Nonlinear Deformation Analyses of Embankments on a Spatially Variable Liquefiable Deposit Modeled Using Conditional Random Fields

Seismic evaluations of embankment dams and levees often involve synthesis and idealization of site investigation data in a simplified spatial structure with equivalent uniform properties assigned to each soil layer or category. The process of assigning equivalent single property values to entire soil layers and categories significantly simplifies the detailed information obtained during the site investigation. This study examines the potential benefits of incorporating the spatial variability of liquefiable soils captured in site investigation data into nonlinear deformation analyses (NDAs) of embankment dams using conditional random fields. Parent models (PMs) are analyzed using unconditional random fields to represent “true” subsurface conditions of a spatially variable, liquefiable soil layer under different size embankments. From each parent model, single continuous columns of soil data are extracted and used to simulate data obtained during site investigations. Conditional embankment models are created using random fields conditioned on the simulated site investigation data to represent possible subsurface conditions. Deformations obtained from sets of conditional model NDAs are compared to those obtained from their parent model NDAs for different embankment sizes and scales of fluctuation to examine the deformation prediction ability of the conditional models.