Progress towards an HF radar wind speed measurement method using machine learning

HF radars are now an important part of operational coastal observing systems where they are used primarily for measuring surface currents. Their use for wave and wind direction measurement has also been demonstrated. These measurements are based on physical models of radar backscatter from the ocean surface described in terms of its ocean wave directional spectrum and the influence thereon of the surface current. Although this spectrum contains information about the local wind that is generating the wind sea part of the spectrum, it also includes spectral components propagating into the local area having been generated by winds away from the area i.e., swell. In addition, the relationship between the local wind sea and wind speed depends on fetch and duration. Thus, finding a physical model to extract wind speed from the radar signal is not straightforward. In this paper, methods that have been proposed to date will be briefly reviewed and an alternative approach is developed using machine learning methods. These have been applied to three different data sets using different radar systems in different locations. The results presented here are encouraging and proposals for further development are outlined.