Recent Vegetation Dynamics in the Po River Corridor Observed by Monthly Satellite Imagery

This research employs monthly Sentinel-2 data along with a two-step machine learning classification approach to explore the current spatiotemporal dynamics of aquatic vegetation in the Po River corridor between 2017 and 2023. High Vegetation Persistence (HVP; ≥80%) and Moderate Vegetation Persistence (MVP; 50–80%) have been employed in the research process as indicators of sediment deposition, riverbank stability, and geomorphological activity. Monthly surface conditions were identified employing a Convolutional Neural Network (CNN) developed on the DenseNet-Tiramisu framework. The preciseness of vegetation recognition in the active river corridor was subsequently improved by modifying the CNN with a Linear Discriminant Analysis (LDA) model. This procedure made it possible to create yearly persistence maps for every class throughout a 340-kilometer stretch of the Po River that was split up into 35 morphologically distinct portions. Three significant reaches – PO_11 (upstream), PO_18 (midstream), and PO_28 (downstream) – were specifically investigated in order to evaluate the various ways in which morphological structure and hydrological variability impact plant patterns. According to the analysis, single-thread morphologies (such as straight and sinuous) demonstrate limited persistence, however transitional morphologies (such as wandering and anabranching) promote larger and more forever HVP coverage. Significant connections were found between year-to-year variations in MVP and HVP throughout the corridor and hydrological data from the Borgoforte station, which includes yearly peak discharge (Q_max) and cumulative flow (Q_cumulated). These results demonstrate that vegetation persistence evaluated remotely is an effective substitute to assess geomorphological evolution and channel stability. This research provides a more dynamic and geographically obvious understanding of river corridor processes by combining hydrological time series, machine learning classifiers, and high-resolution satellite data. This provides essential data for river monitoring, restoration, and adaptive management in complicated regulated structures.