Abstract: According to the remote sensing classification of forest resources, the water and soil conservation forests in Huanglong County of Shaanxi Province, a hilly and gully area of Loess Plateau, were chosen as research objects, and a mixed pixel decomposition method based on coniferous-broad leaved-grass (C-B-G) mode l was introduced for solving the mixed pixel problems in forest information extraction by  ETM⁺ remote sensing images. Using this method, the cover images of coniferous forest, broadleaved forest, bush, and grass were obtained respectively, and the distribution of coniferous and broad-leaved mixed forest was extracted. Erdas 9.1 was used to evaluate the classification accuracy. Comparing with the supervised classification, a general classification method, our method had 20% higher classification accuracy on the coniferous and broad-leaved mixed forest, suggesting its improved effect in vegetation information extraction.

Key words: Weibei dry upland, Apple planting subarea, Soil moisture characteristics, Dried soil layer