Probabilistic River Water Mapping from Landsat-8 Using the Support Vector Machine Method

River water extent is essential for river hydrological surveys. Traditional methods for river water mapping often result in significant uncertainties. This paper proposes a support vector machine (SVM)-based river water mapping method that can quantify the extraction uncertainties simultaneously. Five specific bands of Landsat-8 Operational Land Imager (OLI) data were selected to construct the feature set. Considering the effect of terrain, a widely used terrain index called height above nearest drainage, calculated from the 1 arc-second Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), was also added into the feature set. With this feature set, a posterior probability SVM model was established to extract river water bodies and quantify the uncertainty with posterior probabilities. Three river sections in Northwestern China were selected as the case study areas, considering their different river characteristics and geographical environment. Then, the reliability and stability of the proposed method were evaluated through comparisons with the traditional Normalized Difference Water Index (NDWI) and modified NDWI (mNDWI) methods and validated with higher-resolution Sentinel-2 images. It was found that resultant probability maps obtained by the proposed SVM method achieved generally high accuracy with a weighted root mean square difference of less than 0.1. Other accuracy indices including the Kappa coefficient and critical success index also suggest that the proposed method outperformed the traditional water index methods in terms of river mapping accuracy and thresholding stability. Finally, the proposed method resulted in the ability to separate water bodies from hill shades more easily, ensuring more reliable river water mapping in mountainous regions.