on a formal stratified random field checking methodology overall map accuracy was 81%. Map accuracy is raised to 93% if the 6 agricultural classes are lumped into a single agriculture class.
Error matrices were generated for each county, along with an analysis of each class accuracy and error type. Vegetation/land use class accuracy (the number of correctly classified sample points for each class divided by the total points sampled for each class) was determined on a county and complete map basis. An overall map accuracy of 78% was achieved for the northern half of the Coast Range Ecoregion (Barrett 1998), using aerial videography as accuracy assessment protocol. Accuracy assessment was conducted only within the Coast Range and Willamette Valley Ecoregions. Lack of funding precluded accuracy assessment throughout the rest of the state.
problem of signature extension severely compromises classification effort. Ecoregional partitioning reduces spectral complexity displayed in a full TM scene, and groups vegetation types into more Aprobable@ associations. 2) Construct derivative bands. A normalized difference vegetation index, NDVI, and the first three principal component bands of a Tasseled Cap Transformation algorithm was incorporated with TM bands 1-5 and 7 to form a 10 band image. This image will be the basis of all subsequent spectral analysis. 3) Conversion of TM imagery to TIFF format files. A three band (bands 3, 4 and 5) image was subset from the 10 band image and converted to a TIFF which was downloaded to a lap top computer for field reconnaissance purposes. 4) Conversion of vector format ancillary data. Coverages which assist the analyst during field verification, especially the road and stream networks were converted to a DXF format and brought into the lap top computer to display over the TIFF images.