| Popis: |
Forest mapping over mountainous terrains is difficult because of high relief. Although digital elevation models (DEMs) are often useful to improve mapping accuracy, high quality DEMs are seldom available over large areas, especially in developing countries. In this study, a hierarchical approach coupled with topographic information derived from coarse DEM was developed to improve the efficiency and accuracy of forest mapping over mountainous areas. The overall idea of increasing mapping accuracy over large mountainous areas is to reduce spectral variety over areas to be mapped. The approach consists of three major steps. The first step is to partition a large mountainous area into several small mapping zones. Forest mapping is then conducted in each zone independently. At the second step, forest areas are separated from non-forest areas through a semi-automatic binary classification procedure. At the third step, forested areas are then further classified into detailed forest types by coupling Landsat ETM+ imagery and two topographic variables derived from a coarse DEM (extracted from 1:250,000 digital elevation contour layer, which are readily available). This hierarchical approach was illustrated and evaluated through a case study in Northwest Yunnan, China, a very rugged terrain in the world. Forests and non-forests were separated accurately and efficiently (the overall accuracy is 0.97 and Khat value is 0.94 of whole area). It was found that the inclusion of the coarse topographic data improved the mapping accuracy significantly (overall accuracy from 0.74 to 0.84, from 0.76 to 0.89, from 0.78 to 0.84 in three test areas, respectively), and that the difference in accuracy between the use of coarse DEM data and the use of fine DEM data for the study area is not significant (overall accuracy from 0.84 to 0.86). The results indicate that the hierarchical approach, coupled with coarse DEM information, is effective in increasing the accuracy of forest mapping over very rugged terrains when high quality digital elevation models are not available. |
Full Text from ScienceDirect