A Coupled Bow-Tie and Bayesian Network Framework for CO2 Leakage Risk Assessment in Geological Storage

CO2 geological storage is essential for achieving carbon neutrality, but injected CO2 may migrate beyond the storage complex through wellbores, faults/fractures, or caprock defects, potentially affecting shallow groundwater and marine ecosystems and, in severe cases, reaching the atmosphere. Long-term CO2 leakage risk is uncertain and should be periodically updated as monitoring evidence becomes available. This study proposes a framework that supports evidence updating and diagnostic inference, coupling the Bow-tie model with a Bayesian network. The framework maps the Bow-tie model to a static Bayesian network, represents expert uncertainty using triangular fuzzy numbers, constructs conditional probability tables via a hybrid Noisy-Max and deterministic-override strategy, and is applied to the Lishui 36-1 offshore depleted gas reservoir. Annual conditional probabilities represent operational-period risk under a pre-injection baseline; they are not cumulative post-closure probabilities. Monte Carlo propagation yields a baseline median negligible-leakage probability of 95.65%, with a 5th–95th percentile interval of 95.51–95.80%. Sensitivity analysis identifies wellbore leakage prevention failure, injection overpressure, and management deficiency as dominant factors. Conditional barrier analysis indicates that monitoring and early warning are critical for preventing severe consequences, while long-term management and ecological restoration reduce moderate environmental damage. Overall, the framework provides a structured decision-support tool for risk identification, relative risk ranking, and monitoring prioritization in data-scarce CO2 storage projects.