Application and comparison of advanced supervised classifiers in extraction of water bodies from remote sensing images

Water body extraction plays an important role in monitoring and assessing the existing water resources. It is a complex process that may be affected by many factors. This paper examines the major and advanced supervised classification approaches and ventures into the effectiveness of these techniques in extraction of water bodies from satellite images. The different classification techniques used for this purpose include support vector machine, artificial neural network, K-nearest neighbor, discriminant analysis and random forest. Commonly used normalized difference water index technique has also been examined in the study. Comparisons have been drawn among various variants of these methods and the accuracy in each case has been recorded. Each classification technique has been applied on input images from three different satellite sensors of varying spatial and spectral resolution, to compare their performance on different data sets of three different study areas. The study has found that supervised classifier can extract water bodies with a good accuracy from remotely sensed images even with a fewer number of labeled samples. Additionally, it is seen that the linear classifiers also yield good accuracy in extracting water bodies across different sensor’s data.