Posters

Abstract

In recent years, Scanning LIght Detection And Ranging (LiDAR) instruments have been increasingly employed for wind speed measurements and proved to be a reliable measurement strategy. LiDAR systems are capable of gathering measurements by scanning over wide spatial volumes, resulting in large amount of information and thus incentivising the usage of data driven tools. Gaussian Processes (GPs) are a machine learning regression technique which can be used to interpolate Radial Wind Speed measurements (RWS) at points in space and time of interest. In this work a GP driven tool is used to interpolate RWS measurements from two scanning LiDARs deployed in an onshore measurement campaign in Bremerhaven, Germany. The GP model is set to estimate the RWS at desired spatial locations, using the data from the scanning patterns. The RWS estimates are validated using data from a cup anemometer placed at the cross section of the two LiDARs fields of views. A linear fit of y=1.01x+0.08 between the anemometer wind and GP RWS estimates is found, with a R2 value of 0.934 for the complete duration of the campaign of 23 days. In addition, two independent GP models are further applied so that planar radial wind speed estimates can be combined and used for two dimensional wind speed reconstruction over a 300x300 meter horizontal plane. A 96 hour time widow is analysed and it is seen that both the u and v components can be accurately reconstructed, where a comparison shows R2 values as high as 0.937 and 0.892, for the u and v components respectively. These results suggest that Gaussian Processes applied to multiple LiDARs have the potential to provide high frequency two and three dimensional wind field reconstructions over large volumes as well as a to be a tool to correct RWS measurements in devices with offsets or mistakenly configurated.