RAPID INVENTORY OF INLAND BODIES OF WATER USING LANDSAT TM IMAGERIES AND DIFFERENT FEATURE EXTRACTION TECHNIQUES

ABSTRACT: A good initiative for creating a national database of inland wetlands is identifying possible locations for these water bodies. Four procedures in the detection of water bodies were tested using Landsat TM Imagery and were evaluated to determine which procedure best serves the purpose of efficiently creating an initial wetland inventory for the Philippines. The procedures are namely: classification using Perceptron Model, supervised classification using Maximum Likelihood, unsupervised classification using Iterative Self-Organizing Data Analysis Technique (ISODATA), and classification by thresholding. All four procedures detected major wetlands including lakes and dams, and they also detected areas that are considerably wet such as swamps, fishponds, and agricultural areas. Due to the limited resolution of Landsat TM, small and narrow wetlands were not included in the initial wetland inventory. Considering speed and visual accuracy, classification by thresholding was chosen as the best method to use in creating the initial wetland inventory. The procedure also requires minor supervision from user and therefore minimizes errors coming from the user. As an accuracy check, wetland point locations obtained from the Biodiversity Management Bureau (BMB) were used as reference to validate the created wetland inventory. Out of the 27 wetlands in the database of BMB that lie on the study area, at least 24 wetlands were delineated in the inventory. Upon creating the initial wetland inventory, this will be used to evaluate identified areas for a detailed delineation of wetlands using an established method using Object-Based Image Analysis (OBIA)