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.

Jon Meis EWC Weather Consult GmbH, Germany
Achim Strunk (1) F P Joris Brombach (1) Jon Meis (1) Frank Sehnke (2) Martin Felder (2) Anton Kaifel (2)
(1) EWC Weather Consult GmbH, Karlsruhe, Germany (2) Centre for Solar Energy and Hydrogen Research (ZSW), Stuttgart, Germany

Printer friendly version: printer.gif Print

Presenter's biography

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

After successfully graduating in Geophysics at the University of Cologne, Achim Strunk worked at the Rhenish Institute for Environmental Research on four-dimensional variational data assimilation and its application in atmospheric chemistry. As a post-doctoral researcher he focused on HPC implementations of air quality data assimilation and forecasting. Afterwards he worked on chemistry climate interactions at the Royal Netherlands Meteorological Institute (KNMI). Achim Strunk gained 15 years experience in meteorological model development and application and contributed to various international projects. Since 2012 he is heading the Energy Division at EWC Weather Consult.


A novel MCP method based on deep neural networks for long-term correction in wind resource assessments


The estimation of local wind resources mainly suffers from missing representative information in the vicinity of a site like long-term time series of wind measurements in hub height. The aim of the method under consideration is to allow an accurate and rapid wind resource assessment. Such reliable information is essential for being able to make the right decisions in certain projects and minimize financial risks.


A long-term correction (LTC) is performed which aims at optimally combining long-term reanalysis data and short-term measurements. This is achieved by a novel implementation of a measure correlate predict (MCP) algorithm which is based on state-of-the-art deep neural networks including sophisticated tricks-of-the-trade. This approach is compared to commonly used MCP methods implemented in standard-like software packages.

Main body of abstract

Deep Neural Networks are generally able to represent complex dependencies and are therefore very well suited to be applied to the complex and non-linear relations in the atmosphere. However, when applying a complex model, the main problem is to guarantee the generalisation skill of such a Deep Neural Network by limiting and controlling the overfitting.
In order to evaluate the quality of this approach the results are compared against two other MCP methods which are quite standard in wind resource assessments. Two main experiments are undertaken targeting 1) the dependency of the approach to the length of measurement campaigns and 2) very complex sites for which no reliable reference time series is available.
In both case studies the novel Neural Network MCP clearly outperforms the standard methods when comparing different skill scores on validation data sets. While, for example, the Matrix method achieves a conservation of the observed wind speed frequency distribution, the Neural Network MCP additionally improves the correlation coefficient, especially for complex locations.
The reduction of errors in the expected energy yield reaches about 35% for a mix of different stations and 12 months of observations and about 50% for very complex sites, compared to the next best MCP method.


The presented MCP method helps to significantly reduce uncertainties in wind resource assessments. Moreover, due to the usage of additional physical parameters from reanalysis data sets in the training process, an analysis of very complex sites gets feasible by reconstructing, for example, thermal circulations or other meteorological processes which cannot be covered by reanalysis data sets or downscaling approaches. By that, significant reduction of project uncertainty can be achieved which directly transfers to the economic value of a project.

Learning objectives
Upon completion, the audience will be able to ...
1) ... describe how a state-of-the-art long-term correction (LTC) is performed for wind resource assessment.
2) ... assess different measure correlate predict (MCP) algorithms.
3) ... define what labels a state-of-the-art Neural Network that is generally able to produce outstanding results - not only in the area of wind resource assessment.