Using Conditional Random Fields for a Spatially Variable Liquefiable Foundation Layer in Nonlinear Dynamic Analyses of Embankments

Two-dimensional nonlinear dynamic analyses (NDAs) are performed for a series of hypothetical embankment dams on spatially variable liquefiable foundation layer to evaluate the utility representing with random fields conditioned different levels site characterization information. A set two-dimensional parent models (PMs), each true condition, were generated using unconditional equivalent clean sand, corrected standard penetration test (N1)60cs values. Different then represented by combining numbers local borings (i.e., columns data from PM) optional inclusion constraints geostatistical properties that might come sitewide explorations. NDAs same input motions PM (which represents perfect knowledge soil conditions), realizations alone, and statistics. Embankment deformations obtained conditional compared those potential benefits increasing in terms deformation prediction accuracy. Parametric include varying size, scales fluctuation stratum, number conditioning borings, ground motions. The results these comparisons illustrate beneficial effects generally limited cases horizontal scale approaching base width large (more than three per fluctuation), which may not be practical many situations. Additional limitations spatial layers dam discussed.