Collaborative Research: CAIG: Multi-Task and Multi-Scale Deep Learning Inversion for Geophysical Imaging and Monitoring

University Of Texas At Dallas, Richardson TX

Investigators

Abstract

Understanding the Earth’s subsurface is essential for society’s ability to monitor natural hazards, manage environmental risks, and support sustainable energy development. Seismic waves generated by earthquakes and other sources can provide clues about underground structures, but interpreting these waves remains complex and computationally intensive. This project will advance artificial intelligence (AI) methods for imaging and monitoring the Earth’s subsurface. By combining advances in geophysics and machine learning, the research will enable more accurate and efficient interpretation of seismic data. These innovations may help scientists detect changes beneath the Earth’s surface more quickly and with greater clarity, benefiting efforts in natural hazard monitoring, such as volcanic or earthquake activity, and informing future strategies for environmental management and energy exploration. Aligned with these goals, the project will provide hands-on training in geoscience and AI for students and researchers, fostering cross-disciplinary innovation, education, and collaboration. Technically, the research team will develop a multi-task deep learning inversion framework that jointly estimates subsurface velocity structures and earthquake source parameters using passive seismic data. Conventional full-waveform inversion methods require accurate initial models and typically alternate between updating velocity and source parameters. Existing AI approaches often handle the two tasks independently by assuming that either the velocity model or the source parameters are known, thereby limiting their applicability. This project will address these limitations by introducing a unified framework that captures the intrinsic coupling between earthquake sources and velocity structures in seismic inversion and enables their simultaneous estimation within a single learning system. By integrating deep learning with conventional full-waveform inversion, the proposed approach aims to reduce dependence on initial models and improve computational efficiency after a one-time model training. Additionally, the team will design a multi-scale inversion strategy that enables the AI model to help resolve subsurface features across a range of domain sizes and depths. The framework will be validated using real-world data from seismically active regions where accurate monitoring is essential for public safety. The project will also produce benchmark datasets and open-source tools to support continued research in AI-enhanced Earth imaging and monitoring. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.