Global wind speed retrieval from complex SAR data using scatterometer models and neural networks

Abstract

The global availability of synthetic aperture radar (SAR) wave mode data from the European remote sensing satellite ERS-2 allows to investigate the wind field over the ocean on a global and continuous basis. For this purpose 27 days of ERS-2 SAR wave mode data were processed to single look complex SAR images, representing a total of 34310 images of size 5 km/spl times/10 km, available every 200 km along the satellite track. In this paper two methods for retrieving wind speeds from SAR images are presented and validated, showing the applicability of both methods for global ocean wind retrieval. The first method is based on the well tested empirical C-band scatterometer (SCAT) models which describe the dependency of the normalized radar cross section (NRCS) on wind speed and direction. To apply C-band models to SAR data the NRCS has to be accurately calibrated. This is achieved by a new simple but effective method using a subset of collocated ERS-2 SCAT and model winds from the European Center for Medium Range Weather Forecast (ECMWF). SAR derived wind speeds are compared to the entire set of collocated ERS-2 SCAT and ECMWF model data. Comparison to ERS-2 SCAT results in a correlation of 0.94 with a bias of -0.45 m s/sup -1/ and a root mean square error of 1.21 m s/sup -1/. The second approach is based on neural network algorithms allowing one to retrieve wind speeds from uncalibrated SAR images. For this purpose a neural network is trained using ERS-2 SAR data and collocated wind data from ERS-2 SCAT and the ECMWF atmospheric model. Validation of the neural network retrieved SAR wind speeds to ERS-2 SCAT and ECMWF model wind data is performed. A correlation of 0.95 with a bias of -0.1 m s/sup -1/ and a root mean square error of 1.03 m s/sup -1/ is achieved in comparison to ERS-2 SCAT.