Short description and objective of the project:
Rivers exhibit distinct forms, or morphologies, characterized by three principal variables: channel sinuosity, the presence of depositional bars, and the number of channels comprising a river or river stretch. The main recognized river types are straight/wandering, meandering (sinuous single channel), single channel without a regular sinuous pattern, braided (containing mid-channel bars), and anastomosing (consisting of multiple separate, interweaving channels).
1: Li et al. 2025
With the increasing resolution and coverage of satellite imagery, it is now possible to leverage satellite data to produce a global classification map across these river types. Recent studies have demonstrated the viability of image recognition approaches for classifying river morphologies.
Building on this foundation, and on work by a former Utrecht University Applied Data Science MSc student who demonstrated the potential of self-supervised learning for clustering river morphologies, this thesis aims to develop a method for river segmentation and subsequent morphological clustering using untagged binary images of rivers.
References (optional):
- Li, Y., Zhang, Y., Zheng, N. et al. Global classification of river morphology based on inland water dynamics characterization and digital elevation data. Sci Rep 15, 14258 (2025). https://doi.org/10.1038/s41598-025-99174-7
- Nyberg, B., Henstra, G., Gawthorpe, R.L. et al. Global scale analysis on the extent of river channel belts. Nat Commun 14, 2163 (2023). https://doi.org/10.1038/s41467-023-37852-8
- https://studenttheses.uu.nl/handle/20.500.12932/47120
