High-resolution offshore wind resource assessment at turbine hub height with Sentinel-1 synthetic aperture radar (SAR) data and machine learning

Abstract

Abstract. This paper presents a method for estimating offshore extractable wind power at hub height using Sentinel-1 synthetic aperture radar (SAR) data and machine learning. The method was tested in two areas off the Dutch coast, where measurements from Doppler wind lidars installed at the sea surface were available and could be used as a reference. A first machine learning algorithm improved the accuracy of SAR sea surface wind speeds by using geometrical characteristics of the sensor and metadata. This algorithm was trained with wind data measured by a large network of weather buoys at 4 m above sea level. After correction, the bias in SAR wind speed at 4 m versus buoys was 0.02 m s−1, with a standard deviation of error of 0.74 m s−1. Corrected surface wind speeds were then extrapolated to hub height with a second machine learning algorithm, which used meteorological parameters extracted from a high-resolution numerical model. This algorithm was trained with lidar vertical wind profiles and was able to extrapolate sea surface wind speeds at various altitudes up to 200 m. Once wind speeds at hub height were obtained, the Weibull parameters of their distribution were estimated, taking into account the satellites' irregular temporal sampling. Finally, we assumed the presence of a 10 MW turbine and obtained extractable wind power with a 1 km spatial resolution by multiplying the Weibull distribution point by point by its power curve. Accuracy for extractable wind power versus lidars was ± 3 %. Wind power maps at hub height were presented and compared with the outputs of the numerical model. The maps based on SAR data had a much higher level of detail, especially regarding coastal wind gradient. We concluded that SAR data combined with machine learning can improve the estimation of extractable wind power at hub height and provide useful insights to optimize siting and risk management. The algorithms presented in this study are independent and can also be used in a more general context to correct SAR surface winds, extrapolate surface winds to higher altitudes, and produce instantaneous SAR wind fields at hub height.