Canadian Forest Service Publications
An Illustrated Methodology for Land Cover Mapping of Forests with Landsat-7 ETM+ Data. 2004. Wulder, M.A.; Cranny, J.; Dechka, J.; White, J.C. Methods in support of EOSD Land Cover. Version 3, Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, BC, Canada, March 2004, 35p.
Year: 2004
Issued by: Pacific Forestry Centre
Catalog ID: 34298
Language: English
Availability: PDF (download)
Abstract
The following document describes the methodology used for classification of Landsat ETM+ data as part of Earth Observation for Sustainable Development of Forests (EOSD). Topics include pre-classification image preparation, detailed classification procedures (including aggregation/labelling), and post-classification processing. The classification procedure involves pre-stratifying each image into four broad categories based on a normalized difference vegetation index (NDVI). Each category is then processed using an unsupervised K-means classifier. The input channels for the classification include the 6 optical Landsat-7 ETM+ channels as well as 1 texture measure. The texture is an intra-pixel variance derived from the 15m panchromatic data. The classification is labelled based on existing ground data and user knowledge. The EOSD classification legend consists of 22 land cover classes, plus one for NODATA. These classes have been adapted from those developed for the National Forest Inventory (NFI).