Wednesday 12 March 2025
Deep learning has revolutionized many fields, from self-driving cars to medical imaging. Now, it’s being applied to a field that’s crucial for our understanding of the Earth: seismology.
Seismologists use seismic waves generated by earthquakes or artificial sources to create detailed images of the Earth’s subsurface. However, this process is often limited by the quality and quantity of data available. In many cases, low-frequency data – which contains valuable information about the Earth’s structure – is missing or incomplete.
To address this issue, researchers have developed a new approach that uses deep learning to predict the initial model for seismic full-waveform inversion (FWI). This technique involves creating a smooth, high-velocity anomaly in the subsurface by extrapolating low-frequency data from higher-frequency data. The resulting image is then used as an input for FWI, allowing researchers to refine their models and extract more accurate information about the Earth’s structure.
The team used a convolutional neural network (CNN) to train their model on synthetic datasets and field data from the northwestern part of Australia’s offshore region. They found that their approach significantly improved the accuracy of the initial model, leading to better results in FWI.
One key advantage of this approach is its ability to handle complex geological structures and noise in the data. The CNN can learn patterns and features in the data that would be difficult or impossible for human analysts to identify. This allows it to generate more accurate models than traditional methods, which often rely on manual interpretation and interpolation.
The implications of this research are significant. By improving the accuracy of seismic models, researchers can better understand the Earth’s subsurface structure, which is crucial for a range of applications, from oil and gas exploration to earthquake hazard assessment and environmental monitoring.
In addition, this approach has the potential to revolutionize the field of seismology by providing a new tool for data analysis and interpretation. It could be used in conjunction with traditional methods to generate more accurate models, or as a standalone technique to analyze complex datasets.
The researchers are already exploring ways to apply their approach to other fields, such as medical imaging and computer vision. As the technology continues to evolve, it’s likely that we’ll see even more innovative applications of deep learning in seismology and beyond.
This new approach has the potential to transform our understanding of the Earth’s subsurface structure, and could have significant implications for a range of fields.
Cite this article: “Deep Learning Breakthrough in Seismology: Accurate Modeling of Earths Subsurface Structure”, The Science Archive, 2025.
Seismology, Deep Learning, Convolutional Neural Network, Seismic Full-Waveform Inversion, Subsurface Structure, Earth’S Structure, Oil And Gas Exploration, Earthquake Hazard Assessment, Environmental Monitoring, Medical Imaging
