Use of the azimuth wavelength cut-off to retrieve the sea surface wind speed from Sentinel 1 and COSMO-SkyMed SAR data

2016 European Space Agency (ESA) Living Planet Symposium, 9-13 May 2016, Prague, Czech RepublicThe purpose of this study is to fit the dependence of the azimuth wavelength cut-off (λC) estimated from X and C band Synthetic Aperture Radar (SAR) data from, respectively, the COSMO-SkyMed (CSK) and the Sentinel 1 missions on the incidence angle, the 10-meters ocean wind speed (U10), and the sea state. The parameters used to characterize the sea state are the significant wave height (SWH) and the Charnock parameter. The final aim is to fit a Geophysical Model Function (GMF) to retrieve the wind speed from independent SAR measurements and validate it with independent buoy wind measurements. The training datasets include collocated scatterometer winds and the European Centre for Medium-range Weather Forecasts (ECMWF) ERA-Interim reanalysis SWH and Charnock data, while the validation dataset includes wind and SWH data from the National Data Buoy Center (NDBC) network. All the SAR images are multi-look and vertically polarized (VV) and have been acquired in strip-map mode. In particular, the X-band training dataset consists of 90 images, part of which has been acquired over the Gulf of Guinea and the rest over the Mediterranean Basin. With this choice we can appreciate the differences of the model performances for open ocean cases and for a semi-enclosed basin. For the GMF fit, the CSK images are co-located in space and time with QuikSCAT scatterometer winds. The X-band validation dataset consists of 26 CSK images co-located with the NDBC datalocated in East Bahamas, Hawaii and the Carribean Sea. The C-band training dataset consists of 198 Sentinel 1 images. All of them have been acquired close to the Hawaii archipelago and have been co-located in space and time with the Advanced Scatterometer (ASCAT onboard Metop) winds. The C-band validation dataset consists of 67 Sentinel 1 images co-located with the NDBC data located off-shore the Hawaii archipelago. The λC has been evaluated by means of the auto-correlation function (ACF) through an inverse Fourier transform of the range-averaged SAR image power spectral density. The ACF has been finally fitted with a Gaussian function. The power spectrum has been computed from 10-km side squared image blocks centred on the co-located scatterometer grid point. The dependence of the λC on each of the geophysical parameters has been assumed linear. The dependence of the λC on each geophysical parameter separately as well as on all the possible combinations has been analyzed. All the fitted GMFs have been evaluated and compared in terms of bias and root mean square error with respect to both the auxiliary parameters used in the fit and the independent NDBC measurements. The preliminary results show that for similar wind and sea state conditions, the computed λC values from CSK images are smaller than those computed from Sentinel 1 images. This is expected because of the smaller range to platform velocity ratio of CSK with respect to Sentinel 1. Furthermore, this study confirms that the λC is mainly dependent on the wind speed, on the SWH and on the incidence angle. The dependence on the incidence angle is qualitatively in agreement with the SAR modulation transfer function. A better agreement may be probably achieved by considering also the direction of the sea spectrum. The validation exercises show that the bias of the retrieved wind speed is not significant from a statistical point of view and the rmse is around 2.5 m/sPeer Reviewe