Meet ML-SEISMIC: A Physics-Informed Deep Learning Approach for Mapping Australian Tectonic Stresses with Satellite Data

Understanding the current stress state of the Earth’s crust is imperative for various geological applications, ranging from carbon storage to fault reactivation studies. However, traditional methods face significant challenges, primarily due to the manual tuning of geomechanical properties and boundary conditions. The need for accurate stress orientation information becomes apparent, as it is pivotal for reliable geomechanical models. The manual adjustment processes inherent in these traditional methods hinder the efficiency and accuracy of stress and displacement field estimations. A new research paper from CSIRO, Australia, addresses these challenges by introducing a novel solution, ML-SEISMIC, a physics-informed deep neural network designed to align stress orientation data with an elastic model autonomously.

In geological investigations, conventional inversion processes have long been the norm. However, these processes demand meticulous manual adjustments of geomechanical properties and boundary conditions, making them prone to errors and inconsistencies. The research team introduces ML-SEISMIC as a groundbreaking alternative. This physics-informed deep neural network overcomes the limitations of traditional methods by nearly eliminating the need for explicit boundary condition inputs. The proposed approach signifies a leap forward in geodynamic investigations, promising a streamlined and powerful process.

ML-SEISMIC’s methodology hinges on applying physics-informed neural networks to solve linear elastic solid mechanics equations. The governing equations encompass momentum balance, constitutive relationships, and small strain definitions. The neural network optimizes stress field eigenvalues concerning stress orientations, thus providing a comprehensive understanding of the stress and displacement fields. The application of ML-SEISMIC to Australia serves as a case study, revealing its ability to autonomously retrieve displacement patterns, stress tensors, and material properties. The method proves effective in overcoming the shortcomings of traditional approaches, offering a reliable interpolation framework. Notably, ML-SEISMIC utilizes Global Navigation Satellite Systems (GNSS) observations to revisit large-scale averaged stress orientations and identify areas of inconsistency. The results underscore the adaptability of the approach across various scales, from crystallographic investigations to continental-scale analyses.

In conclusion, ML-SEISMIC emerges as a transformative solution in geological investigations. By autonomously aligning stress orientation data with an elastic model, this physics-informed neural network addresses the inherent challenges of traditional methods. The research team’s innovative approach streamlines the stress and displacement field estimation processes and eliminates the need for explicit boundary condition inputs. The adaptability of ML-SEISMIC across different scales, coupled with its reliance on accurate GNSS observations, positions it as a catalyst for advancements in understanding complex geological and tectonic phenomena. In the ever-evolving landscape of scientific inquiries, ML-SEISMIC promises to be a versatile and powerful tool, ushering in a new era of insights into Earth’s dynamic processes.

Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 35k+ ML SubReddit, 41k+ Facebook Community, Discord Channel, LinkedIn Group, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..