Semiautomatic assignment of high resolution wind directions in SAR images

Abstract

Synthetic aperture radar (SAR) imagery has the potential of producing wind fields that are unique with respect to spatial resolution and coverage. The wind speed retrieval is performed by calibrating the SAR image and comparing the resulting normalized radar cross section (NRCS) with the NRCS from an empirical model, e.g. CMOD4. The backscatter model gives the NRCS as a function of image geometry, wind speed, and wind direction. While the image geometry and NRCS are known, estimation of wind speed by inversion of the backscatter model function still requires an a priori knowledge of the wind direction. In most investigated SAR images wind induced streaks are visible, usually having widths between 500 m and 2000 m. The method presented, defines the wind direction as normal to the gradient of the image amplitudes. The gradient is approximated by finite differences spanning an appropriate length, and the direction is determined by the best and most frequent values of the local gradients within the considered image area. The estimated directions are very close to those extracted by the visual recognition of wind streaks, but they are more often available and at a spatial resolution down to 1000 m. The now available high resolved wind directions allow the retrieval of high resolution wind fields with only a few manual interventions, and thus on a routine level.