Finding Mesoscale Ocean Structures with Mathematical Morphology. (Reannouncementwith New Availability Information)

摘要

We introduce a technique to aid in interpreting infrared satellite images of theNorth Atlantic Ocean Gulf Stream region. Present interpretive methods are largely manual, require significant effort, and are highly dependent on the interpreter's skill. Our quasiautomated technique is based on mathematical morphology, specifically the image transformations of opening and closing, which are defined in terms of erosion and dilation. The implementation performs successive openings and closings at increasing thresholds until a stable division into objects and background is found. This method finds the North Wall of the Gulf Stream in approximately the same place as human analysts and another automated procedure, and does less smoothing of small irregularities than the other two methods. The North Wall is continuous and sharp except where obscured by clouds. Performance in locating warm-core eddies is also comparable to the other methods. However, the present procedure does not find cold-core rings well. We are presently investigating ways to reduce the effects of clouds and delete the unwanted water areas found by the method. We expect to be able to improve the cold-core eddy performance.... Remote sensing, Artificial intelligence, Data assimilation, Satellite data.