Objectives Wide-area three-dimensional stratigraphic models play a crucial supporting role in the survey and design of major infrastructure projects such as railways and highways, the development of underground space resources, and disaster risk assessment. However, wide-area stratigraphic data, including regional geological reports and geological maps, are characterized by varying granularity of stratigraphic information, a lack of structured descriptions of stratigraphic adjacency and spatial distribution relationships, and scattered and poorly associated spatiotemporal and attribute characteristic information. The relational entity organization models commonly used in existing three-dimensional stratigraphic modeling methods struggle to accurately represent multi-dimensional relationships and multi-attribute characteristics, thereby limiting modeling efficiency.

Methods To address this issue, we propose a three-dimensional stratigraphic modeling method guided by an entity association network. First, a stratigraphic entity heterogeneous graph association network is constructed. Stratigraphic characteristic information extracted from geological reports, together with inherent stratigraphic information derived from geological maps using ChatGPT, is integrated to achieve a unified representation of association relationships under the joint constraints of spatiotemporal, attribute, and multi-domain characteristics of stratigraphic entities. Based on the stratigraphic entity heterogeneous graph association network, virtual boreholes are rapidly constructed to guide regional three-dimensional stratigraphic modeling.

Results By selecting typical areas and integrating stratigraphic characteristics and relationship information from digital elevation model, regional geological reports, geological maps, and other data, the stratigraphic entity association network is constructed to guide the generation of a three-dimensional stratigraphic model.

Conclusions The effectiveness and applicability of the proposed method are validated through these case studies.