EOS Data Visualization Using Self-Organizing Map

Description and Objectives:

• Develop a visualization system for multispectral and multivariant EOS datasets, sponsored by JPL 2003 R&TD fund.
• Apply the Self-Organizing Map (SOM) algorithm to reduce a multi-dimensional EOS dataset (e.g. MISR, Multi-angle Imaging SpectroRadiometer) into 3D color space, then use Voronoi Diagram to map the SOM image into a smaller set of colors (i.e. clusters). Validate the effectiveness of the proposed algorithm.
• Build a prototype interactive visualization system using the algorithm developed above.

Examples:

(a) Nadir Image	(b) SOM Image (1000 colors)	(c) Voronoi Diagram (8 colors)

MISR (Multi-angle Imaging SpectroRadiometer) data contains 36 channels (9 cameras each with different viewing angle and four bands per camera). (a) is a MISR image showing three visible bands from the Nadir camera (looking straight down). (b) is a 1000 color false color image generated by SOM that contains all 36 channels of information. (3) is a 8 color image generated by applying Voronoi refinement to (2). You can see clear separation between land (blue and purple) and cloud (green).

The light-blue feather shape feature in yellow circle (in both (b) & (c)) is not visible in (a). In fact, it only shows up faintly in the most forward looking camera. It appears to be a high altitude thin Cirrus cloud. This demonstrates that the SOM algorithm can extract information that is not obviously observable by other visualization methods.

Schedule and Deliverables:

• FY 03: Prototype implementation of the SOM and Voronoi refinement algorithm. Validate the prototype using selected MISR data and benchmark the performance of the algorithm.
• FY 04: Build an interactive visualization system and demonstrate it to MISR science team and other EOS project personnel.
• FY 05: Select a specific science problem and solve it with the system developed in FY04. Test the system using other EOS datasets.

Status: