Efficient CPT locations for characterizing spatial variability of soil properties within a multilayer vertical cross-section using information entropy and Bayesian compressive sensing

Subsurface geological profiles often contain multiple soil layers, and spatial variability of soil properties within such a profile plays a crucial role in geo-structure analysis. Such spatial variability may be characterized through in-situ tests, e.g., cone penetration tests (CPT). However, the number of CPT soundings in a specific site is often small due to time, budget, and/or technical constraints. Because subsurface conditions are often inhomogeneous, CPT at different locations might reveal different spatial variability of soil properties in terms of accuracy. This leads to a question on how to determine efficient locations for a given number of CPT soundings to obtain as accurate as possible information on spatial variability of a multilayer soil property profile. Furthermore, site investigation is a multi-stage process, and additional CPT soundings may be needed in a later stage. In these cases, a similar question arises on selection of efficient locations for additional CPT soundings. This paper proposes information theory-based methods to address these two issues in a vertical cross-section, which is illustrated using both simulated and real-life CPT data. Results show that the locations determined from the proposed methods are very effective with at least 90% probability outperforming randomly selected locations in characterizing spatial variability of multilayer soil properties.