PSI - Issue 84

Francesca Ceccato et al. / Procedia Structural Integrity 84 (2026) 599–606

600

Keywords: Landslide Susceptibility Map; Bridge Vulnerability; Infrastructure Resilience; Machine Learning; Spatial Prioritization; Smart Risk Mapping.

1. Introduction Landslides are among the most pervasive natural hazards in mountainous and hilly regions, causing significant damage to infrastructure and socio-economic systems (Sim et al., 2022). Their impact is particularly relevant in regions characterized by complex terrain and dense transportation networks, where slope instabilities may directly or indirectly interact with civil infrastructures. Landslides occur through different mechanisms, ranging from rapid phenomena such as debris flows to slow-moving processes including earthflows and deep-seated slides, which differ in triggering conditions, temporal evolution, and modes of interaction with structures (Cruden and Varnes, 1996; Hungr et al., 2014). Bridges constitute critical elements within transportation networks, underpinning mobility, emergency response, and economic activities (Nirandjan et al., 2022; Mowll et al., 2024). Recent studies on bridge failures have shown that landslides represent a non-negligible cause of collapse at the international scale. Proske (2019) demonstrated that slope-related phenomena account for a significant share of bridge collapses, second only to hydraulic triggers. This evidence is further confirmed by recent statistical analyses of Italian bridge failures, which identify landslides as one of the dominant natural causes of collapse, particularly when indirect or progressive interactions between slope instabilities and structures are considered (Scala et al., 2025). In Italy, this awareness has directly informed the development of specific national guidelines (MIMS, 2022), which explicitly require the evaluation of landslide hazard and its potential interaction with bridge and viaduct structures. These guidelines are conceived to be applied to extremely large infrastructure inventories, involving tens of thousands of bridges, and therefore promote a multi-level, risk-based approach aimed at prioritizing further investigations rather than performing detailed analyses at the network scale. However, extending landslide risk assessment, at least in a preliminary phase, to the entire infrastructure dataset poses a major operational challenge. Under such conditions, the assessment effort must necessarily remain limited in terms of data requirements, time, and expert judgement, while still being capable of identifying those structures that require further investigation and detailed analyses. From a landslide-risk perspective, hazard and susceptibility maps, together with inventories of known events, such as the IFFI and ITALICA datasets, represent strategic tools for large-scale screening. Nevertheless, these sources are often affected by limitations related to spatial resolution, update frequency, and completeness. In particular, landslide inventories tend to capture documented events, while failing to identify latent or incipient instabilities and the predisposing factors that may suggest the presence of potential landslides in areas without historical records. In this context, machine-learning and artificial intelligence techniques applied to landslide susceptibility modelling can provide effective support to large-scale infrastructure assessment. By recognizing spatial patterns associated with landslide predisposing factors, data-driven models can generate susceptibility maps that estimate the relative likelihood of landslide occurrence and, in some cases, the most probable landslide typology. When combined with infrastructure spatial data, these tools can support evaluators in the preliminary identification of bridges potentially exposed to landslide hazards, even in the absence of documented past events. This study proposes a framework to assess landslide-related hazard to bridges in the Veneto region. Landslide susceptibility for rapid and slow landslide types is computed using machine-learning techniques and normalized to represent landslide intensity. The interaction between landslide hazard intensity (LS) and the infrastructure is evaluated by computing a Combined Infrastructure Landslide Susceptibility Index (CILSI), which represents the average landslide susceptibility in an area surrounding the bridge. The proposed framework aims to support the evaluation of landslide hazard for infrastructures and the definition of the landslide attention class, in line with the risk-based philosophy of the Italian Guidelines. 2. Data and Methodology Selecting appropriate variables for landslide studies is challenging and widely debated, as controlling factors vary across regions (Shahabi et al., 2015). In this study, 23 variables were selected from multiple sources (Table 1). A 30 m Digital Elevation Model (DEM) from the USGS SRTMGL1 v003 dataset, accessed via Google Earth Engine (GEE),