Incorporating Observations to Update the Piping Reliability Estimate of the Francis Levee

Piping failure of levees is one of the main contributors to flood risk. Piping occurs when the head difference over a levee results in uplift of the inland blanket, followed by backward erosion in the aquifer due to high local flow velocities. Whether piping occurs highly depends on local geo-hydrological circumstances which are uncertain due to a combination of high soil variability and limited site investigation. Fragility curves, which show the failure probability as a function of water level, are increasingly used for risk assessment of levees to address this uncertainty. The objective of this paper is to show how pore water pressure measurements and visual field observations reduce or increase the calculated failure probability of a levee. The Francis levee in Mississippi (USA) is used to show the effects since this levee is historically known for being very piping sensitive. The first step is to construct fragility curves based on analytical piping models. The second step consists of using historic piezometer data to update knowledge about aquifer and blanket properties. The final step is to use historic seepage observations and sand boils to update the fragility curve. This paper shows that measurements and observations can significantly reduce the geo-hydrological uncertainties regarding a piping-sensitive subsoil. The failure probability of the Francis levees decreased with a factor 10 after incorporation of all the data. This increases the prediction accuracy of levee performance at high water levels and subsequently the estimation of whether it may require reinforcement or not.