Monitoring river morphology with Sentinel 2 data: limitations and opportunities across scales

Satellite imageries are starting to become for geomorphologists a new tool to monitor medium-large river dynamics at high revisit time (weekly or daily). The Sentinel 2 mission, in particular, provides without charges a multi-spectral image of the earth surface at 10 meters resolution every 5 days (cloud cover permitting). Machine learning algorithms can then classify these images, automatically discriminating those river macro-geomorphic features, i.e. water, sediment and vegetation, that describe how a river responds to different hydrological impulses and boundary conditions. When using these tools (Sentinel 2 images + machine learning algorithm), it is important to first identify what geomorphic processes we can reliably detect, i.e. what are the applicability boundaries dictated by the spatio-temporal resolution of these images. In a dynamic, braided reach of the Sesia River (Northern Italy), we assessed how this inherent uncertainty associated with S2's spatiotemporal resolution can impact the interpretation of the active channel (a combination of sediment and water) delineation and evolutionary trajectory. The analysis demonstrates that water is ∼20% underestimated whereas sediments are ∼30% overestimated. These under- and over-underestimations are not random but a function of the mixed pixels present in each classified macro geomorphic unit. Nevertheless, the results show that these spatial errors are an order of magnitude smaller than the geomorphic changes detected in the 5 years analysed, so the derived active channel trajectory can be considered robust. Within these newly assessed applicability boundaries, in the Po River basin we started to explore in similarly dynamic river reaches new geomorphic indicators able to describe river responsiveness to seasonality and to different flood regimes.