Quantifying Bridge Exposure to Landslide Hazards: A Regional-Scale Framework using Machine-Learning

Quantifying landslide-related risk for bridges and viaducts affected by slope instabilities is inherently complex, as it would require detailed information on hazard intensity, structural vulnerability, and consequences. For this reason, recent Italian guidelines propose a rapid, operational approach aimed at screening and prioritization, rather than full quantitative risk assessment, to evaluate the interaction between landslides and infrastructures. The initial screening phase relies on existing hazard maps and available documentation, followed by field inspections to identify visible indicators of instability. In practice, however, hazard maps may lack sufficient spatial resolution, and inspections may fail to detect critical or incipient phenomena. In this context, this study presents a comprehensive and scalable framework to support landslide susceptibility and exposure screening for bridges at the regional scale, using the Veneto region as a case study. Landslides are first classified into fast-moving processes, which can directly impact infrastructures, and slow-moving instabilities, which may progressively compromise structural performance. A detailed landslide susceptibility map is then developed by training a machine learning model on 23 geomorphological, climatic, and land-use variables, based on available landslide inventories and high-resolution spatial data. Susceptibility is combined with an Infrastructure Spatial Index (ISI), representing the area of influence of each bridge, to derive a Combined Infrastructure Landslide Susceptibility Index, which is categorized into susceptibility or attention classes rather than quantitative risk classes. The proposed framework supports the attribution of landslide-related attention classes for bridges in accordance with the Italian Guidelines and provides a consistent basis for network-level screening and prioritization.