Estimation of wind vectors from Synthetic Aperture Radar (SAR) imagery has been on-going for some time, and various techniques have been developed by researchers to automatically estimate wind direction using spectrum- or gradient-based approaches. The authors have developed and implemented a spectrum-based approach and have compared its performance versus buoy observations for both the C-VV ERS satellite SAR sensor and the C-HH RADARSAT satellite SAR sensor. Wind directions errors (RMS) from 20 to 37 degrees and wind speed errors (RMS) from 1.2 to 2.0 meters per second have been shown using various validation data sets.

Accurate estimation of wind vectors from SAR imagery requires some ocean surface feature aligned with the local wind that the algorithm can use for estimating direction. A large difficulty in utilizing SAR imagery operationally for wind vector estimation is knowing when to ignore the SAR-derived vector because the image subset does not contain any information about the local wind direction; i.e. does not contain any surface features that are aligned with the wind. Operationally accuracy could be significantly increased if only those "good" vectors were used to generate an interpreted field across the whole image.

In this paper, we summarize the work done previously to estimate wind vectors from SAR imagery and validate the performance. Then we present a new algorithm that utilized metrics based on image features to generate a confidence metric for each wind vector that allows us to determine whether a vector is "good" or not. We show comparisons with QuickScat winds as well as buoy observations to validate the performance, and we show comparisons between the SAR-derived fields and model results.