Data-driven identification of three-dimensional spatial distribution of soft soil pockets below seabed for land reclamation using limited cone penetration tests

To protect the environment and minimize reclamation-induced disruption to marine ecosystems in land reclamation projects, soft soils (e.g., marine clay) below seabed may be stabilized in-situ using non-dredged ground improvement methods such as deep cement mixing (DCM). Design of DCM requires accurate information on three-dimensional (3D) spatial distribution of soft soils, including detailed locations of soft soil pockets, to determine the DCM termination depth and ensure a safe and sustainable reclamation. In engineering practice, it is challenging to accurately delineate 3D spatial variations of soft soil pockets below seabed, because subsurface site investigation data (e.g., cone penetration test (CPT) data) is often limited and there is a lack of effective methods for modelling 3D soil stratigraphy from limited CPTs. To tackle this challenge, a data-driven method is proposed in which, a 3D point cloud model is developed based on two cross-correlated CPT quantities, i.e., the normalized tip resistance Q_t and the normalized friction ratio F_R. Consecutively, many 3D random field sample (RFS) pairs of the cross-correlated Q_t and F_R are generated under a Bayesian framework, leading to probable samples of soil behavior types based on Robertson’s soil classification chart at each point within the 3D domain. Ultimately, the 3D spatial distribution of soft soil pockets is delineated automatically in a data-driven manner, with quantified uncertainty. The method is applied to a real reclamation site, and its performance is evaluated. The effect of CPT number on the performance of proposed method is also investigated. © 2026 Elsevier Ltd.