The Emergence and Impact of Scientific Machine Learning in Geophysical Exploration

Scientific machine learning (SciML), often referred to as scientific computing with machine learning, is an emerging interdisciplinary field that integrates traditional scientific computing methods with modern machine learning techniques. The aim of SciML is to augment data-driven learning in scientific applications where traditional machine learning approaches might struggle. Conventional machine learning models typically learn patterns from large quantities of data but may struggle with limited or noisy data sets, or where interpretability, reliability, and robustness are essential. They also often lack the ability to incorporate prior scientific knowledge, and sometimes produce results that, while statistically valid, may be physically impossible. SciML, on the other hand, combines physical models (based on scientific laws and principles) with machine learning techniques. This integration allows the models to effectively learn from smaller or noisier data sets and ensure that the outcomes are consistent with established scientific knowledge. It also offers improved interpretability and generalization capabilities. Applications of scientific machine learning are increasingly being found in a variety of fields, including but not limited to geophysics, climatology, materials science, biology, and fluid dynamics. A number of advancements have been made in recent years in the field of geophysical modeling and inversion using emerging SciML paradigms, including physics-informed neural networks (PINNs), Fourier neural operators (FNOs), and Deep Operator Networks (DeepONets). In this abstract, we review those developments, highlighting the potential impact of such methods and the associated challenges in making these methods mainstream.