Surface-wave methods are widely used to probe subsurface structures because wave dispersion naturally links frequency to depth. Yet traditional workflows remain slow, subjective, and computationally demanding, relying heavily on manual interpretation and iterative inversion. These challenges limit their use in dense monitoring networks and time-sensitive engineering applications. Artificial intelligence has emerged as a powerful alternative, enabling automation and dramatic speedups. However, many AI-based approaches operate as black boxes, raising concerns about physical reliability and generalization across geological settings. Based on these challenges, there is a clear need to conduct in-depth research on how AI can be integrated with physical constraints to ensure trustworthy seismic inversion.
In a review published (DOI: 10.1016/j.bdes.2025.100039) on November 28, 2025, in Big Data and Earth System, researchers from Zhejiang University of Technology, Zhejiang University, and Anhui University of Science and Technology examine how artificial intelligence is reshaping surface-wave seismic methods. The article surveys recent advances in automated dispersion analysis, deep-learning-based inversion, physics-guided modeling, and explainable AI. By systematically comparing data-driven sensitivity patterns with classical seismic theory, the authors assess both the promise and the current limitations of AI-driven seismic imaging.
The review shows that AI has reshaped nearly every step of surface-wave analysis. Deep learning models can now automatically extract dispersion information from complex seismic data, removing the need for time-consuming manual picking. Once trained, neural networks can invert dispersion measurements into shear-wave velocity models far faster than traditional optimization methods, making large-scale imaging feasible.
Crucially, the study emphasizes that speed alone is not enough. By comparing network-derived Jacobians with classical physical sensitivity kernels, the authors reveal that some AI models rely on statistical correlations rather than physically meaningful depth–frequency relationships. This mismatch can lead to misleading interpretations, particularly in poorly constrained depth ranges.
The review also highlights emerging solutions. Physics-guided and physics-informed models incorporate geological knowledge or governing equations into network design, improving stability and interpretability. A featured case study demonstrates how AI-assisted feature analysis can help identify subsurface karst cavities from seismic velocity models more objectively than manual inspection. Together, these results show that AI is most powerful when it complements—rather than replaces—physical understanding.
“AI has clearly changed what is computationally possible in seismic imaging, but accuracy alone is not enough,” the authors note. “Without physical consistency, fast results can still be misleading. Our comparison between data-driven and physical sensitivities shows why interpretability must become a core component of AI-based inversion. Physics-guided learning offers a practical path forward, allowing AI models to remain efficient while preserving the fundamental relationships that govern wave propagation.”
Physics-guided AI surface-wave methods could significantly improve applications ranging from urban hazard assessment and infrastructure planning to groundwater monitoring and environmental studies. Faster, automated workflows 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. 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.
###
References
DOI
10.1016/j.bdes.2025.100039
Original Source URL
https://doi.org/10.1016/j.bdes.2025.100039
Funding information
This study is funded by the National Natural Science Foundation of China (Grant No. 42304155) and Zhejiang Provincial Natural Science Foundation of China (Grant No. LMS25D040001).
About Big Data and Earth System
Big Data and Earth System aspires to be an interdisciplinary beacon, illuminating the integration of Big Data analytics and computational approaches within Earth System Sciences. Our mission is to promote cutting-edge research that leverages advanced data analytics, machine learning, and computational methods to understand, monitor, simulate, and predict Earth system processes and changes. By bridging the gap between data science and earth sciences, the journal aims to facilitate the development of innovative methodologies and applications that can enhance our understanding of Earth system and contribute to global sustainability efforts.