Shanghai urban wetland extraction and classification with remote sensed imageries based on a decision tree model

Urban wetland is an important ecological basis of Shanghai and it is characterized with complex properties. In this study, a decision tree based classification method is used to extract and classify the urban wetland information in Shanghai area. The method uses multispectral bands of Landsat-5 TM image as the main variables, and a series of derivative data as the auxiliary inputs, derived from the Landsat-5 TM images by using respectively K-T transformation, IHS transformation, principal component analysis and textural analysis. With these variables in association with the spatial characteristics of the urban wetland in Shanghai, the method builds a decision tree model for urban wetland extraction and classification. The application of the model shows that the total area of the urban wetland in Shanghai is about 1277.40 km². The rice cultivated area occupies the highest portion up to 65.30% of the total wetland, and the next the area of rivers, ponds, lakes and reed fields. The decision tree model based method has a relative high precision in the urban wetland extraction and classification. The classification result indicates that the overall accuracy reaches 89.05%, more than 10% increase when compared with the maximum likelihood algorithm.