Landsat Etm Sub-pixel Analysis of Urban Landscape Using Fuzzy C- Means Clustering and Differentiated Impervious Surface Classes

Fuzzy c-means clustering (FCM) algorithm has been used to analyze the sub-pixel composition of medium spatialresolution satellite image (i.e., Landsat ETM). As urban landscape shows complex patterns of land cover composition and setting, it is difficult to have high accuracy in estimating urban land cover composition from Landsat image because of the mixed pixel problem. This study evaluates the utility of FCM algorithm in the subpixel analysis of Landsat image with simplified urban land cover classes: impervious surface, lawn, and woody tree. The training pixels of impervious surface are further divided into three sub-classes. The cluster center number and value of FCM is given as the number and the pure pixel spectral value of the three land cover classes. The cluster center value of FCM is defined as the median spectral value of the training pixels of each land cover class and the training pixels of impervious surface is further classified into three subclasses. The accuracy assessment is based on NJDEP LU/LC map that contains DOQQ-based impervious surface estimate value. This study shows how the FCMbased sub-pixel analysis with simplified spectral values of the training pixels estimates accurately the land cover composition of medium spatial resolution satellite image.