Comparison of RADARSAT-1 SAR retrieved wind fields to numerical models

Abstract

An algorithm is introduced and verified, which is designed to retrieve high-resolution wind fields from C-band synthetic aperture radar (SAR) operating at both vertical and horizontal polarization. SAR wind retrieval is a two-step process: In the first, step wind directions are ex- tracted from wind-induced streaks, which have a typical spacing of 200 to 1600 m and are normally aligned with the mean surface wind direction. In the second step, wind speeds are derived from the normalized radar cross section (NRCS) and image geometry of the calibrated SAR images, to- gether with the local wind direction retrieved in the first step. Several semi empirical C-band models are available, which describe the dependency of the NRCS on wind speed, wind direction and image geometry. To ver- ify the algorithm, wind fields were computed from 84 RADARSAT-1 SAR and ScanSAR scenes of the east coast of North America and compared to co-located results from the high resolution numerical model MM5. I. INTRODUCTION Synthetic aperture radars (SARs) are flown on several satellites, e.g., European satellites ESR-1, ERS-2 and EN- VISAT, Japanese satellite JERS and the Canadian satellite RADARSAT-1. Their independence of daylight and cloudi- ness together with their high resolution and large spatial cov- erage make them a valuable tool especially in coastal areas for measuring and observing geophysical parameters, e.g. ocean waves (1) and surface winds (2), (3). The SARaboard the Canadian satellite RADARSAT-1 operates in the C-band at moderate incidence angles. For this wavelength and range of incidence angles the backscatter from the ocean surface is primarily caused by the small-scale surface roughness, which is strongly influenced by the local wind field. Therefore, the backscatter can be empirically related to the wind. In the past few years much effort has been undertaken to develop algorithms for derivation of wind vectors from SAR images. The wind direction can be retrieved from the direction of wind-induced streaks, which are visible in most SARim- ages and are related to the mean wind direction. The direction of these streaks can either be retrieved by using spectral meth- ods or in the spatial domain by a method based on derivation of local gradients. The wind speed is derived from the normal- ized radar cross section (NRCS), which is retrieved from the SARdata, using semi empirical C-band models which were especially developed for vertical (VV) polarization. In case of HH polarization these models have been extended for the polarization ratio. The main objective of this paper is to introduce a recently developed algorithm for wind field retrieval from SAR, oper- ating at C-band with either VV or HH-polarization (2), and demonstrate its application utilizing RADARSAT-1 data. In contrast to the previously used wind direction retrieval algo- rithms based on filtering in the spectral domain (4) here they are retrieved in the spatial domain (2), (5). The spatial domain is an important improvement, because it enables us to exclude individual areas from the wind retrieval, e.g., areas covered by surface slicks, land or sea ice. Therefore, wind fields can be retrieved in coastal areas or areas partly covered by slicks or ice. II. INVESTIGATED DATA The Canadian satellite RADARSAT-1 is positioned on a near-circular, polar and sun-synchronous orbit at a mean al- titude of 790 km. It has a repeat cycle of 24 days with an orbital period of ∼ 100 min. The satellite operates a SAR with a frequency of 5.3 GHz (C-band) and transmits and re- ceives with linear horizontal (HH) polarization in transmit and receive. In this study different RADARSAT-1 SAR modes were used, which offer a possible range of incidence angles, between 20 ◦ and 49 ◦ perpendicular to flight direction. In the ScanSARmode an area of up to 500 km is covered with a reso- lution of ∼ 100 m and in the SARmodes enable a resolution of ∼ 30 m. All utilized RADARSAT-1 ScanSAR data were pro- cessed by the Alaska SARFacility (ASF) into calibrated SAR data.