A PROJECTION METHOD FOR AUTOMATIC ESTIMATION OF WIND VECTORS WITH RADARSAT SAR IMAGERY

A new approach for automatically estimating wind vectors for SAR imagery is presented, which relies on a projection operation to generate wind direction estimates. A threshold can be applied to the projection results in order to eliminate estimation of wind directions in regions where the SAR image contains no visible features. The process also allows multiple possible directions to be generated at a given location using the image features, which are then resolved to a single direction based on uniformity across local regions of the image. The algorithm has been validated using 137 comparisons to in situ buoy observations, giving a direction RMSE of 39 ° (mean error of 10°, error standard deviation of 38°) and a wind speed RMSE of 2.2 m/s (mean error =-1.4 m/s, standard deviation of the error = 1.7 m/s). The largest errors come from automatically utilizing strong features in the image that are not aligned with the local wind, such as atmospheric lee waves are current fronts. If we manually eliminate these from the comparisons, then the direction RMSE is 31 ° (mean error of 20°, error standard deviation of 20°) and the wind speed RMSE is 2.1 m/s (mean error =-1.2 m/s, standard deviation of the error = 1.7 m/s)