Assessing methods of identifying open water bodies using Landsat 8 OLI imagery

Abstract

Remote sensing is an effective technology for monitoring water resources. However, many methods in remote sensing imagery used to identify open bodies of water have often been shown to produce varying water body classification results for the same bodies of water. Therefore, it is necessary to have measures to assess these methods. In this study, we used datasets collected in the field to assess methods for identifying open water bodies using images from the Landsat 8 Operational Land Imager. From this, we clarified the difference in the performance between the use of spectral reflectance images and that of digital number (DN) value images for the classification of water bodies. The results showed that the normalized difference water index (NDWI), calculated using green and near-infrared bands (NDWIGreen/NIR) with reflectance, captured correct control points with an accuracy of greater than 95 % and was therefore the superior method. The result of a comparison in performance in terms of the NDWI between reflectance images and DN value images was consistent with their initial definitions. The NDWI indices calculated by the initial definitions yielded more reasonable results in the classification of water bodies. The optimized threshold, calibrated and validated by 737 field control points, generated water classification results with a higher confidence in this study. We think that it might be better to set the optimized threshold of NDWIGreen/NIR to −0.05 instead of the value of zero used in many studies. However, more optimized thresholds for other regions need to be calibrated and confirmed if data are available. Our results indicated that NDWI methods are more suitable for water body classification than single-band methods when the frequency histogram method is used.