Mapping forest type using Landsat TM images encounters many problems especially when applied in montane landscapes with complex terrain. In this paper we evaluated the effects of selected data inputs and classification methods on the accuracy of forest type mapping in a complex terrain landscape in mountainous southwest China. Results show that the accuracy of a forest type map produced by the original Landsat TM bands data alone is not acceptable, but the integration of topographic data with Normalised Difference Vegetation Index (NDVI) and Principle Components (PCs) improves the mapping accuracy by 15% and 14%, respectively. In addition, the comparison of two-classification methods showed that a GIS expert system (EXPERT) outperforms the maximum likelihood classifier (MLC) by 9%. It is concluded that combination of topographic data together with NDVI or PCs enable production of more reliable and accurate forest maps in landscapes with complex terrain. Where reliable field knowledge is available, expert systems show potential for producing affordable forest type maps as accuracy as those obtained by conventional classifiers.
|Number of pages||13|
|Publication status||Published - Dec 2007|
- forest type mapping
- GIS expert system
- matsutake mushroom habitat
- maximum likelihood classifier