RULE-BASED GEOBOTANICAL CLASSIFICATION OF TOPOGRAPHIC, AEROMAGNETIC, AND REMOTELY-SENSED VEGETATION COMMUNITY DATA

Classifications of rock-type in areas with closed-canopy forests using remotely sensed data must rely on geobotanical associations. However, the influence of substrate on vegetation communities is generally rat her limited, and therefore classifications that are based on geobotani cal associations can be improved greatly by including geomorphological and community spatial relationships, particularly those derived from digital elevation data. In a remote sensing study of Quetico Provincia l Park, Ontario, Canada, digital topographic data was used to divide t he landscape into four microclimate and drainage classes. The Landsat Thematic Mapper data were transformed into the nPDF Deciduous Forest I ndex, which is an estimate of the deciduous-coniferous mixture in each pixel. This was supplemented by digital aeromagnetic data and geologi cal field mapping. Associations of the vegetation communities with geo logy, microclimate, and drainage classes were identified from these da ta, and then used in a rule-based geobotanical classification called T OPOVEG. TOPOVEG achieved an 85% accuracy in a classification of the 10 km x 13 km primary test site. A standard maximum likelihood classific ation of the same area had an accuracy of only 71%, and produces an ou tput that has a distinctive noisy texture compared to the large, homog eneous classes of TOPOVEG. In a neighboring test site to the north, TO POVEG accuracy was similarly high (86%). A neighboring southern test s ite, however, showed changing vegetation associations and consequently a lower accuracy. This suggests that the classification can be extend ed to neighboring unknown areas, except to the south where the classif ication rules should be modified to take into account the changing veg etation associations. Although TOPOVEG was developed for geobotanical exploration, the procedure has potential for investigating ecological community associations, and the patterns of those associations.