Probabilistic investigation of the whole process of backward erosion piping for a two-layered levee in spatially variable soils

Abstract

Backward erosion piping (BEP) at the base of levees is one of the main causes of levee failure. BEP typically occurs in two-layer levees systems and involves multiple stages throughout its development. While random field methods have been widely applied to model soil spatial variability, existing approaches predominantly focus on single-layer soil structures. Analyzing the impact of only single-layer spatial variability on piping development, especially during the initial uplift stage, may not adequately capture the complexities of actual conditions. To address this, this paper proposes a general framework that can be used to generate both single-layer or two-layer random field models. The Karhunen-Loève (K-L) series expansion method has been refined to generate two independent random fields, thereby facilitating the characterization of spatial variability in both the upper and lower soil layers. Monte Carlo Simulation (MCS) is employed to generate realizations of the random fields, which are then incorporated into numerical models to evaluate the effects of spatial variability on the whole process of BEP. The results show that single layer models tend to overestimate or underestimate the probability of piping initiation. The stage of backward erosion is influenced by the spatial variability of the upper and lower layered soil parameters. The single-layer random field model has a tendency to overestimate the probability of slope failure. In addition, the results of several stability analyses under the effects of extreme rainfall and upstream water level rise show that the probability of slope failure increases significantly during this process.