HR: 0830h
AN: NG11A-0174    [PDF]
TI: Localization of Oceanic Fronts and Feature Boundaries Using a Variational Technique
AU: * Sun, W
EM: waltsun@mit.edu
AF: Massachusetts Institute of Technology Stochastic Systems Group (SSG), 77 Massachusetts Ave Room 35-425, Cambridge, MA 02139 United States
AU: Cetin, M
EM: mcetin@mit.edu
AF: Massachusetts Institute of Technology Stochastic Systems Group (SSG), 77 Massachusetts Ave Room 35-425, Cambridge, MA 02139 United States
AU: Thacker, W C
EM: Carlisle.Thacker@noaa.gov
AF: The Atlantic Oceanographic and Meterological Laboratory (AOML) of the National Oceanic and Atmospheric Administration (NOAA), 4301 Rickenbacker Causeway, Miami, FL 33149 United States
AU: Chin, T M
EM: mike.chin@jpl.nasa.gov
AF: Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109 United States
AU: Willsky, A S
EM: willsky@mit.edu
AF: Massachusetts Institute of Technology Stochastic Systems Group (SSG), 77 Massachusetts Ave Room 35-425, Cambridge, MA 02139 United States
AB: Automatic localization of curvi-linear features (boundaries), including oceanic fronts and contours of rings such as those associated with the free-jet portion of the Gulf Stream, from satellite sea surface temperature (SST) maps is a challenging task, especially in the case of missing observations due to cloud cover. Having this as motivation, we explore whether techniques successfully used for non-oceanographic problems can be beneficial in the realm of oceanography. The goal is to apply a generalized version of the Mumford-Shah functional, a variational technique used in photographic and medical imaging applications, to an oceanographic problem. This method performs optimal smoothing jointly with localization of the feature boundaries. The feature boundary partitions the region into two or more subregions. Minimally, it establishes the location of the front. When there are rings, the boundary also separates the interior and exterior of these rings. As a by-product of our boundary localization, we estimate the field, which interpolates across areas of missing data, but maintains the discontinuity at the boundary. Optimal interpolation is commonly performed in data analysis and assimilation; however, the technique presented by us is distinctive in the sense that it incorporates information about the feature boundaries into the field estimation process. We start with an objective functional $E(f,\vec{C})$ which we wish to minimize over both the SST field $f$ and the boundary $\vec{C}$, which need not be a single, connected curve. The functional has a data fidelity term, a field smoothness term, and a curve length penalty term. The curve is evolved using level set methods, which can handle changes in the topology of the boundary. This allows us to locate both rings and the front. The boundary is then identified by the discontinuity, or departure from smoothness, in the estimated field. We present experimental results of our technique on various satellite observations of SST data. Preliminary results show reasonable localization of a particular oceanic front and associated rings.
DE: 4200 OCEANOGRAPHY: GENERAL
DE: 4299 General or miscellaneous
DE: 4500 OCEANOGRAPHY: PHYSICAL
DE: 4528 Fronts and jets
SC: Nonlinear Geophysics [NG]
MN: 2003 Fall Meeting