River network delineation from Sentinel-1 SAR data

Abstract In many regions of the world, especially in developing countries, river network data are outdated or completely absent, yet such information is critical for supporting important functions such as flood mitigation efforts, land use and transportation planning, and the management of water resources. In this study a new method was developed for delineating river networks using Sentinel-1 imagery. Unsupervised classification was applied to multi-temporal Sentinel-1 data to discriminate water bodies from other land cover types then the outputs were combined to generate a single persistent water bodies product. A thinning algorithm was then used to delineate river centre lines which were converted into vector features and built into a topologically structured geometric network. The complex river system of the Niger Delta was used to compare the performance of the Sentinel-based method against alternative freely available waterbody products from USGS, ESA and OpenStreetMap and a river network derived from a SRTM DEM. From both raster-based and vector-based accuracy assessments it was found that the Sentinel-based river network products were superior to the comparator data sets by a substantial margin. The resulting geometric river network was used to perform flow routing analysis which is important for a variety of environmental management and planning applications. The approach developed in this study holds considerable potential for generating up to date, detailed river network data for the many countries globally where such data are deficient.