ABSTRACT
Since the optical images will be combined with the SAR images to estimate impervious surfaces in the following chapters, the selection of optical images
for this chapter was conducted according to the availability of SAR images in the corresponding study area. Similar to the study in Chapter 4, five land use types were defined to conduct the classification procedure according to the landscape of the study area. These land cover types include water (WAT), vegetation (VEG), bare soil (SOI), dark impervious surfaces (DIS), and bright impervious surfaces (BIS). Moreover, in order to analyze the impacts of classification methods to better understand the real spectral confusions in tropical and subtropical areas, two popular machine-learning methods, ANN and SVM, were employed to conduct the LULC classification. The training and test samples were the same sets used in Chapter 4. The confusion matrix was used as the main tool to investigate the spectral confusion between various land cover types. Additionally, we assume that the collected training and test samples had some unavoidable errors, and in order to eliminate the influence of these errors, the analysis and discussion will focus on the misclassifications of more than 10 pixels between two land cover classes.
