Ocean Currents from Successive Satellite Images: The Reciprocal Filtering Technique

Abstract

Cross-correlation analyses of successive satellite images of the ocean surface can provide estimates of surface currents by tracking features of temperature or ocean color. While the technique successfully identifies many true vectors, it also can produce a large number of ?rogue? vectors through the erroneous matching of similar-shaped surface features. Several techniques have been suggested to detect these rogue vectors and eliminate them from the analyses. These include nearest-neighbor matching, velocity filtering, and limits on the minimum cross-correlation coefficient that is acceptable. A pair of concurrent Advanced Very High Resolution Radiometer (AVHRR) images of the Leeuwin Current off the west coast of Australia is used to show that a new technique, based on reciprocity, outperforms other techniques in many situations. The reciprocal technique allows a low correlation coefficient threshold to be set thereby increasing the number of good vectors obtained while also removing those vectors that are obviously wrong. The technique also performs well in detecting small-scale eddies in which neighboring vectors can have markedly different velocities. All filtering techniques are found to perform poorly in regions where a large proportion of obviously bad vectors are present.