Population-based structural health monitoring of historical masonry bridges using spaceborne InSAR measurements

Italy’s railway network includes nearly 10,000 masonry arch bridges, many of which now carry loads far exceeding their original design while undergoing material degradation, posing a major infrastructure management challenge. Conventional structural health monitoring relies on costly on-site instrumentation that cannot be deployed at this scale. This paper proposes a population-based SHM framework that identifies anomalous deformation in masonry arch bridges from satellite interferometric (InSAR) measurements, while keeping physically meaningful, engineering-interpretable indicators. For each bridge, Sentinel-1 line-of-sight displacements and environmental variables (air temperature and soil moisture) are condensed into compact statistical features describing the global displacement distribution, independent of the uncertain position of individual InSAR scatterers. Subspace alignment, a feature-based domain adaptation technique, is used to harmonize the baseline feature distributions across structures with different geometry. To overcome the scarcity of labeled anomalies, a simplified parametric bridge model is used to generate a population of synthetic bridges with imposed anomaly scenarios, on which a bidirectional long short-term memory regression network is trained. Through simulation-to-real transfer, the network is then applied to real data. The model outputs two physically interpretable descriptors, namely anomaly intensity and spatial extent. Validation is carried out on eight masonry arch railway bridges in Liguria, Italy. A persistent localized settlement identified on one structure is consistent with independent velocity maps. The framework offers a scalable, non-invasive tool for regional-scale screening and inspection prioritization of heritage bridge stocks.