Data-knowledge enhanced large geological model

Abstract

Underground space development is an inevitable trend in sustainable urban growth. To advance the capabilities of underground digital twins, urban geological models provide a vital solution by integrating regional geological data and characterizing large-scale stratigraphic variability. In this context, probabilistic tunable large geological models (LGMs) have been previously developed, utilizing local stratification in the form of virtual boreholes (VBs) to mitigate measurement dependency. However, existing studies lack an effective fusion of multi-source data and geological expertise, both of which are crucial for characterizing complex urban engineering geology. Therefore, this study proposes a novel framework to enhance the LGMs by integrating hard data, soft data, and geological knowledge. The area division map is proposed to distinguish sub-regions within urban areas that exhibit similar stratigraphic characteristics from a macro perspective. Then, a data-driven optimization approach is developed to objectively determine the configurations of VBs for individual sub-regions. Based on the VBs constructed by experienced geologists, a Large-Scale Random Field-Based (LS-RFB) method is introduced to incorporate topographic and superficial information, improving the characterization of stratification similarity across large areas. The proposed framework is applied to construct the first tunable LGM including mountainous and reclaimed regions of the Macao Peninsula, covering a total area of approximately 8.1 km2. Furthermore, the framework is validated using a real-world case, demonstrating the enhanced LGM’s superiority in stratigraphic prediction with sparser measurement inputs at the initial stages of borehole planning, providing a practical solution to reduce data costs.