ABSTRACTThe ability to non‐invasively detect and monitor internal erosion and piping in earthen embankments is of great value to government agencies and municipalities responsible for maintaining large numbers of civil structures. Internal erosion can be a difficult failure mode of dams and levees to assess using traditional techniques. Surface nuclear magnetic resonance is a promising technology for monitoring these structures due to its direct sensitivity to liquid water. This is a new application for the method, and requires the formulation of a robust modelling framework capable of incorporating three dimensional water content models and topography. Adaptive octree meshes represent an efficient means to perform these calculations, as they are able to adapt and refine in areas where the kernel function is changing rapidly. Given the non‐trivial challenges associated with the development of realistic water content models, we focus on first‐order effects and present six 2.5D flow models representing seepage through a loaded earthen embankment under varying levels of erosional piping. Simulation results show that the employment of multiple surface nuclear magnetic resonance transmitter and receiver loops can provide early detection and effective monitoring of piping under the assumption of a 2.5D water content model. Therefore, the surface nuclear magnetic resonance technique has promise as a means for detection and/or monitoring of internal erosion and piping in earthen embankments.