Extracting impervious surfaces from multi-source satellite imagery based on unified conceptual model by decision tree algorithm

Abstract

Extraction of impervious surfaces is one of the necessary processes in urban change detection. This paper derived a unified conceptual model (UCM) from the vegetation-impervious surface-soil (VIS) model to make the extraction more effective and accurate. UCM uses the decision tree algorithm with indices of spectrum and texture, etc. In this model, we found both dependent and independent indices for multi-source satellite imagery according to their similarity and dissimilarity. The purpose of the indices is to remove the other land-use and land-cover types (e.g., vegetation and soil) from the imagery, and delineate the impervious surfaces as the result. UCM has the same steps conducted by decision tree algorithm. The Landsat-5 TM image (30 m) and the Satellite Probatoire d’Observation de la Terre (SPOT-4) image (20 m) from Chaoyang District (Beijing) in 2007 were used in this paper. The results show that the overall accuracy in Landsat-5 TM image is 88%, while 86.75% in SPOT-4 image. It is an appropriate method to meet the demand of urban change detection.