A Novel Azimuth Cutoff Implementation to Retrieve Sea Surface Wind Speed From SAR Imagery

In this paper, the synthetic aperture radar (SAR) azimuth cutoff method is thoroughly revised and a new and general implementation is proposed. The key roles of the pixel spacing, the size of the image box, and the texture of the SAR scene are analyzed and optimized in terms of azimuth cutoff (lambda(c)) estimation. The reliability of the lambda(c) estimation is analyzed by measuring the distance between the measured and fitted autocorrelation functions. This analysis shows that it is of paramount importance to filter unfeasible/unreliable lambda(c) values. To identify those values in an objective way, a criterion that is based on the chi(2) test performed over a large data set of Sentinel-1 SAR imagery is defined and proven to be effective. The new robust implementation of the lambda(c) estimation at about 1-km grid spacing is then used to produce averaged lambda(c) at about 10-km grid spacing. The performance of the new estimation procedure, analyzed using a lambda(c)-to-wind-speed forward model, is shown to provide improved wind speed retrievals, with a root-mean-square error of 1.8-2 m/s when verified against independent numerical weather prediction model output and scatterometer winds.