3D lithological modelling in GIS: the influence of borehole scope and spatial pattern

Abstract

Over the past decades, with the development of digital tools and technologies, geotechnical engineering practice has shifted from traditional ground modelling procedures toward data-rich 3D models capable of representing geological-geotechnical complexity more realistically. The foundation of such models is represented by the field investigation data. In this regard, a key question remains open: what investigation scope is required, and how should it be optimally distributed in space to achieve reliable model predictions and maximize the information gain? In this study, Empirical Bayesian Kriging is employed in a GIS environment to generate 3D voxel-based lithological models. Subsets ranging from 10 to 70 investigation boreholes are analyzed, grouped into different spatial distribution patterns. The accuracy of lithology prediction is evaluated against an independent set of validation boreholes and relative to a quasi-reference model based on all 94 boreholes. The standard error of prediction is used as an indicator of uncertainty, introducing a threshold-based approach. The results show increasing prediction accuracy and decreasing uncertainty with a higher number of boreholes and with their optimal spatial arrangement, especially for models with poor investigation coverage.