Levee Slope Reliability Based on Cross-Correlated Random Fields

Aging river levees are heterogeneous flood-defense earthworks whose reliability is affected by limited site data, interparameter dependence, and spatial variability. This study develops an integrated framework combining normal information diffusion (NID), Copula-based dependence identification, cross-correlated random fields (CCRFs), and Monte Carlo reliability analysis. Reliability was evaluated using Monte Carlo simulation based on a simplified infinite-slope benchmark model. Data were obtained from the authors’ own site investigations and laboratory tests. NID yields the lowest Kolmogorov–Smirnov statistics for all documented variables and better preserves local fluctuations and lower-strength tails than four classical marginals. Gaussian Copula is preferred for the representative bivariate cases and slightly outperforms the multivariate t Copula for Hongpaihe. For strongly dependent Beiwei data, the correlated model gives a mean factor of safety of 1.235 and a failure probability of 0.153, whereas independence gives a similar mean factor of safety of 1.238 but increases failure probability to 0.200. The weakly dependent main-dike case changes negligibly, while the φ–γ pair dominates the trivariate response. Illustrative CCRF realizations show how pointwise dependence can propagate into spatially continuous weak zones. Failure probability is therefore more diagnostic than mean factor of safety when stochastic model form is uncertain. The main novelty lies in the consistent integration and propagation of marginal, dependence, and spatial uncertainties within a unified levee reliability framework.