Home » APPENDIX » Appendix D


Over the past decade, there has been an ongoing industry debate over different correlation methodologies which can be used for the prediction of the long-term mean wind speed at a site. All correlation methods have a common feature in that they:

  1. Establish a relationship between the concurrent data recorded at the site and reference station; and
  2. Apply the relationship to the historical data recorded at the reference station to predict the long-term wind regime at the site.

Such methodologies are commonly called measure-correlate-predict (MCP) analyses. Variables in such correlation analyses mooted over the past decade include those defined in Tables D.1 and D.2.

Table D.1: Prediction Methodologies Based on Ten-Minute or Hourly Data

Technique Option 1 Option 2 Others…
Directional bin size 30 degrees Other  
analysis technique
Principal component analysis Least squares fit  
Fitting method One parameter fit Two parameter fit Non-linear
Low wind speed cut-off Exclude lowest wind speed data Include lowest wind speed data  

Table D.2: Prediction Methodologies Based on Longer Term Data

Technique Option 1 Option 2 Others…
Averaging period Monthly Daily  
Fitting method One parameter fit Two parameter fit Non-linear
Threshold for data coverage Varies    

The tables present a bewildering array of options. While the technical merit of some methods over other methods can be argued, experience has shown that where the wind regimes at the site and reference meteorological station are well correlated, the results obtained tend to be relatively insensitive to the specific correlation methodology adopted. For cases where the correlation between the site and reference station is less good, then significant divergence is sometimes seen between the results obtained with different methods. In such circumstances, careful checks are required to ensure that the correlation is sufficiently good to justify the use of the reference meteorological station at all. Due consideration also needs to be given to the interpretation of the uncertainty associated with a specific correlation methodology.

The methods based on ten-minute data or hourly data typically use the long-term wind rose recorded at the reference meteorological station. Those based on daily or monthly correlations are dependent on the site wind rose. It is often pragmatically observed that where hourly or ten-minute correlations between a site and reference station are poor, a reasonable correlation is observed over longer time periods such as a month.

Detailed Description of a Measure-Correlate-Predict Analysis

A detailed description of the steps within a measure-correlate-predict analysis is described below based on hourly data from the site and reference station. As indicated in the previous section, different approaches may be used. In the following discussion the proposed wind farm site is referred to as the ‘target site’ and the meteorological station is referred to as the ‘reference site’. The first stage in the approach is to record, over a period of about one year, concurrent wind data from both the target site and the nearby reference site for which well-established long-term wind records are available. The short-term measured wind data are then used to establish the correlation between the winds at the two locations. Finally, the correlation is used to adjust the long-term historical data recorded at the reference site to calculate the long-term mean wind speed at the site.

The concurrent data are correlated by comparing wind speeds at the two locations for each of twelve 30 degree direction sectors, based on the wind direction recorded at the reference site. This correlation involves two steps:


  1. Wind directions recorded at the two locations are compared to determine whether there are any local features influencing the directional results. Only those records with speeds in excess of, say, 5 m/s at both locations are used.
  2. Wind speed ratios are determined for each of the direction sectors using a principal component analysis.

In order to minimise the influence of localised winds on the wind speed ratio, the data are screened to reject records where the speed recorded at the reference site falls below 3 m/s (or a slightly different level) at the target site. The average wind speed ratio is used to adjust the 3 m/s wind speed level for the reference site to obtain the different level for the target site, which ensures an unbiased exclusion of data. The wind speed at which this level is set is a balance between excluding low winds from the analysis and still having sufficient data for the analysis. The level used only excludes wind speeds below the cut-in wind speed of a wind turbine, which do not contribute to the energy production.

The result of the analysis described above is a table of wind speed ratios, each corresponding to one of 12 direction sectors. These ratios are used to factor the wind data measured at the reference site over the historical reference period to obtain the long-term mean wind speed at the target site. This estimate therefore includes the following influences:

  • ‘Speed-up’ between the target site and the reference site on a directional basis; this can be a very important characteristic, and sometimes speed-ups differ by a factor of as much as two; and
  • The wind patterns at the reference site have been translated through the correlation process, so the long-term pattern at the target site has also been established.



  Acknowledgements | Sitemap | Partners | Disclaimer | Contact

coordinated by


supported by

Intelligent Energy Europ

The sole responsibility for the content of this webpage lies with the authors. It does not necessarily reflect the opinion of the European Communities. The European Commission is not responsible for any use that maybe made of the information contained therein.