Posters

Abstract

The long-term wind speed plays a key role in estimating the wind potential of a site and, consequently, its profitability. Therefore, it is essential to estimate this long-term wind speed as accurately as possible, which is ensured by installing measurement devices several years prior to the project. In the context of preliminary studies, no robust measurement campaigns are conducted, and the data available for the developer usually comes from wind atlases. While these provide useful input, they may yield misleading mean wind speeds that can be significantly different from the true values on site. With over 150 projects where reliable measurement campaigns have been conducted, a statistical comparison between the actual measured wind speed and the wind speed predicted by various wind atlases has been performed. This analysis reveals significant discrepancies depending on geography, topographic variables, and roughness. It has contributed to a better understanding of the behavior of wind atlases and has been used to recalibrate predicted wind speeds using reliable measurement data along with open-source topographic and roughness data. Numerous statistical and machine learning models have been investigated, and evaluation metrics have been established in advance. These metrics include statistical measures of the errors made by each model, as well as business metrics such as the percentage of predictions that outperform the wind atlases.  The results demonstrate that the combination of atlases and their recalibration with project-specific variables provides better outcomes than using raw wind atlases in approximately 80% of cases. Although this work does not replace reliable measurement campaigns, it facilitates higher-quality preliminary studies. Further improvements could include modeling wind distribution instead of average speed and modeling wind shear using similar methods.