Enhancing Active channel Delineation in Alluvial Rivers using Monthly Aggregation of Sentinel-2 Imagery

In aerial and satellite imagery, the active channel of an alluvial river encompasses water channels and exposed sediment bars, delineating areas of geomorphic activity over a defined time window. While increasing satellite data availability enables monthly active channels delineations, multi-year analyses often rely on synthetic composites (e.g., annual medians) to reduce computational costs and intra-annual variability. The potential of monthly information to improve active channels delineation accuracy and geomorphic interpretation remains largely unexplored. In this work, we delineated yearly active channels for the Po River (Italy) by aggregating monthly Sentinel-2 (S2) classifications based on pixel-level occurrence frequencies for river and sediment classes, derived from a pre-trained global Fully Convolutional Neural Network applicable across river morphologies. Monthly variations in water and sediment classifications reveal both model classification biases and geomorphic dynamics. Results show that: 1) Monthly-aggregated information can enhance the accuracy of annual active channel delineations once the model classification biases are known; 2) In dynamic reaches, monthly active channel areas can vary substantially due to intra-annual sediment bar dynamics; these variations are masked in delineations based on single high-resolution orthophotos or on synthetic S2 annual medians. In contrast, active channel delineations on less dynamic reaches show minimal differences across methods. These findings highlight how different temporal aggregation should be considered for active channel delineations across different river morphologies, with dynamic rivers more dependent on high-revisit frequency data.