Posters

Abstract

The accuracy of wind resource assessments plays a critical role in wind energy project development. One challenge has been effectively merging the advantages of low-resolution mesoscale models with the detailed accuracy of high-resolution microscale models. Mesoscale models offer more meteorologically accurate inputs, which better represent regional wind conditions. However, they typically achieve a resolution no finer than 100 meters horizontally. On the other hand, microscale models are able to capture the impact of topography, achieving horizontal resolutions typically between 10 and 20m in wind resource assessments. To bridge these two worlds, we propose a two-dimensional convolution method to average each data source to a relevant resolution where the mesoscale model dominates. Then, we apply the ratio between these results to the high-resolution microscale wind map. Supporting case studies from operating wind farms in Norway were used to evaluate this methodology. For example, at Kjøllefjord wind farm, where there is a strong wind speed gradient not covered by a microscale model, the cross-prediction error was reduced from 19.4% to 6.9% using the mesoscale correction. At Kvenndalsfjellet site, with 5 onsite met masts available, the mean absolute cross prediction error was reduced from 7.9% to 1.0%. However, we will also show that the mesoscale model is not always improving the results as results vary depending on local conditions; for example, at Storheia, the mean absolute cross prediction error slightly increased. Further analysis of SCADA data revealed that while the mean absolute error at the turbine level improved, there was still a trend of overprediction in the northern part of the wind farm with the mesoscale correction model.