On Determination of Wind Vectors for C-band SAR Will Perrie (1) , Yijun He (2) , and Hui Shen (1,2) (1) Bedford Institute of Oceanography, 1 Challenger Dr., Dartmouth Nova Scotia, Canada (2) Ocean Remote Sensing Lab., Inst. Oceanology, Chinese Acad. Sci., Qingdao, China Abstract In this paper, we propose a gradient method (GM) to estimate wind vectors from SAR images. Wind speed and direction are estimated using the normalized radar cross sections (NRCS) from 2 neighboring subimage blocks, having slightly different incidence angles. 1. INTRODUCTION Ocean wind speed and wind direction are estimated simultaneously using the NRCSs corresponding to two neighboring (25-km) blocks, within a given SAR image, having slightly different incidence angles, motivated by the methodology used for scatterometer data. The wind direction ambiguity is removed by using the direction closest to that given by a buoy or some other source of information. We demonstrate this method with ENVISAT ASAR and RADARSAT SAR images from coastal waters of the North Atlantic. Estimated wind vectors are compared with wind measurements from buoys and scatterometer data. We show that this method can surpass other methods, even those with notable visible wind-induced streaks in SAR images, to extract wind vectors. The method is described in Section 2, and discussion and conclusions in Section 3. 2. METHODOLOGY The CMOD-IFR2 [1] geophysical model function for VV polarization images from the ASAR (Advanced Synthetic Aperture Radar) aboard the ENVISAT [2], which may be expressed in terms of ) , , ( 0 u V φ θ σ , where u is the wind speed at 10 m reference height, φ is the angle between the radar look direction and the wind direction, and θ is the incidence angle. A scatterometer can measure several NRCSs 0 σ , with different look directions, for a given location. Moreover, some instruments like QuikSCAT take 0 σ measurements at different incidence angles and polarizations. Because ASAR only measures NRCS 0 σ in the radar look direction, it is impossible to use a scatterometer methodology to retrieve wind vectors from ASAR images. Thus, we divide a given ASAR image into N by M blocks along the radar look direction and the platform flight direction, respectively. As 0 σ depends on wind speed and direction, wind vectors can be retrieved by using the 0 σ s from two neighboring blocks, along the radar look direction, having slightly different incidence angles. In order to apply the ERS satellite C band scatterometer GMF model, the size of each block is 25 25 × km 2 . Therefore an ASAR image can be represented in terms of overlapping sub-images, each of which consists of two neighboring blocks along the radar look direction, where each sub-image is also partially covered by the next sub-image. For a given ASAR image, M N × − ) 1 ( wind vectors result, giving the wind speeds and directions for all sub-images. Sub-images may be labeled ) , ( j i , where i is the index for the coordinates of the N-1 radar look directions and j is the index for the M platform flight directions, with 1 , , 1 − = N i L , and M j , , 1 L = . For each sub-image ) , ( j i , we define a cost function ( ) ( ) 2 0 2 2 2 0 1 1 ) , 1 ( ) , 1 ( ) , ( ) , ( ) , ( j i j i j i j i j i J m m + − + + − = σ σ σ σ (1) consisting of blocks that we represent symbolically as “1” and “2”, where 0 σ and m σ are NRCS observed and calculated backscatter images using the CMOD_IFR2 model. In this methodology, we assume the wind vectors of the two adjacent blocks along radar look direction are equal, while ) , ( 1 j i m σ and ) , ( 2 j i m σ are calculated using the CMOD_IFR2 model. For each sub-image, the cost function is minimized by requiring that the partial derivatives with respect to wind speed ) , ( j i u and wind direction )) , ( cos( j i φ be zero, which determines the set of wind vectors for the sub-images constituting the ASAR image. The difference in incidence angles for two neighboring blocks, making up a given sub-image, is normally bigger than 1 o , for ENVISAT ASAR images. Assuming the same wind vector for the two