CPT-Based Probabilistic Characterization of Three-Dimensional Spatial Variability Using MLE

AbstractEngineering geological characterization, subject to spatial variability of soil properties, is a three-dimensional (3D) problem in reality, although it is often simplified as one- or two-dimensional. Direct characterization of 3D spatial variability is a challenging task due to the scarcity of geotechnical data and a satisfactory characterization method. To address such a problem, this paper develops a cone penetration test (CPT)–based probabilistic approach for characterizing 3D spatial variability underlying the framework of maximum likelihood estimation (MLE). A matrix decomposition technique is applied to enhance the practical application of MLE for high-dimensional and spatially correlated data. Results of a case study and three virtual site analyses indicate that MLE provides more accurate estimates of random field parameters with smaller statistical uncertainty than the commonly used method of moments with best fitting, particularly for the estimation of scale of fluctuation. In addition, simultaneous vertical and horizontal characterization based on multiple CPTs is a feasible way for 3D spatial variability characterization in the presence of limited data, such as the limited sounding issue and the thin layer issue. The sampling strategy having some closely located CPTs is preferable for 3D spatial variability characterization.