Abstract
Boreal forest, which covers 15% of the Earth's land area, is subjected to disturbances such as fires and logging. Because such forest is covered by snow during the long winter, the window of opportunity for observation in summer, when the trees are fully foliated, is short. The use of snow-covered Landsat images for forest change detection was studied as alternative to the use of summer images. A time series of five Landsat images taken between 1980 (MSS) and 1999 (TM) in the Russian Far East was used. Changes were detected by comparing certain indices in successive images. The indices were RED, normalized RED (N-RED) NIR, MIR, NDVI, NDSI, Tasseled Cap Brightness, Greenness and Wetness and SAITO and YAMAZAKI's V2 and S3. Indices from summer images taken in 1995 were compared with indices from winter images. The majority of winter indices showed points of change for much longer periods after disturbance than did the summer indices, although the winter indices derived by deviation were less stable than the summer indices. Winter RED, N-RED and Wetness were particularly suitable for change detection, while summer Wetness was suitable for estimating the succession stage after disturbance. The detection accuracies were 66% to 100%.
