Although it is widely acknowledged that a three-dimensional (3D) model of geotechnical property variation is the cornerstone of underground digital twin, its adoption in practice remains limited, probably due to the scarcity of site-specific geotechnical data collected from site investigation. During a site investigation program, geophysical surveys are carried out sometimes, allowing for a rapid, nondestructive recognition of subsurface conditions over large areas. The abundant geophysical data obtained provides valuable complementary information for developing a reliable 3D geotechnical property model. However, no practical and rational method is available currently for a quantitative fusion of geotechnical and geophysical data in a 3D space. To tackle this challenge, this study develops an innovative 3D fusion method for efficiently integrating limited geotechnical data with high-resolution geophysical data in a 3D space. The auto- and cross-correlation of 3D geotechnical and geophysical data are exploited using joint sparse representation in the proposed method, enabling efficient exchange of information between geotechnical and geophysical data in a data-driven manner. The proposed method is demonstrated and validated using both synthetic and real-world datasets. A significant improvement is observed in the construction of 3D model of geotechnical properties after an integration of geotechnical and geophysical data.