A depth-responsive EBK3D framework for risk-informed digital geotechnical mapping in heterogeneous subsurface conditions

Most digital geotechnical maps (DGMs) reduce three-dimensional subsurface conditions to two-dimensional representations, limiting their value for risk-informed design. This study develops a depth-responsive Empirical Bayesian Kriging framework (EBK3D) for volumetric geotechnical mapping. The framework centres on two calibrated vertical-adaptation controls: the Elevation Inflation Factor (EIF), which rescales vertical separation, and the Vertical Trend Order (VTO), which removes deterministic depth trends. These controls are integrated with direction-sensitive semi-variograms, local empirical Bayesian updating, and covariance-informed three-dimensional neighbourhoods. A quality-controlled regional database was used to generate DGMs of standard penetration test values (SPT-N), soil classes, and plasticity index (PI), which were benchmarked against alternative interpolation methods. The K-Bessel-based EBK3D configuration achieved up to 66% lower error and approximately 28% higher correlation than the benchmark methods, with borehole-wise leave-one-out cross-validation (LOOCV) correlations of 0.97–0.98. The calibrated models also maintained stable uncertainty, with ASE values of 1.32–1.67. Focal-scale analysis showed up to 46-fold greater preservation of localised spatial variability relative to diffusion-based interpolation. Five-fold spatial block validation retained correlations of 0.89–0.94 under complete spatial withholding, demonstrating useful regional transferability. The resulting 3D-DGMs provide a robust basis for subsurface zoning, geotechnical risk assessment, uncertainty-aware planning, and preliminary foundation design in heterogeneous depositional environments.