This study proposes a new river bank (RB) detection methodology for Landsat satellite data in the context of the Vietnamese Mekong Delta (VMD). Additionally, an algorithm for the calculation of the surface change between dierent data sets is presented. The new method is developed by modifying and combining known concepts in order to be eective in the complex environment of the VMD. In a rst step, the performance of ve existing spectral indices (NDWI; MNDWI; WNDWI; AWEInsh; NDVI) combined with three dierent segmentation methods (default thresholding; Otsu's method; mathematical morphological (MM) techniques) for the detection of the RB position is analyzed. The study area covers a 20 km long segment of the Mekong river including two large islands around Cao L~anh in the upper part of the VMD. As reference, geometrically corrected Google Earth data sets are used. The best performing RB detection method is identied and in a second step, the benet of pixel unmixing techniques (e.g. spectral mixture analysis (SMA)), to achieve sub-pixel accuracy, is assessed. Results show that the use of MM techniques combined with (M)NDWI increases the mean accuracy of surface change estimation between two data sets by around 22% compared to the other water indices. Compared to the vegetation index, the accuracy is increased by around 61%. The application of SMA increases the mean accuracy of this surface change estimation by a further 20%. In a nal step, the new method is applied over a 100 km long segment of the Mekong river in the VMD. This analysis reveals the locally big impact of water level uctuations on the position of the land-water interface during dry season. In order to assess the RB erosion in the VMD, it is therefore essential for further studies to work with satellite data collected between June to July. With its intermediate water level, the position of the land-water interface would most likely correspond to the actual RB.