CWO Data Mining

Abstract

The Navy Research Laboratory's Coastal Ocean Model (NCOM) is a realistic, large-scene simulation that runs daily and generates massive amounts of data. The data must be analyzed and/or reduced to provide pertinent information. This may be achieved through data mining by performing feature detection and/or region-of-interest detection. Data reduction using data mining techniques is not a new idea, especially when the objects of interest are ocean eddies. There are on the order of 20 methods to "data mine" for eddies. However, no one method has been tested on all models, few have been tried on multiple models or model types, and different methods require different data fields (e.g., salinity, temperature, horizontal velocity, vorticity). Our objective was to examine the most attractive eddy detection methods for NCOM and then determine which method provides the best results. We implemented and evaluated two eddy detection methods for NCOM data. The first is an algorithm created at Mississippi State University, which utilizes critical points in ocean flow. The algorithm was developed for the Navy Research Laboratory's Layered Ocean Model (NLOM) and performed well. The second algorithm is based on the Marr-Hildreth edge detection. We evaluated our results by comparing the detected eddy locations to eddies identified in ocean color from SeaWiFS in the northwestern Arabian Sea and Gulf of Oman