Probabilistic assessment of multi-mode failure of levees under climate change

Levees are critical flood defense infrastructures, yet their reliability is increasingly threatened by climate change. To address this challenge, this study develops a novel framework that couples non-stationary flood frequency analysis with mechanism-based failure modeling to quantify multi-mode failure probabilities of levees under climate change. The framework employs the time-varying moments (TVM) model with temperature and precipitation as covariates, incorporates CMIP6 climate projections under multiple Shared Socioeconomic Pathway (SSP) scenarios, and applies Monte Carlo simulation to assess overtopping, piping, slope sliding, and system failures. Application to the Shijiao levee validates the framework against historical failure events and demonstrates its predictive power under climate change. Results show that by 2100, system failure probability increases by factors of 3.6 (SSP126), 5.7 (SSP245), and 21.7 (SSP585) relative to the historical baseline (Ps f = 0.009). Piping dominates under low- and medium-emission scenarios, whereas high emissions lead to synergistic multi-mode failures. Stationary design standards are found to severely underestimate risks: even with anti-piping reinforcement, the safety threshold will be exceeded decades earlier under all scenarios, with Ps f reaching 0.974 by 2100 in SSP585. This study establishes the first probabilistic assessment framework for multi-mode levee failure under climate change, offering a scientific basis for climate-adaptive risk management.