PSI - Issue 84

Luciano Pavesi et al. / Procedia Structural Integrity 84 (2026) 583–590

588

Once the infrastructures have been identified and parameterized from the high-resolution DEM, the DEM is resampled and processed to a computationally advantageous coarser resolution (Fig. 1c) without losing this information, while the detected infrastructure elements are directly integrated into the hydrological–hydraulic model (Fig. 1d). Levees are explicitly represented through lateral flow confinement, modifying conveyance scales and stage discharge relationships relative to non-leveed conditions. Similarly, bridges are represented as effective cross-section constrictions that influence rating curves and associated backwater effects, which are explicitly modelled within the hydraulic framework. Indeed, in order to accurately capture the influence of levees and bridges on inundation dynamics, it is important to represent them through their hydraulic behaviour rather than solely through their geometric presence in the DEM. These structures induce variations in water depth that propagate not only locally, but also cascade along the channel network, substantially modifying inundation patterns throughout the system. The most critical example is when multiple bridges are positioned at short distances from each other and confined by levees: the backwater effect generated by a bridge can interact with the flow confinement imposed by nearby levees, creating compounded hydraulic perturbations that propagate upstream and downstream. A purely geometric representation in the DEM would fail to capture these complex hydraulic interactions and their reach-scale effects.

Fig. 3. Application of the levee detection algorithm proposed by Khanh et al. (2025) to a reach of the Tiber River located downstream of the city of Rome: a, f) examples of the terrain elevation along two areas of the river reach; b, g) relative elevation (meter); c, h) terrain slope (degrees); d, i) terrain aspect; e, l) profile curvature, and (right upper and lower panels) automatically detected levee location and elevation. The reader is referred to Khan et al. (2025) for the definition of the above morphological parameters. Once the main infrastructures are identified (Fig. 1b) and the DEM is processed (Fig. 1c), flood maps are produced through the RESCUE model (Fig. 1d) that consists of four main components: (1) geomorphological analysis using the DEM to extract and segment the river network into nodes and reaches; (2) cross-section definition and infrastructures corrections using the Height Above Nearest Drainage (HAND) model to derive hydraulic properties and generate rating curves via the Manning equation as in Zheng et al. (2018); (3) hydrological load estimation using the rational formula to calculate discharge in all river reaches for given return periods; and (4) flood map generation by solving the 1D gradually varied-flow equation under steady-state conditions along the river network as detailed in Pavesi et al. (2022). Water levels at segment nodes are linearly interpolated and spatially propagated using the HAND map to identify flooded areas based on terrain elevation and water depth. The model outputs (Fig. 1e) include probabilistic flood inundation maps that integrate the uncertainty from hydraulic parameters through Monte Carlo analysis as detailed in Pavesi et al. (2024). By coupling flood hazard estimates with vulnerability curves and exposure data, we produce comprehensive risk assessments on population and on all different assets (such as infrastructures, buildings, agriculture, population etc.). The framework provides not