Preliminary Neural Network Detection of a Gulf Stream in Images of Sea Surface Temperature Gradients. Phase 1

摘要

A concept is tested for automatically identifying Gulf Stream surface temperature gradients among the myriad fronts discernable as edges in satellite infrared imagery. The concept is to use the techniques of neural networks in concert with a principal component climatology of Gulf Stream axes. Sample neural networks were constructed that successfully produced mode coefficients for the first three components for a large set of well defined Gulf Streams. One network, operating on a jumble of edges from a real composite sea surface temperature image, produced a 3-mode Gulf Stream sufficiently close to the actual Gulf Stream edges as to hold promise for identifying the Gulf Stream's gradients automatically. Keywords: Neural networks; Objective analysis; Satellite imagery; Image processing; Oceans graphic fronts. (edc)