Modeling seismic impacts on bridge networks and accessibility: a scenario-based framework for resilient transportation systems

• Scenario-based framework links bridge damage to accessibility loss. • Ordinal logistic regression predicts bridge vulnerability by magnitude. • Accessibility modeling reveals 70 % loss under M9.5 earthquake scenario. • Network redundancy strongly influences regional resilience outcomes. • Framework supports equity-focused and resilience-based planning. Bridges serve as critical links in regional transportation networks, where seismic disruption can generate cascading impacts on mobility, accessibility, and emergency response. This study develops a transferable, scenario-based modeling framework to quantify how varying earthquake magnitudes influence bridge performance and regional connectivity. Structural, geotechnical, and seismic indicators are integrated within a formally defined ordinal logistic regression (OLR) model calibrated using post-event bridge inspection data from the 2018M7.1 Point Mackenzie earthquake in Southcentral Alaska. Predictor variables—including bridge age, condition, soil uniformity, and magnitude-scaled peak ground acceleration (PGA)—were used to estimate ordered damage probabilities across four earthquake scenarios (M6.5–M9.5). Results indicate a nonlinear escalation in bridge vulnerability with increasing magnitude, with the highest damage concentrations along primary corridors characterized by limited redundancy. Under upper-bound scenarios, predicted collapse probabilities reach 20–25 %, producing potential isolation of communities reliant on single-route access. Accessibility modeling shows that even moderate structural degradation can propagate into disproportionately large regional mobility losses. The framework supports resilience-oriented planning by linking probabilistic damage modeling with geospatial accessibility analysis, enabling agencies to identify corridor-level risks, prioritize retrofit investments, and evaluate emergency response strategies. Although demonstrated in Alaska, the approach is adaptable to other seismically active or infrastructure-limited regions. Key methodological assumptions and uncertainties, particularly those related to PGA scaling and model extrapolation—are explicitly acknowledged to guide appropriate interpretation of results.