DEM-based modification of pixel-swapping algorithm for enhancing floodplain inundation mapping

Floodplain inundation plays a key role in riparian ecosystems. Remote sensing provides an advanced technology for detecting floodplain inundation, but the trade-off between the spatial and temporal resolutions of remotely sensed imagery is a well-known issue. Sub-pixel mapping is an effective way to mitigate the trade-off by improving the spatial resolution of image classification results while keeping their temporal resolution. It is therefore useful for improving the mapping of highly dynamic flood inundation using coarse-resolution images. However, traditional sub-pixel mapping algorithms have limitations on delineating the extent of floodplain inundation that reveals linear and complex characteristics. A modified pixel-swapping (DMPS) algorithm which is based on a digital elevation model (DEM) is thus developed in this study. It is built on the widely accepted pixel-swapping (PS) algorithm and one of its derivatives, the linearized pixel-swapping (LPS) algorithm. A Landsat image recording a significant flood inundation event in the Chowilla Floodplain of the Murray-Darling Basin in Australia was used as a case study. The results show that the DMPS algorithm outperformed the original PS and LPS algorithms both in accuracy and rationality of the resultant map. It improves the accuracy and the kappa coefficient by about 5% and 0.1, respectively, in comparison with the PS algorithm. The spatial pattern of inundation derived from the DMPS algorithm reveals fewer breakpoints and errors along the river channels. Moreover, it is observed that the DMPS algorithm is less sensitive to some critical parameters compared with the PS and LPS algorithms. It is hoped that the proposed DMPS algorithm will broaden the application of coarse-resolution sensors in floodplain inundation detection, which would thereby benefit the ecological studies in floodplains.