Spectral matching based on discrete particle swarm optimization: A new method for terrestrial water body extraction using multi-temporal Landsat 8 images

Abstract

Terrestrial water, an important indicator of inland hydrological status, is sensitive to land use cover change, natural disaster and climate change. An accurate and robust water extraction method can determine the surface water distribution. In this paper, a new method, called the spectrum matching based on discrete particle swarm optimization (SMDPSO) is proposed to recognize water and nonwater in Landsat 8 Operational Land Imager (OLI) images. Only two parameters, the standard water spectrum and the tile size, are considered. These parameters are sufficiently stable so it is unnecessary to change their values for different conditions. By contrast, in supervised methods, samples are chosen based on conditions. Eight test sites covering various water types in different climate conditions are used to assess the performance relative to that of unsupervised and supervised methods in terms of overall accuracy (OA), kappa coefficients (KC), commission error (CE) and omission error (OE). The results show that: (1) SMDPSO achieves the highest accuracy and robustness; (2) SMDPSO has lower OE but higher CE than the supervised method, which means that SMDPSO is the least likely to misclassify water as nonwater, but is more likely to misclassify nonwater as water; (3) SMDPSO has advantages with respect to removing shallows and dark vegetation and preserving the real distribution of small ponds, but cannot recognize shadows, ice, or clouds without the help of other data such as DEM. In addition, a case of flooding in northeastern China is analyzed to demonstrate the applicability of SMDPSO in water inundation mapping. The findings of this study demonstrate a novel robust, low-cost water extraction method that satisfies the requirements of terrestrial water inundation mapping and management.