates the differences in topography between reality (photo) and the reanalysis model (yellow line). The topography induced speed-up in reality and in the microscale model are thus quite different from that in the reanalysis model.
The details of the position of the coastline are misrepresented by the coarse reanalysis grid. The right figure illustrates the differences in the location of the coastline in reality (Google Earth map) and that in the reanalysis model (green and blue squares).
To avoid double counting these effects when coupling the mesoscale or global model results to the microscale model, effects in a similar scale to that simulated in the microscale model must be removed. We call this process “generalization” within the WAsP-thinking and wind atlas method developed at DTU Wind Energy.
Illustration of the conflict between representation of terrain (left) and land/ocean mask (right) in a global reanalysis and in nature.
The wind atlas method is based on the generalization of the wind climatologies derived from mesoscale or reanalysis global modeling. This generalization post-processing method has been used extensively in a number of wind resource assessment studies, particularly within the KAMM-WAsP method . The method was used for the first time with WRF model simulations in the Wind Atlas for South Africa (WASA) project .
The starting place for applying the generalization procedure to the reanalysis data is the reanalysis TAB netcdf files described in Datasets. These are identical file type for each reanalyses but with different horizontal grid dimensions. These files contains the statistical information required for further downscaling.
Four main parameters can be derived from the surface characteristics of the reanalysis model grid (e.g., the terrain and surface roughness length) as described in :
- a factor that accounts for how the model description of topography impacts the local flow (δAo),
- a parameter that takes into account how the topography alters the wind direction (δφo),
- a parameter that accounts for the local flow perturbation on wind speed due to roughness length variations (δAr), and finally a
- grid point dependent upstream roughness length (ẑ0).
These four parameters are computed from the reanalysis grid description (Table 1), that includes the topographic height and a time averaged surface roughness length. These parameters are stored in a NetCDF file and used in the generalization of the reanalysis TAB netcdf file (see Eqs. 1↓, 2↓, 3↓, and 4↓).
A similar code to that used with the KAMM/WAsP and WRF mesoscale models is used to derive the generalization parameters. A few modifications were introduced to account for the specifics of the reanalysis grids:
- The LINCOM flow model  used in the calculation of the local flow perturbation assumes that the horizontal grid is equally spaced in distance in the x × y directions. The reanalysis grid is equally spaced in latitude × longitude, and in the MERRA data contains different spacing in latitude and longitude. To overcome this issue, the globe was divided into smaller patches that overlap each other and where a constant ΔxorΔy assumption is not too far off. The size of the patches in terms of degrees latitude × longitude was 6○ × 18○ in the CFDDA, 8○ × 18○ in the CFSR, and 8○ × 18○ in the MERRA. On each of these patches we assumed dx = dy = 2πR(dlon ⁄ 360)cos(clat), where R is the radius of the Earth, dlon is the longitude spacing, and clat is the latitude of the center of the patch. No significant differences are found in these parameters along the edge of each patch.
- In the direction-dependent calculation of the upstream roughness, a correction is made according to the actual distance (in km) between neighboring grid points which varies with latitude according to the angle, i.e. ds = dx|cosφ| + (1 − |cosφ|)dy, where φ is the direction of the sector, and dx and dy are the actual dimensions of the grid in kilometers.
Figure 5↓ displays the geographic distribution of two of these generalization parameters for the MERRA reanalysis for a patch centered over Denmark. Along the southern coast of Norway and Sweden, large changes in surface roughness length (Fig. 5↓a) are visible (from water to forest landcover). δAr values are as large as 1.2. In opposition, the smooth topography in the MERRA reanalysis (Fig. 5↓b) results in quite small values of the topographic correction factor, e.g. |δAo| < 1.0065 (Fig. 5↓d).
Surface fields: (top-left) Surface roughness length (m), (top-right) surface elevation (m), and generalization parameters: (bottom-left) DAROU (δAr) and (bottom-right) DAORO (δAo), for the 180○ sector and 100 m height for the MERRA reanalysis.
Basic generalization equations
We describe here the generalization procedure. In the first step, each center value of wind speed, u = u(z, φ), and wind direction ,φ, is corrected for orography and roughness change, which are a function of wind direction and height. The intermediate values, û and φ̂, are given by
(1 + δAo)(1 + δAr)
= φ − δφo,
where δAo, δφo and δAr and are generalization factors for orography in wind speed and direction and roughness change, respectively, described in the generalization factors section.
From the corrected wind speed value we obtain an intermediary friction velocity, û*
= (κû)/(ln(z ⁄ ẑ0))
where z is the height, ẑ0 is the downstream surface roughness length for that sector and κ is the von Kármán constant. The stability of the surface layer is not taken into account in these calculations because it is not easily available from the reanalysis data and at the dimensions of a global model square does not have the same sense as for a local measurement.
In the next step, we use the geostrophic drag law, which is used for neutral conditions, to determine nominal geostrophic wind speeds, Ĝ, and wind directions, δφG, using the intermediate friction velocity and wind direction:
= (û*)/(κ)√((ln(û*)/(|f| ẑ0) − A)2 + B2),
= − sin − 1⎛⎝(B û*)/(κĜ)⎞⎠,
where A = 1.8 and B = 5.4 are two empirical parameters and f is the Coriolis parameter, and φ̂G is the angle between the near-surface winds and the geostrophic wind.
To obtain a new generalized friction velocity, û*G, for a standard roughness length z0, std, Eq. 4↑ is reversed by an iterative method,
= (û*G)/(κ)√((ln(û*G)/(f z0, std) − A)2 + B2)
Finally, the generalized wind speed, uG
, is obtained by using the logarithmic wind profile law
= (û*G)/(κ) ln⎛⎝(z)/(z0, std)⎞⎠
It is assumed that the initial sector of the direction bin remains unchanged, because each sector is 30○ wide and in only very rare instances δφG is larger than ±15○. This might not the case if more direction sectors are considered and the topography has larger horizontal gradients.
Once a generalized wind speed, uG, has been found for each direction sector and standard roughness and wind speed bin a Weibull fit is done for each sector and standard roughness.
Weibull distribution fit
In this section the way the generalized wind speed data is converted into a compact file containing the generalized wind climatology is described. The resulting file is called a WAsP lib-file.
The frequency distribution of the horizontal wind speed can often be reasonably well described by the Weibull distribution function :
) = (kw)/(Aw)⎛⎝(u)/(Aw)⎞⎠kw − 1
where F(u) is the frequency of occurrence of the wind speed u. In the Weibull distribution the scale parameter Aw has wind speed units and is proportional to the average wind speed calculated from the entire distribution. The shape parameter kw( ≥ 1) describes the skewness of the distribution function. For typical wind speed distributions, the kw-parameter has values in the range of 2 to 3.
From the values of Aw and kw, the mean wind speed U ( m s-1) and mean power density E (W m-2) in the wind can be calculated from:
AwΓ⎛⎝1 + (1)/(kw)⎞⎠
(1)/(2)ρA3w⋅Γ⎛⎝1 + (3)/(kw)⎞⎠
is the mean density of the air and Γ
is the gamma function. We use the moment fitting method as used in the Wind Atlas Analysis and Application Program (WAsP) for estimating the Weibull parameters. The method is described in detail in 
. Basically this method estimates Aw
to fit the power density in the time series instead of the mean wind speed.
The Weibull fit is done for the ensemble of generalized wind speeds in each wind direction bin (usually 12 direction sectors), each height (50, 100 and 200 m) and each standard roughness lengths (usually 5 roughness: 0.0002 (water), 0.03, 0.1, 0.4, 1.5 m).
This sector-wise transformation of Weibull wind statistics---i.e. transforming the Weibull Aw
parameters to a number of reference heights over flat land having given reference roughnesses---uses not only the geostrophic drag law, but also a perturbation of the drag law, with the latter part including a climatological stability treatment. The transformation and stability calculation is consistent with that implemented in WAsP and outlined in 
, with further details given in 
. The transformation is accomplished via perturbation of both the mean wind and expected long-term variance of wind speed, such that both Weibull-Aw
are affected. When purely neutral conditions (zero stability effects) are presumed for the wind statistics to be transformed, there is still a perturbation introduced, associated with the generalized (reference) conditions in the wind atlas. This perturbation uses the default stability parameter values found in WAsP; it is negated upon subsequent application of the generalized wind from a given reference height and roughness to a site with identical height and surface roughness, using WAsP with its default settings. The climatological stability treatment in the generalization depends on the unperturbed Weibull parameters and effective surface roughness 
, as well as the mesoscale output heights and wind atlas reference heights (though the latter disappears upon application of wind atlas data via WAsP).
In practice, a choice has to be made when regarding whether a correction for the stability has been done or not in the calculation of Aw and kw for each sector. We set ain_neut = True in the generalization options (i.e. the data has been transformed to neutral stratification) and no correction owing to a mean heat flux is done. The remaining stability-induced correction are done using heat fluxes of -40 W m-2 over land and 15 W m-2 over water.
In this chapter the generalization proceedure has been described. In particular the way the method has been adapted for the generalization of reanalysis data is highlighted. Also outlined is the way the generalized wind climatology is processed and formatted to be made ready for application in the microscale modelling.
Contributing authors Andrea N. Hahmann, Jake Badger
Calculation of local wind climates
In this chapter the microscale modelling is described. It is here that the downscaling from the generalized wind climates to the local climates is performed every 250 m. This very large area calculation is run by a system of software and servers called Global Wind Atlas Frogfoot. The method is very similar to that used in the WAsP software. For example, the flow modelling for orography, roughness and roughness change is the same as in the WAsP software. However, the Global Wind Atlas calculation differs in a number of ways in order to allow a very large area to be covered. For example, local wind climate calculations are based on more than a single generalized wind climate, and terrain data is input as raster maps rather than vector maps.
WAsP microscale effects
The WAsP software contains flow models for orography, roughness and roughness change effects, and obstacles. In the Global Wind Atlas obstacles are not included.
Schematic diagrams Fig. 5.1↓
illustrate the change of wind flow caused by a hill. The maximum speed-up is at the top of the hill, the magnitude of the speed-up and the height above surface of the maximum speed-up is related to the geometry of the hill. WAsP uses the BZ-model 
to calculate the orographic speed up. The flow model uses a high-resolution, zooming, polar grid, centered on the calculation node. More details can be found in 
(a) Streamlines of winds flowing over a hill. The more closely spaced streamlines at the top of the hill are associated with a speed up of the winds. (b) The vertical profile of wind speed upwind and on top of a hill, from 
. The speed-up is a function of height above the surface. The height of maximum speed up (l
) is related to the geometry of the hill.
Surface roughness length, z0, is a property of the surface which can be used to determine the way the horizontal wind speed varies with height, assuming a homogeous surface and neutral stability, according to
) = (u*)/(κ)
is the height above surface and κ
is the von Kármán constant. Figure 5.2↓
a shows the influence roughness has on the vertical profile of wind speed assuming the same large-scale wind forcing. The wind speed at a given height decreases with increasing surface roughness. The equation above and the figure assume a homogeneous surface over several kilometers. However it is very common to have a heterogeneous surface and this complicates the vertical profile somewhat. Internal boundary layers develop and the profile of wind speed departs from the logarithmic wind profile.
b shows an example of the vertical profile of wind speed 4 km downwind of a surface roughness change from 0.02 cm to 20 cm. The wind speed profile does not change at all heights immediately downwind of the roughness change. At first, only the lowest parts of the profile change, with the change progressively reaching higher and higher with increasing downwind distance from the roughness change. The impact of a roughness change can be felt many kilometers downwind. As a rule of thumb, at 100 m above the terrain a surface roughness change 10 km upwind may still have an influence on wind speed.
The WAsP roughness change model can account for these internal boundary layer effects due to inhomogeneous surface roughness. More details can be found in 
(a) Different surface roughness lengths result in different vertical profiles of wind speed. The y-axis is height above surface and the x-axis is wind speed. For a range of surface roughness lengths, the curves show the wind speed profile for neutral conditions and a geostrophic wind of 10 m s-1. (b) This graph illustrates how a roughness length change (at x = 0) impacts the downwind profile of wind speed. For x < 0 the roughness length is 0.02 cm, for x > 0 the roughness length is 20 cm. The two red lines show the heights of the internal boundary layers. The upper red line shows the height below which the wind profile departs from the 0.02 cm roughness profile. The lower red line shows the height where the profile is adjusted to the 20 cm roughness profile.
The high-resolution resource calculation system
The calculation system used for the Global Wind Atlas is called Frogfoot. It has been developed in association with the software development company World In A Box
http://www.worldinabox.eu/index.html . The motivation for the development of Frogfoot was to allow high resolution WAsP-like calculations of predicted wind climates to be made over large areas, using a large number of generalized wind climates. This need came about because of numerical wind atlases being carried out on nation-wide scale which generated generalized wind climates on a grid with a spacing of, typically, 5 km.
As stated before, the Frogfoot system employes the same flow modelling as WAsP. Unlike the present WAsP, the terrain description can be input using raster maps, rather than vector maps. This is convenient for the Global Wind Atlas calculation because typically the global topographical data is in raster formats. Unlike the present WAsP, the starting point for describing the large scale wind forcing is any number of geographically distributed generalized wind climate files (lib-files), whereas WAsP can only use one.
Frogfoot is a system of programs and interlinked servers developed and set up to allow for very large geographical coverage of high-resolution wind resource maps, with inclusion of changing large-scale wind forcing and microscale flow effects.
The core components of the Frogfoot system are:
Ancillary components are:
Job Management Console
Climate Data Manager
The core components are essential to carry out a Frogfoot calculation. The ancillary components are needed to set-up the configuration of a Frogfoot job, as well as import and export data into or out of the system.
The framework of Frogfoot can be understood by considering the elements required to carry out a WAsP calculation of a predicted wind climate at a single location. These elements are generalized wind climate data, roughness and orography data (in the form of maps) for the area around the location of interest, and flow models inside the WAsP software. The roughness and orography data is used by the flow models to determine flow effects at the location, and these flow effects are used to modify the generalized wind climate. These elements are also represented in the Frogfoot core components, see Fig. 5.3↓
. For Frogfoot though, instead of considering a single point, the Job service dispatches a very large number of points within an area of interest to be calculated.
For any particular application of Frogfoot, the Job Service is set up by the Job Management Console. Here the user specifies the map data to be used, selected from map data inside the Terrain Service. Here the user also specifies the generalized wind climate data to be used, selected from generalized wind climate data inside the Climate service. The generalized wind climate data consists of a number of geo-referenced WAsP generalized wind climate files.
Within the Job Management Console the definition of the area to be calculated is specified, by a map containing a single closed contour outlining the boundary of the calculation area. The user also needs to set the grid spacing and origin of the calculation nodes. The Job Management Console will then split the job into a number of tiles which are 10 x 10 calculation nodes in size (i.e. 100 calculation nodes in all). A tile makes up a set of calculations that will be dealt with separately by distribution of the tile to a WAsP Worker. The WAsP Worker is a standalone installation of the WAsP flow models, without the user interface.
In order for the WAsP Worker to calculate the predicted wind climate at the 100 nodes, tile maps of roughness and orography are prepared by the Terrain Service. The maps are given an extent sufficient for the tile by extending the map boundary with a 25 km buffer around the extent of the tile.
The calculation also needs generalized wind climate data. This is provided by the Climate Service in the form of 100 lib-files (one for each calculation node). For each calculation node a libfile is calculated by the Climate Service based on an interpolation of the 3 nearest libfiles from the selected generalized wind climate dataset. The interpolation weighting of the libfiles is inversely proportional to distance. For each direction sector the wind speed distribution is calculated, based on the weighted combination of the Weibull distributions for the 3 nearest libfiles, then with this a new Weibull fit is performed to provide the Weibull parameters for the interpolated libfile.
shows how the different components are related to each other and the flow of data from one component to the next. Screen shots of the components are shown in Fig. 5.4↓
. The Job Service feeds tile data to the WAsP workers. The WAsP workers are installed on computers within the same local area network as the Frogfoot servers. Once the WAsP worker has finished the calculations for one tile, the tile results are sent to the Results Service. The Job Management Console allows users to get an overview of the current jobs running on the system, as well as jobs that have been completed or paused. Once the Frogfoot job has completed, the Result Exporter is used by the user to export output in the desired format for subsequent analysis or plotting.
Schematic diagram showing the relationship and flow of data between the components of the Frogfoot system. Blue boxes represent core components of Frogfoot, red boxes represent ancillary components, purple boxes represent data that is input into the system, and the green box represents the result outputs.
Setting up the global calculation
In order to carry out the Global Wind Atlas calculaton a convenient way of breaking the globe into managable size blocks was required. The Military Grid Reference System (MGRS) provided the basis of the job structure, see Fig. 5.5↓
Still the MGRS zones were too big, so these were divided into 4 pieces, to make what is called a job tile. The result is a total of 4903 job tiles. However, in order to cover land and 30 km offshore, 2439 job tile are needed. In Fig. 5.6↓
the Global Wind Atlas Frogfoot job tile layout is show in Google Earth.
Google Earth image showing the breakdown of the globe into Global Wind Atlas job tiles (red rectangles). Two job tiles are highlighted in South Africa, with the smaller WAsP Worker calculation tiles shown in yellow. Calculations are performed on or within 30 km of land.
For the 2439 job tiles the necessary Frogfoot input data are prepared. This entails preparation of 2439 orography and surface roughness maps on the calculation projection system (Universal Tranverse Mercator), and preparation of calculation inclusion maps, in order to limit the calculation to land and 30 km offshore.
In this chapter the microscale modelling system has been described. Its similarities to the WAsP software in terms of the flow modelling are highlighed. Its differences to the WAsP software, in terms of ability to cover large areas, use multiple lib-files, use raster topography maps, are outlined. The Frogfoot calculation system and the job tile configuration for the Global Wind Atlas are described.
Contributing authors Jake Badger, Niels G. Mortensen
In this chapter verification of the elements of the Global Wind Atlas are described. The large spatial coverage of the Atlas means that verification is a challenge as the number of data points in relation to the number of measurement points is very large. We address this issue by considering verification in number of ways.
We can verify Global Wind Atlas winds against remote sensing data from Synthetic Aperture Radar (SAR). This method gives maps of wind climate data over large areas, in a way an output similar to the Global Wind Atlas. The limitations of this method include that only offshore areas can be mapped, and the extrapolation of wind speeds to Global Wind Atlas heights introduces uncertainty.
We can verify Global Wind Atlas winds against numerical wind atlas results from other projects. This has the advantage that the verification can be done over land. The limitation of this method is that a comparison is being made against results of modelling, so it is not a comparison against measurements.
For locations where there are measurements, verification is possible at the location, but the challenge then is to infer what is gained from that verification to all other locations. We can also verify Global Wind Atlas winds against high resolution resource maps generated from measurement based generalized winds.
This chapter attempts to give a qualitive and quantitative assessment for the verification of the Global Wind Atlas. It is important to note that the verfication is not stating whether one reanalysis dataset is better than another. It is instead a comparative assessment of the result of the generalization of the reanalyses and the high resolution modelling. So any poorer performance must be seen in this large context. Improvement in the generalization methodology, or changes to the surface topographical description would alter the results.
Synthetic aperture radar ocean derived winds
Synthetic Aperture Radar (SAR) data were used to verify Global Wind Atlas outputs over coastal seas where DTU Wind Energy has collected SAR scenes through previous projects. Additional scenes were downloaded and processed for the seas around South Africa.
The areas used for verification of offshore winds are:
Aegean Sea  
The Great Lakes 
Southern China  
This section describes the SAR data and the applied procedure for retrieving winds and wind resource maps from SAR at the standard level of 10 m above sea level.
Satellite SAR data
SAR data were obtained from the European Space Agency (ESA). The gateway to accessing ESA data is http://earth.esa.int/. Approximately 15,000 scenes from the Envisat ASAR mission in 2002-11 were selected from DTU’s archive to be used for the Global Wind Atlas verification. All the SAR scenes were acquired at C-band with either vertical or horizontal polarization in transmit and receive (i.e. VV or HH polarization).
Figure 6.1 Examples of the Envisat ASAR coverage over South Africa for four individual scenes. Image courtesy Google Earth.
Envisat was a polar-orbiting satellite scanning the Earth surface in a 400-km swath. A scene represents a fraction of the swath and can be several hundreds of km long but with a fixed width of 400 km (Fig. 6.1↑
). The SAR principle of operation is that radar pulses are transmitted towards the Earth surface. The proportion of backscattered signal determines the image brightness and it depends on the surface properties as well as the radar viewing geometry. Over ocean surfaces, the return signal is mainly determined by Bragg scattering 
from capillary and short-gravity waves, which have wavelengths proportional to the radar wavelength. These waves are wind-generated and there is thus a relationship between the instant wind speed and the radar backscatter. In order to eliminate effects of longer-period waves and random noise in the SAR images, pixels are averaged to 500 m or larger cells before the wind retrieval.
SAR wind retrieval
Geophysical Model Functions (GMFs) are empirical equations for the backscatter-to-wind relationship, which are originally developed for scatterometers 
. If the wind direction is known, these equations can be used to retrieve the wind speed from SAR observations 
. Here we used the GMF called CMOD5.n 
to retrieve Equivalent Neutral Winds (ENW) at 10 m. The winds represent neutral atmospheric conditions. The wind retrieval processing was performed with the APL/NOAA SAR Wind Retrieval Software (ANSWRS) with input wind directions from the US Navy Operational Global Atmospheric System (NOGAPS). The model wind directions were available at 6-hourly intervals with 1 ○
latitude and longitude resolution and they were interpolated spatially to match the SAR data. Figure 6.2↓
shows examples of retrieved SAR wind maps over South Africa.
For areas poleward of 50○ latitude a sea ice mask was applied to filter out ice covered seas where the SAR wind retrieval could not be trusted. Ice mask data from the IMS Daily Northern Hemisphere Snow and Ice Analysis at 4 km Resolution by the US National Ice Center (http://nsidc.org/data/docs/noaa/g02156_ims_snow_ice_analysis/) were used to eliminate areas with ice cover.
Figure 6.2 Two examples of wind fields retrieved from Envisat ASAR over South Africa. Left: morning pass acquired May 9, 2011 at 07:47 UTC; right: evening pass acquired June 27, 2010 at 20:59 UTC. Wind barbs indicate the model wind vectors from NOGAPS.
SAR Wind resource mapping
For each verification area, all available SAR wind maps were combined in a statistical analysis in order to map the offshore wind resources. The data was first insert into a MySQL data base and organized into 1-km grid cells. Once the data is stored in the data base it can easily be retrieved based on criteria for location and time coverage. The tool S-WAsP by DTU handles the insert procedure, the data retrieval, as well as computation of wind resource statistics. The entire processing chain needed to transform raw SAR observations to wind resource maps is illustrated in Figure 3.
Figure 6.3 The processing chain for SAR wind retrieval and resource mapping.
Over each focus area, a Weibull function was fitted to the data points available for each cell in a 0.02 degree latitude and longitude grid. This gave maps of the mean wind speed, Weibull shape and scale parameters, and wind power densities as well as uncertainty estimates for the 10-m level.
SAR case study results
The SAR winds are given at 10 m above the sea surface. The winds are extrapolated to 100 m above the surface by using the logarithmic law and a surface roughness of 0.2 mm. This assumes a neutral profile and a single water roughness. Both assumptions are likely to cause uncertainty in the estimation of the SAR 100 m winds. An improved method, that corrects for the effects of a distribution of boundary layer stability is the subject of ongoing research. A scientific paper which describes the method is in review.
shows the MERRA reanalysis wind interpolated to 100 m above surface, the Global Wind Atlas winds at 100 m, and the SAR winds at 100 m. The MERRA winds tend to be comparatively lower in the offshore coast regions, whereas the Global Wind Atlas winds are higher close to the coastline. The SAR results suggests slighly lower winds in the Danish inners seas, and a more gradual change in wind speeds off coastlines compared to the Global Wind Atlas.
(b) GWA, background MERRA
(c) SAR, background MERRA
Figure 6.4 Selected area within Scandinavian seas showing 100 m winds [ m s-1] from MERRA, from the Global Wind Atlas and from SAR. (a) MERRA 100 m winds, (b) Global Wind Atlas 100 m winds, and (c) SAR 100 m wind. In (b) and (c) MERRA 100 m winds are given in the background to allow comparison at the edge of the Global Wind Atlas and SAR winds.
shows the MERRA winds tend to be comparativily lower in the region, whereas the Global Wind Atlas winds and even more so the SAR winds are indicating higher winds. The SAR results show very noticable enhanced winds in the gaps between islands, and some headland areas on Turkey west coasts. The Global Wind Atlas is missing these features, having a more homogeneous wind speed. The reason these features are missing is that the reanalysis model is too coarse to capture the terrain of the islands and headland features that cause the enhanced wind speed. This aspect of missing mesoscale variability is something that occurs in other examples in this chapter.
(b) GWA, background MERRA
(c) SAR, background MERRA
Figure 6.5 Aegean Sea, otherwise the same as Fig. 6.4↑
The Great Lakes
shows the MERRA and Global Wind Atlas winds tend to be weaker than the SAR winds are indicating. Near the shoreline the disagreement between the Global Wind Atlas and SAR winds is lesser, but in the interior of the Great Lakes, the difference is large. The reason for these differences needs to be investigated further. Later in Fig. 6.22↓
it can be seen that none of the reanalysis data suggest such high winds over the Great Lakes. This may point to the issue being how the 100 m SAR winds are extrapolated, in particular stability and surface roughness may diverge from the used assumptions. Also the sampling of the SAR scenes is the area could favour seasons with higher winds (winter). In 
a method which corrects for any season bias in the same collected images, shows annual mean winds more in line with the reanalysis and Global Wind Atlas winds.
(b) GWA, background MERRA
(c) SAR, background MERRA
Figure 6.6 Great Lakes, otherwise the same ass Fig 6.4↑
Southern China Sea
shows MERRA winds tend to be comparativily lower in the offshore coast regions, whereas the Global Wind Atlas winds are higher close to the coastline. The SAR results suggests slighly higher winds. The South China Sea is in the tropics and has with a high sea surface temperature. This induces unstable atmospheric conditions and deviations from the neutral wind profile assumption.
(b) GWA, background MERRA
(c) SAR, background MERRA
Figure 6.7 Southern China Sea, otherwise the same ass Fig 6.4↑
South Africa, Western Cape coast
shows MERRA winds tend to be comparativily lower, whereas the Global Wind Atlas winds are higher close to the coastline. The SAR results suggests much higher winds off the western most part of the Western Cape, but similar winds close to the coastline elsewhere. The high winds indicated by SAR off the western most part of the Western Cape could be due to mesoscale spatial variability and the influence of the cold Benguela Current. Another perspective is that this cold current creates a large departure from the assumption of neutral stability used for the extrapolation of SAR winds to 100m, there by adding a bias to the SAR winds estimate.
(b) GWA, background MERRA
(c) SAR, background MERRA
Figure 6.8 South Africa, Western Cape coast, otherwise the same ass Fig 6.4↑
In this section, high resolution results from the Frogfoot system are compared for a number of different regions. The wind speed and power density are compared in two ways. The first is to compare exact grid point location results with each other. The second is to compare the ranked results. This method puts less emphasis on where precisely the wind resource is, but tests how the distribution of wind resource is captured in the test area.
Figure 6.11 Alaiz region case (30T-1) 100 m wind speed [ m s-1] (upper) and wind power density [W m-2] (lower) scatter plots comparing Frogfoot results from different reanalyses. (a) and (b) MERRA v CFDDA, (c) and (d) MERRA v CFSR. In (a) and (c) the grid point to grid point scatter plot is given. In (b) and (d) the ranked values are plotted against each other.
The Alaiz test area is composed of Global Wind Atlas job tile 30T-1. The location is northern Spain. Figure 6.11↑
shows that in the lower portion of the wind power density distribution, results from the MERRA and CFDDA are not so different, however for higher power sites the generalized CFDDA gives higher powers than the generalized MERRA. The power density distributions of the MERRA and CFSR based results are rather similar, over the entire range.
Figure 6.12 Denmark region case (32U-4), otherwise same as 6.11↑
The Denmark test area is composed of Global Wind Atlas job tile 32U-4. Figure 6.12↑
shows that the results from the MERRA and CFDDA are not so different. The power density distribution of the MERRA and CFSR based results show a tendency for the CFSR based results to have higher power.
Figure 6.13 Egypt region case (36R-3), otherwise same as 6.11↑
The Egypt test area is composed of Global Wind Atlas job tile 36R-3. Figure 6.13↑
shows tresults from the MERRA and CFDDA indicating CFDDA to give higher estimates of wind power density. The generalized MERRA and CFSR based results also sho a tendency for the CFSR based to have higher power compared to MERRA, but less so than CFDDA.
Figure 6.14 Columbia Gorge region case (10T-4), otherwise same as 6.11↑
The Columbia Gorge test area is composed of Global Wind Atlas job tile 10T-4. It is located in the northwest USA. Figure 6.14↑
shows that for the least windy portion of the wind power density distribution results from the MERRA and CFDDA are similar; in the middle portion, MERRA estimates higher than CFDDA, and in the higher portion CFDDA estimates higher than MERRA. The power density distribution of the MERRA and CFSR based results indicate MERRA estimating higher than CFSR throughout the distribution.
Figure 6.15 Mali region case (30P-4), otherwise same as 6.11↑
The Mali test area is composed of Global Wind Atlas job tile 30P-4. It is located in central Mali. Figure 6.15↑
shows that the wind power density distribution results from the MERRA and CFDDA indicated CFDDA to give higher estimates of wind power density. The power density distribution of the MERRA and CFSR based results also shows a tendency for the MERRA based results to have higher power compared to CFSR.
Figure 6.16 South Africa region case (34H-4), otherwise same as 6.11↑
The South Africa test area is composed of Global Wind Atlas job tile 34H-4. Figure 6.16↑
shows that the wind power density distribution results from the MERRA and CFDDA indicate CFDDA to give higher estimates of wind power density. The power density distribution of the MERRA and CFSR based results also show good agreement.
Figure 6.17 Aegean sea region case (35S-4), otherwise same as 6.11↑
The Aegean Sea test area is composed of Global Wind Atlas job tile 35S-4. Figure 6.16↑
shows that the wind power density distribution results from the MERRA and CFDDA indicate CFDDA to give higher estimates of wind power density, especially in the upper part of the distribution. The power density distribution of the MERRA and CFSR based results also shows CFSR to give higher estimates of power density.
Comparison other wind atlases
In this section the Global Wind Atlas results are compared with results based on different generalized wind climates. This is possible in regions of interest that have been covered by either observation wind atlas studies or numerical wind atlas studies.
Part of Denmark
shows three independent results for part of Denmark at high resolution (job tile 32U-4). The generalized wind climates used in the calculation are different, whereas the topographical descriptions are the same. The results appear very similiar irrespective of whether the generalized wind climates come from MERRA (as in the Global Wind Atlas), from the European Wind Atlas 
, or from mesoscale modelling using the KAMM/WAsP methodology 
. The KAMM/WAsP study for Denmark is verified and described in 
(a) input: generalize MERRA
(b) input: European Wind Atlas
(c) input: KAMM/WAsP
Figure 6.18 Comparison of high resolution mean wind speed [ m s-1] based at 100 m on different sources of generalized wind climates for part of Denmark (job tile 32U-4). Input from generalized wind climates based on (a) MERRA, (b) European Wind Atlas, and (c) KAMM/WAsP.
Part of Mali
shows four independent results for part of Mali at high resolution (job tile 30P-4). As in the Danish case, the generalized wind climates used in the calculation are different, whereas the topographical descriptions are the same. Result from the MERRA and CFSR generalized wind climates are similar, whereas those from generalized CFDDA show stronger winds. Compared to the verified results from the KAMM/WAsP study 
, MERRA and CFSR give the closest results.
Alaiz is one of the test sites for the CENER, Spain. It has been the location for a number of flow studies, including a study using the KAMM/WAsP method 
. Figure 6.20↓
shows four independent results for Alaiz at high resolution (job tile 30T-1). Results from the MERRA and CFSR generalized wind climates are similar, whereas those from generalized CFDDA show slightly stronger winds. Compared to the verified results based on the mesoscale KAMM/WAsP study 
, CFSR appears to give the closest results. The mesoscale based results (Fig. 6.20↓
d) show greater spatical variability, with lower wind speed values in the low wind speed areas caused by mountain lee effects.
Part of Vietnam
shows four independent results for part of Vietnam; three at high resolution (job tile 48Q-2 and 48Q-3) and one result from WRF mesoscale modelling 
, a project funded by the World Bank. Results from CFSR generalized climates give the weakest winds. Results from the MERRA generalized climates give slightly stronger winds. Results from CFDDA generalized climates give higher winds compared to the other reanalysis; over the sea there is better agreement with SAR winds. Compared to the WRF mesoscale modelling results at 5 km resolution (i.e. no micrsocale modelling performed) in the low land areas, where microscale effects are smaller, MERRA and CFDDA appear to give best agreement.
Part of Illinois and Great Lakes
shows four independent results for part of Illinois at high resolution (job tile 16T-1 and 16T-2). Illinois has been mapped using generalized wind climates based on observations 
, shown here in Fig. 6.22↓
d. Results from CFSR and MERRA generalized climates match quite well the observational wind atlas results. Results from CFDDA generalized climates give higher winds compared to the other reanalysis. None of the results suggest the very high winds indicated by the SAR winds shown in Fig. 6.6↑
Part of South Africa
shows four independent results for part of South Africa at high resolution. Western Cape has been mapped in the WASA project using WRF mesoscale modelling 
. The WASA project results are verified against measurements. Results from CFSR and MERRA generalized climates show similar magnitude winds. Over land, results from CFDDA generalized climates give higher winds compared to the other reanalysis. The WASA generalized wind climates give results with much more spatial variability, including a region of low winds that none of the reanalyses capture.
In Fig. 6.25↓
, the high resolution results based on the generalized reanalyses are compared in scatter plots to the high resolution results based on the WASA generalized wind climates. It appears that the CFDDA results give good agreement in the higher portion of the distribution, where it is over sea areas, and also ridge and mountain tops. But it overestimates the winds where WASA suggests low wind areas. MERRA and CFSR indicate a similar ability to capture the distribution of wind speeds, with reasonable agreement in the lower to middle part of the distribution but underestimation in the upper part of the distribution.
Figure 6.25 South Africa region case (34H4) wind speed scatter plots comparing Frogfoot results using generalized wind climates from WASA project and different reanalysis, namely, CFSR, MERRA and CFDDA. The grid point to grid point scatter plot is given then the ranked values are plotted against each other.
As one can infer from the test cases shown above, there are some differences between the re-analysis datasets, and subsequently between the generalized wind-climate statistics derived from them.
The differences between re-analysis datasets extend beyond wind velocities and subsequent power densities. This includes not only spatial and temporal variability due to differing underlying model characteristics, but also surface parameters; of these, one is important in the generalization procedure: the roughness length.
The generalization of wind speeds invokes the geostrophic drag law and a (perturbed) log-law, both of which involve the roughness length. Together, these relations lead to an increase in generalized wind speed for an increase in background roughness, and vice-versa. The three reanalysis datasets each have different background roughnesses, which contribute to some of the differences in generalized winds shown. Most simply, the tendency of the CFDDA to give higher wind speeds over land, along with prescribing larger roughnesses (relative to the other datasets) over modest terrain leads to larger generalized winds. Over modest and flat terrain the CFSR roughness can be yet larger, which can lead to generalized CFSR winds larger than generalized MERRA winds, in locations where the CFSR and MERRA winds are comparable.
Further refinements of the generalization process and/or reanalysis roughnesses can lead to yet more compatible results between generalized wind data from different re-analyses; this is the subject of ongoing research.
Contributing authors Jake Badger, Merete Badger, Mark Kelly, Xiaoli Guo Larsén
 : “Wind-Climate Estimation Based on Mesoscale and Microscale Modeling: Statistical-Dynamical Downscaling for Wind Energy Applications”, J. Appl. Meteorol. Climatol., pp. 1901—1919, 2014. URL http://journals.ametsoc.org/doi/abs/10.1175/JAMC-D-13-0147.1.
 : ESA ResGrow: Trial cases for SAR lifting. Aegean Sea, 2015.
 : “RZGM2013-14”, , 2013.
 : “Offshore wind mapping Mediterranean area using SAR”, Energy Procedia, pp. 38—47, 2013.
 : “Offshore Wind Resources Assessment from Multiple Satellite Data and WRF Modeling over South China Sea”, Remote Sensing, pp. 467—487, 2015.
 : “Applicability of Synthetic Aperture Radar wind retrievals on offshore wind resources assessment in Hangzhou Bay, China”, Energies, pp. 3339—3354, 2014.
 : “Wind retrieval from synthetic aperture radar-an overview”, 4th SAR Oceanography Workshop (SEASAR 2012), 2012.
 : “The ERA-Interim reanalysis: configuration and performance of the data assimilation system”, Q. J. R. Meteorol. Soc., pp. 553—597, 2011. URL http://doi.wiley.com/10.1002/qj.828.
 : “Satellite winds as a tool for offshore wind resource assessment: The Great Lakes Wind Atlas”, Remote Sensing of Environment, pp. 349—359, 2015.
 : “LINCOM wind flow model: Application to complex terrain with thermal stratification”, Physics and Chemistry of the Earth Part B - Hydrology Oceans and Atmosphere, pp. 839—842, 2001.
 : Mesoscale modeling for the wind atlas for South Africa (WASA) Project, 2014. URL http://orbit.dtu.dk/services/downloadRegister/107110172/DTU_Wind_Energy_E_0050.pdf.
 : “A Reanalysis System for the Generation of Mesoscale Climatographies”, J. Appl. Meteor. Clim., pp. 954-972, 2010.
 : “Offshore wind climatology based on synergetic use of Envisat ASAR, ASCAT and QuikSCAT”, Remote Sensing of Environment, pp. 247—263, 2015.
 : “Mapping Offshore Winds Around Iceland Using Satellite Synthetic Aperture Radar and Mesoscale Model Simulations”, I E E E Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015.
 : “Comparison of C-band scatterometer CMOD5. N equivalent neutral winds with ECMWF”, Journal of Atmospheric and Oceanic Technology, pp. 721—736, 2010.
 : European Wind Atlas. Published for the Commission of the European Communities, Directorate-General for Science, Research, and Development, Brussels, Belgium by Risø National Laboratory, 1989.
 : “Weibull-k Revisited: ``Tall'' Profiles and Height Variation of Wind Statistics”, Boundary-Layer Meteorol., pp. 107-124, 2014.
 : “The SAR measurement of ocean surface winds: an overview for the 2nd workshop on coastal and marine applications of SAR”, Proc. of the 2nd workshop on SAR coastal and marine applications, pp. 8—12, 2004.
 : “Observation of tropical cyclones by high-resolution scatterometry”, Journal of Geophysical Research (JGR)-Oceans, pp. 7767—7786, 1998.
 : “MERRA: NASA's modern-era retrospective analysis for research and applications”, J. Clim., pp. 3624—3648, 2011.
 : “Global Distribution and Characteristics of Diurnally Varying Low-Level Jets”, J. Clim., pp. 5041—5064, 2010. URL http://journals.ametsoc.org/doi/abs/10.1175/2010JCLI3514.1.
 : “The NCEP Climate Forecast System Reanalysis”, Bull. Am. Meteorol. Soc., pp. 1015—1057, 2010. URL http://journals.ametsoc.org/doi/abs/10.1175/2010BAMS3001.1.
 : “Scatterometer data interpretation: Estimation and validation”, Journal of geophysical research, pp. 5767—5780, 1997.
 : “The characteristics of wind velocity that favor the fitting of a Weibull distribution in wind-speed analysis”, J. Appl. Meteor. Clim., pp. 124—134, 1984.
 : “Theories for the interaction of electromagnetic and oceanic waves?A review”, Boundary-Layer Meteorology, pp. 61—85, 1978.