Introduction
Ground investigations prior to large infrastructure projects were traditionally limited to geotechnical drillings. Availability of modern geophysical equipment, both ground-based and airborne, allows for dense, high-resolution sampling of the subsurface. However, analysis of such large amounts of geophysical data can be time-consuming and costly. In this paper we show that automatic, machine learning (ML) based, interpretation of densely sampled geophysical data is feasible and can be conducted in a time efficient manner.
In this study, we use subsurface resistivity models derived from transient electromagnetic (TEM) measurements as input for our ML based analysis. The data were acquired in 2022 as part of a feasibility study investigating the usability of both ground-based and airborne TEM data for aerial subsurface mapping prior to the possible construction of a large tunnel system for the Einstein Telescope in the Euregio Meuse-Rhine region (www.einsteintelescope.nl).
Conclusion
In this study we used a towable time-domain tTEM system, to collect high-resolution transient electromagnetic data in the Euregio Meuse-Rhine region of the Netherlands. The data were processed and inverted to create a 3D resistivity model. Using few manual interpretation points as a-priori training data, a specialized artificial neuronal network interpreted the 3D resistivity model to produce a bedrock surface.
High resolution geophysical surveys often contain large amounts of data. Manual interpretation of these datasets is often tedious and labour intensive. Using a specialized artificial neural network, trained on a-priori information, we were able to automatically interpret a bedrock surface at every model location in a time efficient manner. This method is scalable to large regional surveys, as the training of the neuronal network is computationally expensive, but the interpretation using the trained network is rather quick. This method is also applicable to other parameterized, TEM and non-TEM based, models.
The small scale tTEM survey demonstrated the principal feasibility of a project-wide airborne survey to accurately map overburden thickness, however the amount of expected infrastructure couplings reduces the overall achievable model coverage.
