What is it about?

This work introduces a new artificial intelligence (AI) method for processing geophysical data, specifically electrical resistivity tomography (ERT), which helps us map what lies beneath the Earth’s surface. Traditional methods for this task can be slow and often fail to provide meaningful information about the confidence in the results. Our method combines deep learning with physics to deliver faster solutions, and includes an efficient way to measure the confidence in those results. The approach works well on both simulated and real-world data, providing clear images of underground features with realistic uncertainty estimates — all while keeping computational costs low.

Featured Image

Why is it important?

Accurate estimation of the subsurface properties is essential for groundwater exploration, environmental monitoring, and engineering projects. However, conventional approaches tend to be computationally demanding to offer uncertainty quantification. By combining physics and machine learning, our framework not only improves the reliability of subsurface models but also helps geoscientists assess the confidence of their predictions in affordable computational times. This enables better, faster decision-making in the field and has the potential to transform how we plan exploration surveys, assess risks, and protect natural resources.

Perspectives

This research shows that the future of geophysical inversion lies in hybrid approaches that blend physics with data-driven models. Our physics-guided deep learning technique opens the door to near real-time subsurface imaging with uncertainty quantification — something previously only achievable with costly probabilistic methods. In the long term, we envision deploying such models directly in the field to guide data acquisition and make informed decisions on the spot.

Felipe Rincón
Universita degli Studi di Pisa

Read the Original

This page is a summary of: Physics-guided deep learning DC-resistivity inversion with uncertainty quantification, Geophysics, April 2025, Society of Exploration Geophysicists,
DOI: 10.1190/geo2024-0434.1.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page