Efficient probabilistic tunning of large geological model (LGM) for underground digital twin

Urban large geological models (LGMs) are essential for characterizing subsurface conditions for underground digital twins, facilitating informed decision-making. Incorporating uncertainty and efficient tuning methods for LGMs are indispensable technologies for enhancing reliability with dynamic geotechnical databases, yet these aspects are not fully addressed in current studies. This research proposes a novel framework to develop the first probabilistic tunable LGM, integrating local stratification knowledge and real borehole measurements. Local stratifications are collected from experienced engineering geologists and interpreted as virtual boreholes. These virtual boreholes are inputted into the stratum-informed random field-based method (SI-RFB) to develop geological prior for the LGM. Then, the spatial sequential Bayesian updating (SSBU) algorithm is utilized to partially tune the LGM with on-site borehole data. The influence zones of updating are mathematically predetermined based on project-specific borehole spacing. The effectiveness of the proposed framework is demonstrated through a simulated 3D case referencing a site in Macao. Furthermore, the proposed model is applied to develop a tunable urban LGM for the landfill region in the Macao Peninsula covering 6.4 km². The results emphasize the framework's ability to effectively tune the LGM, enhancing details and reducing uncertainty. Importantly, the method is computationally efficient, accounting only for up to 0.3 % of the conventional reconstruction cost for the same area, thereby providing an economically viable solution for underground digital twins.