Title: Earthquake monitoring based on graph neural networks and foundation models
Abstract: Earthquake monitoring is a fundamental aspect of seismological research, aiming to detect, locate, and analyze earthquake events to generate high-precision earthquake catalogs. These efforts enhance the understanding of focal mechanisms as well as earthquake risk assessment, and provide a solid scientific foundation for early warning systems. The global expansion of monitoring stations has led to an enormous accumulation of observational data, posing significant challenges for data processing and analysis. While deep learning has recently made substantial contributions to seismology, existing models often focus on specific tasks, limiting their generalizability across different regions and neglecting physical constraints. We examine two strategies to address these limitations. The first strategy integrates physical constraints into a graph neural network to perform earthquake monitoring, thereby enhancing the detection of low-magnitude events and improving overall earthquake monitoring accuracy. The second strategy introduces a foundational model based on contrastive learning, pretrained on multimodal data to effectively process seismic spectra and extract key features of the specific events. Compared to networks trained for individual tasks, this foundational model demonstrates superior capabilities in event classification, earthquake localization, and focal mechanism analysis. The findings highlight the potential of incorporating physical knowledge and foundational models to advance earthquake monitoring and underscore their feasibility in seismological research.
Biography: I am a Postdoctoral Researcher at GT, supervised by Prof. Zhigang Peng. Recently, I got my Ph.D degree at The University of Science and Technology of China, supervised by Prof. Xinming Wu and cooperated with Prof. Zefeng Li . Before that, I received the M.Sc degree of Geophysics at CUGB in 2020 and B. Eng degree of Geophysics at China University of Geosciences in 2017.
My research lies at Geophysical application using Graph Neural Network(GNN), Foundation Model and Convolutional Neural Network(CNN).
To join virtually: Zoom
Contact: xsi@gatech.edu
Website: https://sixu0.github.io
Recording: Zoom Recording (will be available within a week after the seminar)