A multi-subregions decision tree land cover classification approach using Landsat8 image

Coal fires burning caused serious environmental, economic and safety catastrophe in Wuda district, North China. The land cover change research helped to evaluate the extent of coal fire damage. The image data of Landsat8 satellite offered the possibility of detecting and studying land cover/use in coal fire area. Five subregions were divided from one Wuda image based on topographic, landform and land surface radiation characteristics. Corresponding to each subregion, five different decision tree models with different parameters were respectively constructed based on a general sole decision tree for the whole research area, which was built by spectral characteristics analysis, height, slope and infrared information. By contrasting with a general sole decision tree and other four common classification methods applied to the whole area, land cover accuracy of multi-subregions decision tree classification approach derived higher overall accuracy(87.63%) and Kappa coefficient(0.86) because subregions decreased land-cover confusions. In particular, the accuracy of building and coal ash classification mapping showed a marked increase.