Water body extraction from LANDSAT ETM+ image using MNDWI and K-T transformation

Abstract

Rapidly and accurately extracting water body information from remote sensing satellite images is critical for the investigation of water resources, planning watershed, dynamically monitoring, water environment and water conservation, etc. With an image of Landsat-7 ETM+ in Lushan Mountain region, the spectral characteristics of water body and other main surface features were analyzed in the paper. A method of modified normalized difference water index (MNDWI) combined with K-T transformation was proposed to distinguish water body from mountain shadow. First, MNDWI of original image was calculated and the threshold values were set across the interval (0.25, 0.8). Then the original image was processed by K-T transformation method, and the pixels were extracted using the third component (wetness) threshold value of less than -80. Finally, the two images were added together after mask processing, thus water information was identified. The results showed that this method of combing MNDWI with K-T transformation could not only remove the influence of mountain shadow of MNDWI method alone, but also avoid the confusion of water and buildings existing in image after K-T transformation. The extraction accuracy of water body was up to 93%, which suggests that this approach has better performance than some previous methods such as single-band threshold technique and the spectrum-photometric method.