Near-real-time Probabilistic Assessment of Riverbank Structures using Monitoring-integrated Numerical Models

Abstract

Riverbank hydraulic structures are typically monitored using deterministic thresholds, whereas their reliability is assessed through probabilistic analysis during design. This separation limits the ability of monitoring systems to quantify evolving structural risk under uncertain geotechnical conditions and stochastic excitations. To bridge this gap, this study proposes an integrated probabilistic monitoring framework that couples offline probability density evolution (PDEM) with online surrogate-based distribution correction.
An offline baseline database of response probability density surfaces (PDS) is established by incorporating spatially variable geotechnical parameters and stochastic excitation models within a reduced-dimensional representation. This baseline captures the temporal evolution of structural response distributions under representative environmental scenarios. For near-real-time application, a lightweight surrogate model is developed to infer distribution-level correction parameters from monitoring-derived features, enabling rapid reconstruction of updated PDS without repeated dynamic simulations. The framework is validated through a numerical case study of a Π-shaped anti-scour wall slope. The surrogate-reconstructed PDS demonstrates strong agreement with direct probabilistic solutions, with Jensen-Shannon divergence and Earth Mover's Distance generally below 0.1, while reducing computational cost by more than two orders of magnitude. The proposed method enables real-time extraction of distribution-based risk indicators and provides a probability-informed pathway for monitoring and early warning of hydraulic structures subjected to coupled environmental uncertainties.