Comparison of the methods of extraction of information on wetland in inland river basin of northwest China: a case study in Zhangye city

摘要

Taking the Zhangye City in inland river basin of Northwest China as the research area and selecting TM image of Landsat 5 as data source,the wetland information were extracted by using Iterative Self-Organizing Data Analysis Technique(ISODATA),Maximum Likelihood Classification(MLC),Support Vector Machine(SVM),Artificial Neural Network(ANN)and Object Oriented Approach(OOA).Comparison of the extraction results showed that each method is suitable for extracting the wetland information.Some conclusions obtained are: 1)The overall accuracy of OOA was the maximum(75.1％)with perfect continuity.The accuracy of ISODATA was the minimum(56.67％)and many over extraction regions.2)Although the overall accuracy of MLC,SVM,and ANN was lower(63.26％,72.56％and 72.21％,respectively)than OOA,they also had reference for extraction.3)There were some differences in results of shallow water area,marshes and small ponds.The result not only of OOA but also of ANN and SVM had the same phenomenon,which gave imperfect information on patches of small ponds.Near the regions of shallow water area and marshes,some farmlands were wrongly identified as wetland by ISODATA and MLC.