Boolean stochastic stratigraphic modelling of geological cross-section from sparse cone penetration tests

Land reclamation is vital for urban development in coastal cities such as Hong Kong. However, the long-term consolidation settlement of fine-grained materials poses significant risks to infrastructure safety and serviceability. It is imperative to accurately delineate subsurface stratigraphic distributions from sparse data for reliable geotechnical design. Conventional approaches typically assume stationary stratigraphic patterns of multiple layers within a single geological cross-section, neglecting the inherent heterogeneity of different strata and the importance of site-specific engineering judgement. In this study, a divide-and-conquer framework is developed to predict nonstationary subsurface stratigraphic patterns from limited cone penetration tests (CPTs), explicitly incorporating engineering judgement. Distinct stratigraphic characteristics of soil layers are identified and segmented from sparse CPTs using space transformation methods, e.g. signed distance function, before applying layer-tailored methods, such as Bayesian compressive sensing, for spatial interpolation. The predicted strata are then combined via Boolean operations for stochastic stratigraphic modelling. CPT data collected from a recent reclamation site in Hong Kong are adopted to illustrate the performance of the proposed method. A subsequent comparative study demonstrates that the proposed method excels in capturing intricate stratigraphic patterns with explicit uncertainty quantification, significantly outperforming traditional methods, such as direct BCS, Ordinary Kriging, and Indicator Kriging.