I have paper on researchgate (2015) on expanding stream network with LiDAR, but it is alot of hand digitizing if you want to get into GIS. Anastomosed streans are stable braided channels, but the other braided channels from excess sediment load are unstable. In areas without tree overstory, the braids are easier to see remotely, and a GIS expert may be able to separate out the water detail much like land use typing with small pixel size raster coverage. The infrared coverages often separate out water and wetlands the best for aerial photos. I have seen Landstat 8 a few times, but have not tried to apply them to this type detail to know the potential.
Many thanks for your help, William. It seems that hand digitizing is the most safety way for me considering knowledge background. I hope in future big companies like Google could offer dataset of all water-body in the world for research purpose. It will be very useful for analysis of river evolutions.