A Comparison of Water Indices and Binary Thresholding Techniques for Water Surface Delineation for St. Croix Watershed Area

Water indices have been successfully utilized to delineate open water surfaces from land surfaces in various landscapes. This study explores the applicability of satellite-derived indices for extracting open water surface pixels in a humid environment with wetlands as the background noise source. The St. Croix Watershed area situated in Minnesota and Wisconsin provided a good test site because of it large number of glacial lakes, wetlands with varying amount of vegetation, and a humid continental climate. The study used Landsat TM data to generate spectral water indices such as Normalized Difference Water Index (NDWI), Modified NDWI (MNDWI), Normalized Difference Moisture Index (NDMI), Tasseled Cap (TCW) and Normalized Difference Vegetation Index (NDVI) for St. Croix Watershed Area. Additionally, the study compared two different binary thresholding techniques—zero and Otsu thresholding—to separate open water from other land covers. For validity assessment, reference points from Collect Earth were used. The accuracy of the water indices was tested and compared using error matrices, kappa coefficient, and overall accuracy. Overall, MNDWI yielded the best result among all the water indices in extracting open water surface, while NDMI and TCW wetness index did poorly. There was no significant difference between the Otsu and zero thresholding methods of extracting water from land surfaces.