Automated Quantification of River Dynamics from Sentinel-2 Imagery

Spaceborne techniques offer unprecedented opportunities for monitoring network-scale river changes, providing near-weekly observations of medium-large rivers globally. This data can revolutionize our understanding of river dynamics and inform river management. However, current processing methods often focus on wetted channel changes and centerline migration as proxies for planform dynamics. This approach overlooks the ability to map and quantify changes across the entire geomorphic active channel, which includes not only the wetted surface but also exposed sediment bars and newly established vegetation.

We present an automated methodology based on Sentinel-2 imagery and a Convolutional Neural Network (CNN) for generating time series of geomorphic active channel masks, enabling inter-annual comparisons to identify floodplain reworking and abandonment areas. Lateral channel mobility is then automatically classified as either permanent or transient, based on the spatial extent and temporal persistence of changes.

Application to the Po River (Italy) demonstrates the method's ability to: (i) predict short-term channel trajectory revealing areas abandoned or recently activated by the river channel, (ii) identify reaches with a narrowing/widening trend, or those with limited lateral mobility, and (iii) relate active channel changes to the hydrological forcing and the planform morphology, disclosing morphological behaviors of the river system. The Sentinel-2 historical series available (2017-2023) is used to map stable trends of progressively abandoned or activated areas of the river channel, limiting misclassification errors. Results reveal that confinement, induced by the presence of bank protections and levees, plays a crucial role in river mobility alongside planform morphology. In highly confined reaches, the 7 year-variation in active channel width is limited to 5%, compared to 12% in less confined reaches. Single-thread reaches are particularly affected, with activated areas remaining scarce and abandoned zones showing a more intermittent history, driven by fluctuations in water levels and the variable establishment of new vegetation on sediment bars. 

This systematic monitoring provides quantifiable and visually interpretable insights into river dynamics, enhancing our ability to predict future channel trajectories. Such information can be used to monitor medium-large rivers globally and automatically explore their dynamics and morphological response to climate and environmental changes, as well as to inform restoration plans and risk mitigation strategies.