Abstract

Backward erosion piping (BEP) is a form of internal erosion and common failure mode along levees. Despite over a century of study, predicting where BEP will initiate is still a considerable challenge. This study proposes a new model for predicting BEP initiation focused on the widest range of applicability. A logit model is trained using data from 15 sites along the Lower Mississippi Valley. The included parameters are independent of geography or geological regime and exhibit recorded or suspected correlations to BEP. Three significant factors (95% confidence interval) are retained for the final model: cumulative clay thickness within the blanket (odds ratio (OR) 0.520), critical gradient (OR 0.001) and exit gradient (OR 63.15). Receiver operating characteristics analysis indicates an area under the curve of 0.823. The model demonstrates 71% classification accuracy, a dramatic 10% increase over previous logit model attempts. Model results are most applicable within 150 m of the levee toe to predict new incidents of BEP initiation. The final model is a useful tool for BEP assessment and mitigation efforts.