A New Decision Tree Classification Approach for Extracting Urban Land from Landsat TM in a Coastal City, China

Extraction of urban land is one of the necessary processes in the change detection of urban growth. In this paper, a new decision tree Classification (DTC) approach was developed to automatically extract urban land based on spectral and geographic features from Landsat TM images. The method integrates multi-spectral features such as SAVI (Soil adjustment vegetation index), MNDWI (Modified normalized water index), MNDBaI (Modified normalized difference barren index) and WI (Witness index), with geographic features including DEM and slope. The multi-feature decision tree approach achieved more than 45% higher overall classification accuracy for urban land than NDBI (Normalized difference built-up index) method when both were implemented simultaneously in Xiamen, located on southeast coast of Fujian Province, China. One reason for the improvement is that DTC approach can well extract urban areas from barren and bare land, e.g., beach, a typical landuse type of a coastal city. In addition, DTC has no assumption that a positive NDBI value should indicate a built-up area while a positive NDVI value should indicate vegetation.