13.9
Detection and Tracking of Vortices and Saddle Points from SST Data
Qing Yang, LBNL, Berkeley, CA; and B. Parvin
To improve long-term exploration of climate-change activities, we are constructing a system for automated feature extraction of significant events from extended simulation runs or satellite images. Climate modeling and corresponding satellite imaging generate massive amounts of data that pose a difficult performance and comprehension challenge for a problem solving environment (PSE). Furthermore, existing comparative analyses of climate codes are based on aggregate measures such as average temperature and rainfall, and they ignore higher-level feature activities (e.g., vortices, fronts, patches of uniform motion, pools of warm water). The proposed automated feature extraction will generate an intelligent summary of raw data and thus facilitate hypothesis testing, data mining, model validation, and efficient visualization of meaningful information (e.g., vortices and fronts and their corresponding attributes).
Ocean vortices are an important component of global circulation because they are an efficient transport and mixing mechanism for salt/freshwater, heat, plankton communities, nutrients, and momentum. Furthermore, these (compact) singularities can provide a lossy reconstruction of the original data. Intelligent data mining may occur at several scales of hierarchy. The lowest scale is at the raw image level, where we have developed an algorithm to compute the velocity of feature vectors from consecutive frames of data, followed by localization of singular events from the underlying vector field. In our current model, the velocity field is computed along the temperature gradient (perpendicular to isothermal) from SST data, while motion continuity and fluid incompressibility are used as physical constraints. Our implementation uses a pyramid representation for efficient computation of large image sizes. Additionally, we have developed an efficient algorithm to detect vortices and saddle points from underlying velocity field measurements. In the following image, the arrows show the velocity of feature vectors normal to isothermals, the underlying intensity encodes the magnitude of the feature vector, and the red dots denote the position of vortices.
Session 13, Applications of IIPS Using Satellites, Other Observation Platforms, and Their Associated Data Processing Systems (Parallel with Sessions 11 & 12)
Thursday, 13 January 2000, 10:30 AM-5:15 PM
Previous paper Next paper