AI Revolutionizes Seismic Imaging But Needs Physics to Stay Grounded

A groundbreaking synthesis of seismic research reveals that artificial intelligence is revolutionizing how scientists image Earth's subsurface structures, but with crucial caveats about maintaining physical reliability. Researchers from Zhejiang University of Technology, Zhejiang University, and Anhui University of Science and Technology demonstrate in their review published in Big Data and Earth System that AI can dramatically accelerate surface-wave analysis workflows—from automated signal extraction to inversion and interpretation—enabling major gains in speed, consistency, and scalability for applications ranging from urban hazard assessment to groundwater monitoring. However, the study warns that purely data-driven AI models can produce results lacking physical meaning, even when appearing accurate, highlighting the need for physics-guided AI frameworks that balance computational efficiency with interpretability.

The review, published on November 28, 2025, shows AI has reshaped nearly every step of surface-wave analysis, with deep learning models now automatically extracting dispersion information from complex seismic data and neural networks inverting measurements into shear-wave velocity models far faster than traditional methods. Crucially, the authors emphasize that speed alone isn't enough—by comparing network-derived Jacobians with classical physical sensitivity kernels, they reveal some AI models rely on statistical correlations rather than physically meaningful depth-frequency relationships, potentially leading to misleading interpretations. The study highlights emerging physics-guided and physics-informed solutions that incorporate geological knowledge or governing equations into network design, improving stability and interpretability while maintaining efficiency.

This research matters because physics-guided AI surface-wave methods could significantly improve real-world applications through faster, automated workflows that enable near-real-time analysis from dense sensor networks, including emerging distributed acoustic sensing systems. At the same time, interpretable AI models help practitioners identify uncertainty and avoid overconfidence in automated results, offering a more reliable foundation for seismic imaging in both research and practical engineering applications. As standardized datasets and physically informed architectures continue to develop, AI-driven seismic imaging is poised to move from experimental innovation to routine, reliable practice in Earth science and engineering, fundamentally transforming how we understand and interact with subsurface environments.