Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'Wind speed predictions: Are we at the limit of our knowledge or can we improve?' taking place on Wednesday, 12 March 2014 at 11:15-12:45. The meet-the-authors will take place in the poster area.

Carolin Schmitt juwi Energieprojekte GmbH, Germany

(1) juwi Energieprojekte GmbH, Woerrstadt, Germany (2) EWC Weather Consult GmbH, Karlsruhe, Germany (3) ZSW, Stuttgart, Germany

Printer friendly version: printer.gif Print


How to get a reliable long-term reference in complex, data void regions


Long term wind speed prediction for energy yield estimation normally includes a combination of on-site wind measurements with long reference time series by measure-correlate-predict (MCP) methods. Numerous long-term data sets are made available, but usually they don’t serve for a reliable long-term estimate in complex terrain. Moreover, complexity-reduced MCP methods don’t allow for inferring the local characteristics such that long-term correction don’t reach acceptable error ranges. According to site specific conditions, optimal choice of long-term data and MCP method can be crucial to the accuracy of final data and therefore strongly influence the value of a wind project.


In the first step of this study the local characteristics of a complex site with specific meteorolocial conditions are introduced by means of daily wind speed cycle and governing physical processes in general. The measurements from met-masts are compared to long-term datasets from different providers. These data sets are subject to very different methodologies and temporal and spatial resolution. The suitability for an MCP with the observational data is discussed. As a final part we evaluate the performance of different MCP techniques when compared with long-term data of moderate representativeness.

Main body of abstract

The site under consideration shows very specific orographic and climatological conditions. It is situated in a steep coastal valley showing very predominant inter-diurnal wind speed cycles forced by land-sea-breeze effects. Moreover, non-logarithmic wind profiles are obtained in the valley due to channeling effects. These effects normally can’t be reproduced by most of the available long-term data. An assessment of wind energy resource is therefore governed by two major challenges:
1) Finding datasets with correlations values that allow further processing
2) Finding appropriate MCP methods to reproduce the characteristics properly
A comparison of measured and long-term data sets reveals different stages of quality. The representation of the diurnal cycle and vertical wind profile are of poor to moderate skill. Therefore only low correlation coefficients could be found.
In order to get a valid and reliable long-term correction estimate, different MCP methods are combined with one of the long-term data sets. To this end, the observational time series is split into training and test set. The former is used as input to a number of MCP methods, while the latter is withdrawn from the process and used as independent validation data set. Therefore the performance of the different MCP methods can be evaluated on the same data and objectively compared. The skill of an approach is assessed by mean error, correlation, and wind speed frequency distribution error. It is shown that a novel Neural Network MCP method outperforms standard methods


Despite the fact that numerous long-term data sets are available for locations word-wide today, it cannot be guaranteed to find a representative time series for sites in complex terrain. Standard MCP methods don’t allow producing a prediction with acceptable skill when using long-term data of moderate correlation as reference. A combination with a state-of-the-art MCP technique based on Neural Networks achieves very good results for the site under consideration, both in terms of correlation as well as for the wind speed frequency distribution. Therefore the uncertainty in energy yield a

Learning objectives
The audience will understand
• why long-term data might not be representative for sites in complex terrain,
• how difficult it is to choose a suitable MCP method,
• what gets feasible by using a novel, state-of-the-art MCP technique and
• what this means for the value of a wind energy project.