Prior geological knowledge enhanced Markov random field for development of geological cross-sections from sparse data

Qualified subsurface geological cross-sections are indispensable for efficient site planning and risk management of underground infrastructure. Nevertheless, delineating a qualified geological cross-section from sparse site-specific data is challenging, especially when dealing with heterogeneous str...

全面介紹

Saved in:
書目詳細資料
Main Authors: Qian, Zehang, Shi, Chao
其他作者: School of Civil and Environmental Engineering
格式: Article
語言:English
出版: 2024
主題:
在線閱讀:https://hdl.handle.net/10356/180899
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
機構: Nanyang Technological University
語言: English
實物特徵
總結:Qualified subsurface geological cross-sections are indispensable for efficient site planning and risk management of underground infrastructure. Nevertheless, delineating a qualified geological cross-section from sparse site-specific data is challenging, especially when dealing with heterogeneous strata patterns and inconsistent stratigraphic information revealed from adjacent boreholes. To address this challenge, this study proposes a knowledge-based geological modelling paradigm to enhance the traditional Markov random field (MRF)-based stratigraphic modelling method. An informed initial stratigraphic configuration is first inferred from a single training image that reflects prior geological knowledge. The configuration is then integrated with the MRF algorithm for stochastic simulations of geological cross-sections. The MRF parameters are estimated from the initial stratigraphic configuration and undergo automatic regularization and updating with constraints from site-specific boreholes in an unsupervised Bayesian manner. The performance of the proposed method is illustrated using real-world examples, including the Zhu-Hai subsea tunnel and Nan-Chang Urban Rail Transit tunnel projects. The results indicate that the model can accurately predict the presence of complex geological patterns, such as interlayer structures and isolated boulder stones, from sparse data with quantified stratigraphic uncertainty. Moreover, the advantages of the proposed model over existing approaches are demonstrated, and the practical significance of prior geological information for stochastic stratigraphic modelling is discussed.