Knowledge-based imaging spectrometer analysis and GIS for forestry

In 1993 and 1994, AVIRIS images were taken by NASA's ER-2 over test sites in British Columbia as part of the SEIDAM (System of Experts for Intelligent Data Management) project. To support these acquisitions, extensive ground measurements were made including calibration with ground-based spectrometers and uniform targets. A method for calibration of the AVIRIS sensor has been developed which uses these ground spectra and a JPL-modified MODTRAN 2 radiative transfer model. The ingest, calibration and geocoding of AVIRIS imagery is a sophisticated task for which expert systems have been constructed. The geocoded AVIRIS imagery are interfaced to a GIS and attribute data base. Expert systems generate spectral end member images, atmospheric images, band moment images, and principal component images. SEIDAM contains a meta data database for imagery and GIS files. The system uses case-based methods to solve user-specified goals and product selection, In analyzing AVIRIS imagery, SEIDAM follows the analysis paths which a domain expert has taught it or selects alternative paths based on similar training cases. SEIDAM controls the image analysis software, the GIS software, and the database software. Analysis results are available as GIS file updates, tables, images, or visualizations, SEIDAM uses an expert system shell written in PROLOG and runs on a SUN computer. It controls software on SUN, SGI, and VAX computers. The paper describes the expert system methodology and the AVIRIS examples for scenes over the 15 km by 23 km Greater Victoria watershed test site.