A novel spatial–spectral extraction method for subpixel surface water

Abstract

ABSTRACTThe spectral index provides a feasible and efficient method for extracting land surface water from satellite images. However, existing methods at the pixel scale consider only water bodies but disregard complicated mixtures. Emerging subpixel–scale methods improve water mapping a detailed description, but they still fail to achieve precise identification of various types of mixed pixels. This deficiency restricts the further development and application of the spectral index method. In this study, we proposed a novel spatial–spectral extraction method that utilizes double thresholds in describing pixels at the subpixel scale. This method combines spatial and spectral attributes to refine and decompose mixed water–land pixels. In this manner, a stepwise approach is adopted to achieve the optimal result instead of the common single–judging method. The experiments use moderate spatial–resolution images, such as those taken by Landsat–8 Operational Land Imager, GF–1’s wide field view multispectral camera and the Sentinel–2A satellite. Results confirm that the proposed method can be applied to common moderate spatial–resolution satellite images. This method achieves better accuracy than existing methods. The insensitivity of threshold selection ensures that the proposed method maintains high precision in actual applications.