Global ocean wind fields from SAR data using scatterometer models and neural networks

Abstract

Three weeks of ERS-2 SAR wave mode data, representing a total of 34000 SAR images of 5 km /spl times/ 10 km size, were utilized to verify wind retrieval algorithms on a global basis. Wind speeds are retrieved from calibrated SAR normalized radar cross section (NRCS) as well as uncalibrated SAR intensity images. In case of the calibrated NRCS the well-tested empirical C-band scatterometer (SCAT) model CMOD4 is used, which describes the dependency of the NRCS on wind. Therefore the SAR data are calibrated, which is performed by utilizing a subset of co-located ERS-2 SCAT data. SAR derived wind speeds are compared to co-located winds from the ERS-2 SCAT and model results of the European Centre for Medium-range Weather Forecast (ECMWF). The comparison to ERS-2 SCAT results in a correlation of 0.95 with a bias of -0.01 ms/sup -1/ and a root mean square error of 1.0 ms/sup -1/. In case of SAR intensities a Neural Network (NN) is used that allows to retrieve wind speeds from uncalibrated SAR images. Comparison of NN retrieved SAR wind speeds to ERS-2 SCAT wind speeds result in a correlation of 0.96 with a bias of -0.04 ms/sup -1/ and a root mean square error of 0.93 ms/sup -1/.