Subsurface characterization for Seoul through optimized geotechnical database based on geospatial interpolation with adaptive grids

In the era of big data, enhancing the reliability of large-scale geotechnical datasets is crucial for accurate subsurface characterization. Although various statistical interpolation techniques have been developed, significant challenges remain in addressing spatial variability and uncertainty inherent in subsurface conditions. This study presents an advanced spatial interpolation framework that integrates carefully preprocessed borehole datasets with adaptive grid-based modeling to improve the precision of subsurface mapping. The borehole data were classified based on elevation discrepancies from a Digital Elevation Model (DEM), temporally segmented by project year and type, and standardized using the 3-sigma rule to minimize outlier-driven distortions. The interpolation process combined kriging with localized averaging strategies and systematically varied grid resolutions to assess performance sensitivity. Leave-one-out cross-validation, using geological layer thickness as the reference metric, demonstrated that finer grids significantly reduced interpolation error near the surface, while deeper layers exhibited increased uncertainty. Notably, the implementation of an adaptive grid system, capable of dynamically adjusting spatial resolution according to data density and terrain complexity, proved essential in mitigating the smoothing effect often associated with kriging. Furthermore, in data-sparse regions, the integration of localized averaging within adaptive cells helped stabilize estimation accuracy. This adaptive approach offers a powerful enhancement to conventional spatial modeling techniques by enabling more faithful representation of geological heterogeneity and by reinforcing the robustness of predictions under variable data availability, ultimately contributing to more informed geotechnical decision-making.