Jump to Main Content
Coupling the Modified Linear Spectral Mixture Analysis and Pixel-Swapping Methods for Improving Subpixel Water Mapping: Application to the Pearl River Delta, China
- Liu, Xulong, Deng, Ruru, Xu, Jianhui, Zhang, Feifei
- Water 2017 v.9 no.9
- Landsat, algorithms, disasters, flooded conditions, monitoring, remote sensing, river deltas, rivers, China
- High-resolution water mapping with remotely sensed data is essential to monitoring of rainstorm waterlogging and flood disasters. In this study, a modified linear spectral mixture analysis (LSMA) method is proposed to extract high-precision water fraction maps. In the modified LSMA, the pure water and mixed water-land pixels, which are extracted by the Otsu method and a morphological dilation operation, are used to improve the accuracy of water fractions. The modified LSMA is applied to the 18 October 2015 Landsat 8 OLI image of the Pearl River Delta for the extraction of water fractions. Based on the water fraction maps, a modified subpixel mapping method (MSWM) based on a pixel-swapping algorithm is proposed for obtaining the spatial distribution information of water at subpixel scale. The MSWM includes two steps in subpixel water mapping. The MSWM considers the inter-subpixel/pixel and intra-subpixel/subpixel spatial attractions. Subpixel water mapping is first implemented with the inter-subpixel/pixel spatial attractions, which are estimated using the distance between a given subpixel and its surrounding pixels and the water fractions of the surrounding pixels. Based on the initialized subpixel water mapping results, the final subpixel water maps are determined by a modified pixel-swapping algorithm, in which the intra-subpixel/subpixel spatial attractions are estimated using the initialized subpixel water maps and an inverse-distance weighted function of the current subpixel at the centre of a moving window with its surrounding subpixels within the window. The subpixel water mapping performance of the MSWM is compared with that of subpixel mapping for linear objects (SPM<inf>L</inf>) and that of the subpixel/pixel spatial attraction model (SPSAM) using the GF-1 reference image from 20 October 2015. The experimental results show that the MSWM shows better subpixel water mapping performance and obtains more details than SPM<inf>L</inf> and SPSAM, as it has the largest overall accuracy values and Kappa coefficients. Furthermore, the MSWM can significantly eliminate the phenomenon of jagged edges and has smooth continuous edges.