Hydrophytes extraction in Taihu Lake, China: an approach of integrating decision tree with water depth based on Landsat TM and SPOT

摘要

When multispectral images are used to extract the area of aquatic vegetation in Taihu Lake, because of the influence of suspended matter and algae, different objects may have the same spectrum and make it difficult to mapping the distribution of aquatic vegetation exactly. Many different methods, including unsupervised classification and supervised classification, are used, but the classification accuracy didn't improve obviously. The growth of aquatic vegetation is closely to the water depth. So we try to use water depth data to increase the extraction accuracy. The whole Taihu Lake is classified into three types: open water, emerged vegetation and submersed aquatic vegetation. Suppose the DN (Digital number) of each type satisfies normal distribution. Numbers of sample points of each type in single band or combined bands are selected and put down there DNs, and then statistical method is adopted to acquire the maximum and minimum which are used to build decision tree to fulfill the classification. The single band or combined bands in which maximum and minimum interval of each type have small intersect set are considered as the suitable bands for classification. Two methods, classification based on spectral characteristics and classification based on spectral characteristics and water depth data, are used. The classification accuracies of the two methods are compared. The results show the water depth data can improve the classification accuracy and resolve the different objects with same spectrum problem partially.