Extraction of land cover information based on decision tree method: A case of Wuxi and Changzhou city, China

Based on the Landsat-5 TM image of May, 4 and January, 12 in 1997, this paper extracts the land cover information in Wuxi and Changzhou city which is belong to the heavily polluted area of Taihu Lake, China. Firstly, the paper collected and analyzed the spectral characteristics of main land cover type, and then determined the main indexes used in decision tree combining with the characteristics of topography. The main indexes are MNDWI, NDVI, NDBI-NDVI, TM1+TM2, TM5-TM4, TM2+TM3, K-T transformation, and slope. This paper shows that the overall accuracy of main land cover types reaches 85.21%. The area proportion of dry farm, tea plantation, waters, garden, paddy field, forest, construction land and other land is 7%, 1%, 24%, 1%, 34%, 7% 17% and 9%, respectively. The relative high error occurred in between garden and forest because their similar spectral and growth characteristics.