Datasets in the Global Wind Atlas

The data used in the Global Wind Atlas was chosen from the best availible global datasets for each required category. This meant the datasets need to both be of high quality, but also high enough resolution that the downscaling process would not be missing large amounts of information.

Reanalysis data has been used in the wind energy sector, particularly wind resource assessment, for some years. The increasing resolution of these datasets has made the Global Wind Atlas methodology a possibility. Here details of the reanalysis data used in the Global Wind Atlas are given and the ways we reformat the reanalysis data for the purposes of Atlas are described.

A reanalysis is a scientific method for developing a comprehensive record of how weather and climate are changing over time. In a reanalysis observations around the globe and a numerical weather prediction model, which simulates one or more aspects of Earth system, are combined objectively to generate a synthesized estimate of the state of the system. In the 2010s four large atmospheric reanalysis projects took place. These are listed in Table 2.1↓. All provide output at a horizontal resolution below 1^° × 1^°. Three of these reanalysis (i.e., CFDDA, CFSR and MERRA) were generalized and later used in the global wind atlas.

Many of the models described in Table 1↓ were run at a high spatial resolution but the data was truncated (e.g. CFSR, ERA-Interim) or interpolated (e.g. CFDDA) to a lower resolution when made available to the users. This means that the data is not available in the native model grid (e.g. there is no information on the real surface elevation and land use) and has implication for the generalization procedure carried out. See the Generalization section for more information about how the reanalysis datasets were generalized.

The reanalysis datasets come in varying meteorological data formats. Therefore it is important to harmonize the file formating for later processing of the reanalysis datasets. For the Global Wind Atlas it is the climatological properties of the datasets that are of primary interest, rather than the datasets as a time series. Therefore the reanalysis data is compiled into a tabular form. The table content is similar to the WAsP observed wind climate file format and contains the following information. This table data stored in a single large netcdf file for each reanalysis dateset.

• Latitudes [°], Longitudes [°] and heights 50 m, 100 m, 200 m
• Sector-wise frequencies of occurrence [%] for 12 direction sectors
• Upper limit for speed class 1 (1 ms^− 1), sector-wise frequencies in class 1
• Upper limit for speed class 2 (2 ms^− 1 ), sector-wise frequencies in class 2
• .. .. ..
• Upper limit for speed class 40 (40 ms^− 1), sector-wise frequencies in class 40

This file is called the reanalysis TAB netcdf file and contains the statistical information required for further downscaling.

Figures 1↓, 2↓, 3↓, 4↓ show the mean wind speed at 100 m above the surface for the MERRA, CFDDA, CFSR and ERA-Interim reanalyses. The general patterns of the winds are similar. Over the oceans the strong westerly in the mid-latitudes, and the trade wind belts in the subtropics are shown. However there are differences in exact placement of these features and magnitudes. For example winds over land masses appear to be lower for the ERA-Interim and CFSR reanalyses, and higher for CFDDA reanalysis. Part of the explanation for this is the difference in boundary layer parameterization schemes and surface descriptions in the reanalysis models. This is discussed in more detail the Generalization section of the methods.

Figure 1 

Mean wind speed at 100m from MERRA reanalysis. Period 1979-2013.

Figure 2 

Mean wind speed at 100m from CFDDA reanalysis. Period 1985-2005.

Figure 3 

Mean wind speed at 100m from CFSR reanalysis. Period 1979-2010.

Figure 4 

Mean wind speed at 100m from ERA-Interim reanalysis. Period 1979-2012.

Although climatological properties are the primary interest, the Global Wind Atlas also serves information about how a location’s large scale forcing varies on average throughout the day, and throughout the year. This is useful information in order to approximately gauge how wind resources are distributed temporally. For this purpose the reanalysis data is averaged according to month of year and hour of day. A tabular form is used with the following content:

• Latitudes [°], Longitudes [°] and heights 50 m, 100 m, 200 m
• Months of year
• Hours of day
• mean wind speed for month 1
• standard deviation of wind speed for month 1
• mean wind speed for month 2
• standard deviation of wind speed for month 2
• .. .. ..
• mean wind speed for hour 1
• standard deviation of wind speed for hour 1
• mean wind speed for hour 2
• standard deviation of wind speed for hour 2
• .. .. ..

This file is called the reanalysis CYCLE netcdf file. An example of the average daily and annual cycle from the CFDDA reanalysis is shown in Fig. 5↓. The wind speed values are normalized by dividing by the annual mean wind speed. This is done to make sure that the wind speed value is not misunderstood as an expected wind speed for the location. It can be seen for this location there is a strong daily cycle, with maximum wind speeds at around 12 UTC. There is also a moderate annual cycle, with maximum winds around July.

Figure 5 

Average daily and annual wind speed cycle, and wind direction frequency rose at 50 m from CFDDA reanalysis for a location around Northeastern Lake Victoria (0.4°S, 34.0°E). The shading on the cycle plots shows the standard deviation of monthly and hourly means. Period 1985-2005.

The reanalysis datasets have been described. The way they are formatted into netcdf files containing the required information for the Global Wind Atlas downscaling and annual and daily cycle charactistics has been outlined. How the reanalysis TAB file is used is described in the methodology section. Information about the different map tools that can use the reanalysis data can be found in the tutorials.

Contributing authors Jake Badger, Andrea N. Hahmann, Bjarke T. Olsen

Public datasets describing the earth’s topography have become available at increased resolution in recent years. These impressive datasets make the Global Wind Atlas feasible. For the purpose of the Global Wind Atlas, the topography description can be split into two parts:

• the description of the surface elevation; refered to here as orography, as in the WAsP terminalogy.
• the description of the surface land use or class

The Global Wind Atlas used digital elevation models from Viewfinder Paroramas. This data was developed in support of their downloadable software that draws panoramas from viewpoints. The data has a 3 arc-second resolution and is mainly based on data collected by the Shuttle Radar Topography Mission (SRTM). SRTM data has been availible for use in WAsP through the WAsP map editor, however, it does not have any coverage north of 60°. This would have prevented the modeling of large parts of Scandinavia, Russia and Canada. Additionally it is known that the raw SRTM data has no-data (void) areas in mountainous and desert areas and has some other errors. These have been corrected in other datasets, but the 60° N restriction still applies in those datasets. The viewfinder DEM has voidfilled those areas using the best availible alternative sources, which depended greatly on the region being voidfilled. Detailed information about the datasets used for voidfilling can be found at http://www.viewfinderpanoramas.org/dem3.html

The viewfinder DEM was provided in the WGS 1984 coordinate system (EPSG: 4326), as a global raster that was chopped into 1° by 1° tiles. Therefore the data had to be reprojected to UTM rasters to match the global wind atlas tile format. This interpolation was done using the Geospatial Data Abstraction Library (GDAL). In addition to reprojecting the data, the data was warped to a 150m resolution, which cooresponds to the effective resolution of the SRTM data. This conversion used the nearest neighbor resampling algorithm from the gdalwarp tool.

Roughness length in the global wind atlas was derived from the GlobCover 2009 land cover map. The GlobCover dataset was created by ESA and the Université catholique de Louvain (UCL) to convert MERIS FR (Medium Resolution Imaging Spectrometer Instrument Fine Resolution) surface reflectance mosaics into 22 land cover classes as defined by the United Nations Land Cover Classifcation System (LCCS). The GlobCover dataset was selected as it is a relatively recent land-cover map, with a consistent approach applied to all parts of the globe. It has a 10 arc-second ( 300 m) resolution and was provided in the WGS 1984 coordinate system (EPSG: 4326). The data was converted to the GWA tiles using gdalwarp as described in the Orography section, but the spatial resolution of 300m was retained.

While GlobCover is a full global raster, one of the classification types is no-data. This datatype was mostly in areas North of 60°. To voidfill these regions we used the 0.5 km MODIS-based Global Land Cover Climatology. This dataset was based on 10 years of data 2001-2010, and had 17 land cover classes. These classes were mapped to the GlobCover classes and then used to fill any no-data points.

Once the data was reprojected and the no-data points were filled, the data was converted to roughness length by defining a specific height to each of the land use classes. A reproduction of this mapping is provided below. The mapping was created based on the class name as well as looking at a global map of the classes and using expert input from DTU Wind Energy modelers.

Use of these global datasets for orography and land use will not give as accurate a description of topography as dedicated land surveys, such as those from large scale mapping from national cartography institutions. Although 150 m and 300 m resolution for orography and roughness length gives an impression of high resolution when related to global coverage, an actual wind turbine site will not be well represented at this resolution. Therefore the Global Wind Atlas is not appropriate for specific site assessment.

Another limitation is due to the conversion of land use or class to a single surface roughness length. This discretization of land use type is somewhat unrealistic. Furthermore, that a single land use can be assigned a single surface roughness is unlikely to be correct because a single land use may actually represent a range of vegitation covers.

For these reasons we expect uncertainty to be introduced to the wind climate calculations. However, this uncertainty needs to be balanced against the uncertainty that is inherent in not considering the high resolution topography at all. Therefore we advise that the appropriate use of the Global Wind Atlas is for initial aggregated area studies and not for site specific assessments.

The relevance of the surface description has been outlined, and broken into orography and land use or class. The sources of the topography data have been described. The limitations of the topography data and method have been discussed. The topography will be used for the calculation of local wind climates. The topography data is also shown in the Atlas.

Contributing authors Neil Davis, Jake Badger

The principle of the Global Wind Atlas is to make available the data relevant to the calculation of the wind climatologies. This is ensure transparancy of methodology. For example, the user can see what has been used for surface roughness length in their region of interest. The user can also see the flow modelling effects and the terrain ruggedness. This allows the user to see inside the methodology and assess the model behaviour. It also allows the user to assess the quality of the input data, which has direct influence on the quality of the output data.

These datasets tell the user about the flow modelling characteristics. Details of the flow modelling can be found 5↑. Using these data, the user can gain an understanding of the microscale phenomena that are important in their region of interest. For example, one can see if roughness change effects have an impact, or whether there is a significant orographic speed up effect. Here also the ruggedness index (RIX) is given. This is useful for users to assess uncertainty in the modelling output.

Roughness effects

• WAsP Meso roughness per direction sector
  This dataset gives the roughness length that is used as the basis when calculating the roughness speed up effects. It is dependent on direction sector. The dataset is from the 250 m resolution dataset sampled at 1 km spacing.
• WAsP Roughness speed up per sector at 100 m
  This dataset gives the roughness speed-up as a percentage at 100 m. From these values you can see how the actual surface roughness distribution gives an increase or decreasing of the wind speed compared to a surface with constant roughness equal to the meso roughness. The dataset is from the 250 m resolution dataset sampled at 1 km spacing.

Orography effects

• WAsP Orographic speed up per sector at 100 m
  This dataset gives the speed up as the percentage speed increase due to orography as compared to a completely flat terrain. The dataset is from the 250 m resolution dataset sampled at 1 km spacing.
• WAsP Ruggedness Index
  This dataset gives the percentage of the terrain within 3.5 km which is steeper than a gradient of 0.3. The dataset is from the 250 m resolution dataset sampled at 1 km spacing. (see [44] and [45])

Modelling set-up information

• MRGS Tiles
  Here the locations of the Military Grid Reference System (MGRS) zones, subdived into strips 2 ^○ in latitude (typically 4 subdivisions), are displayed. This reference system gives each of the Global Wind Atlas job tile a unique identity. This is useful if a user wants to find out the particular job tile number for a particular region.
• Computation Boundaries
  Here the computation area within a Global Wind Atlas job tile is displayed. This shows explicity where calculations are performed on nodes with a 250 m spacing. It can be seen that the computation area boundary follows a 30 km buffer around coastlines and that they are divided along modified MGRS zones.

These data are direct values from the microscale local wind climatology caculation. Due to the size of the dataset, only a subsample set of data are served on the website. Those interested in applying the full calculation data should contact DTU Wind Energy.

High-resolution wind speed and power density

• Direct result 1000 m sampling at 50, 100, 200 m
  This dataset gives the value of the wind speed directly from the microscale modelling result, without any aggregation, sampling the 250 m spaced calculation nodes every 1 km. The locations of these sampled values correspond to the locations the modelling data is given.

In order to give the user quick access to the richness of the Global Wind Atlas results, an aggregation method has been devised. With the dataset listed below the user can see the characteristics of the windiest sites inside a 1 x 1 km area, and can see how much variation of resource there is within that area.

High-resolution wind speed

• Aggregated mean wind speed at 50, 100, 200 m
  This dataset gives the mean wind speed for a 1 x 1 km area, composed of 16 calculation nodes with a spaced of 250 m. In proximity to a computational boundary there may be fewer than 16 calculation nodes. The resolution is 1 km.
• Aggregated Top-Quartile mean wind speed at 50, 100, 200 m
  This dataset gives the mean wind speed for the windiest 4 calculation nodes for a 1 x 1 km area, composed of 16 calculation nodes with a spacing of 250 m. This value represents the mean wind speed for the best sites in the 1 x 1 km area. In proximity to a computational boundary there may be fewer than 16 calculation nodes. The resolution is 1 km.
• Aggregated Top-Quartile Bottom-Quartile mean wind speed difference at 50, 100, 200 m
  This dataset gives the difference between the mean of the windiest 4 calculation nodes and the mean of the least windy 4 calculation nodes for a 1 x 1 km area. This indicates the variation of wind resource within the 1 x 1 km area. In locations where this difference is small, the resource is homogeneous, and wind turbine micrositing may be less important. In locations where this difference is large, the resource is heterogeneous on small scales, and micrositing of wind turbines is likely to be very important.

High-resolution power density

• Aggregated mean power density speed at 50, 100, 200 m
  
• Aggregated Top-Quartile mean power density at 50, 100, 200 m
  
• Aggregated Top-Quartile Bottom-Quartile power density wind speed difference at 50, 100, 200 m