BOSTON UNIVERSITY GRADUATE SCHOOL OF ARTS AND SCIENCES Thesis BIOME LEVEL CLASSIFICATION OF LAND COVER AT CONTINENTAL SCALES USING DECISION TREES by ALEXANDER LOTSCH

Land cover plays a key role in terrestrial biogeochemical processes. Therefore many problems require accurate information on the distribution and properties of land cover. A decision tree classi cation algorithm is used to generate a land cover map of North America from remotely sensed data with 1 km resolution in a 6-biome classi cation scheme. To do this, the normalized di erence vegetation index (NDVI) data from the Advanced Very High Resolution Radiometer (AVHRR) is used in association with ancillary data sources. Training sites required for this approach were generated by the Boston University Land Cover and Land-Cover Change Research Group and improved in ve preprocessing steps. Accuracy assessment of the map produced via decision tree classi cation yields a site-based map accuracy of 73%. The map is compared with maps generated from the same data, but classi ed using the International Geosphere Biosphere Program (IGBP) classi cation scheme. Biome classes are mapped with approximately 5% higher overall accuracies than IGBP classes. The biome map will be useful for remote sensing-based retrievals of leaf area index (LAI) and the fraction of absorbed photosynthetically active radiation