Back to the programme printer.gif Print

Wednesday, 12 March 2014
11:15 - 12:45 Wind speed predictions: Are we at the limit of our knowledge or can we improve?
Resource Assessment  

Room: Ponent
Session description

No new long-term correction methods have appeared for years and it is possible that current techniques are optimal. There are several issues which affect any long-term correction analysis:

… from the mundane: the optimum reference period? how do we measure success? re-analysis data or ground based stations? non-integer years of data?

… to the exotic: atmospheric stability, climate change decadal variations, sun spots activity and solar cycles.

It is likely that long-term correction techniques which consider these may provide more reliable predictions than has previously been possible.

This session describes new long-term correction methodologies and compares the results with those of conventional methods. Innovative techniques to improve the representativeness of long-term data series are discussed, different long-term data series are compared and conclusions on the decadal-scale variability of the wind speed are presented. The overall objective of this session is to give insight on how these developments contribute to a greater certainty in future wind speed predictions.

Learning objectives

  • Evaluate innovative methods to improve the representativeness of long-term data series and the overall accuracy of long-term extrapolations
  • Compare new long-term correction methods to traditional methods
  • Understand how a more accurate description of the decadal-scale variability of the wind speed contributes to the reduction of the uncertainty in the long-term corrected wind speed
Lead Session Chair:
Sónia Liléo, Kjeller Vindteknikk AS, Sweden

Steve Ross, 3Tier
Erik Holtslag Ecofys, The Netherlands
Anthony Crockford (1) F P
(1) Ecofys, Utrecht, The Netherlands

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

Erik Holtslag is Team Leader Wind resource Assessment within the Ecofys-wind unit and he is working as Operational Manager for the Test Site Lelystad prototyping & certification field. After obtaining a master’s degree in Meteorology at the Wageningen University, he has worked in wind energy since 2002. His specializations are in wind measurements using met masts and LiDARs and in feasibility studies. Additional expertise is wind farm design and micro-siting, both for onshore and offshore wind farms. Amongst others he was responsible for the full wind resource trajectories of over 1000 MW of projects worldwide.


In-depth validation key to acceptance of mesoscale results


As a precursor to a bankable wind resource assessment, Ecofys has validated the accuracy of six mesoscale models. The mesoscale model outputs were compared in detail with measured data from five coastal wind measurement campaigns with excellent data quality. The statistical analysis showed that accuracy of all models was good, while some models outperformed others substantially on key metrics. It was thus possible to select the best model for the wind resource assessment, and evaluate the uncertainty of the model.


Following a market study of leading mesoscale modellers and an evaluation of the various modelling approaches, Ecofys selected the best available data on the market – including data from AWS Truepower, INNOSEA and Vortex. Time series of hourly mesoscale data was commissioned for each of the validation sites.

The validation sites represent five high-quality wind measurement campaigns in similar terrain. A number of statistical tests were performed for the concurrent datasets. A battery of over 30 statistical tests (including sector-wise analysis) was performed on the measured and modelled time series, focusing on correlation, bias and the quality of the distributions.

Main body of abstract

Following a detailed quality assessment, data from five campaigns were selected including LiDAR and tall met masts, with measurement periods between 6 months to 3 years. The measurement locations were visited by an Ecofys engineer to verify the on-site measurement set-up, surroundings and expected impact on data quality.

The measured wind data was filtered according to detailed documentation, and corrected to the reference time zone (UTC). Data in highly disturbed wind direction sectors has been discarded. The provided datasets were thoroughly investigated to determine the data quality and its suitability for use in mesoscale model validation.

A selection of statistical results shows our key findings. Two statistical tests evaluated the relationship between time series, in terms of wind speed and direction: wind speed correlation; and wind direction correlation. The correlation coefficients for all models were high, ranging from 80-90%.

Three tests reveal any bias in the modelled data and the magnitude of the difference: mean difference in wind speed (bias); standard deviation in wind speed difference; and mean absolute difference in wind speed. Again, the results across all models were comparable, with some models consistently performing better.

Since the accuracy of a wind resource assessment is primarily concerned with the wind speed distribution rather than the time series, two further tests are important: Kolmogorov-Smirnov test statistic; and difference in calculated energy yield for representative wind turbine. Clear differences were found between models; two of the six models produced consistently more accurate wind speed distributions.


While all mesoscale models were comparable in terms correlation and bias, it was found that the accuracies of the overall distribution were not all equal. This is particularly important for wind resource assessments, where accuracy in wind speed distribution will significantly improve the accuracy of energy yield predictions. Thus, the two distribution tests were crucial in the differentiation between models.

Based on this validation exercise, it was possible to make an informed selection of mesoscale model for the subsequent wind resource assessment. The validation also helped to evaluate the uncertainty in wind speed.

Learning objectives
Rather than focusing on the site-specific results, this presentation intends to show the framework for evaluation and validation of mesoscale models. The accuracy of the models was high, but the results highlight that there remain significant differences between commercially-available datasets. The statistical process outlined here could be applied to any wind farm intending to use mesoscale data, provided that a thorough validation is possible with nearby measured data.