Tuesday 04 March 2025
The quest for more accurate and efficient surface wave tomography has led researchers to explore innovative approaches, including the development of a transformer-based neural network called DispFormer. This new method has been shown to outperform traditional techniques in both zero-shot and few-shot learning scenarios.
Surface wave tomography is a crucial tool for geophysicists, as it allows them to reconstruct the Earth’s subsurface structure by analyzing seismic waves that travel through the planet’s crust and mantle. However, this process can be challenging due to the complexity of the Earth’s interior, limited data availability, and computational requirements.
DispFormer addresses these issues by leveraging a transformer architecture, which is typically used in natural language processing tasks. In this context, the transformer blocks extract period-related features from the dispersion curves, allowing DispFormer to process data of varying lengths without requiring adjustments to its structure or alignment between training and test datasets.
The authors evaluated DispFormer using both synthetic and real-world datasets. In the zero-shot scenario, where no labeled data was available, DispFormer produced inversion profiles that closely matched the true models. When fine-tuned with limited labeled data, DispFormer outperformed traditional global search methods, delivering comparable or superior results.
One of the key benefits of DispFormer is its ability to handle datasets with varying period ranges, missing data, and low signal-to-noise ratios – common issues in real-world surface wave tomography applications. This versatility makes DispFormer a promising approach for broader applications in geophysics.
The authors also demonstrated DispFormer’s effectiveness in assessing uncertainty in inversion results. By applying the model to a real-world dataset, they showed that DispFormer can provide reliable estimates of uncertainty, even when faced with noisy data. This capability is critical for ensuring the credibility of surface wave tomography results and informing decision-making processes.
The development of DispFormer highlights the potential benefits of integrating techniques from artificial intelligence and machine learning into geophysics. By leveraging these approaches, researchers may be able to develop more efficient and accurate methods for solving complex problems in Earth sciences. As the field continues to evolve, it will be exciting to see how DispFormer and similar techniques are applied to tackle some of the most pressing challenges facing our planet.
The authors’ work on DispFormer serves as a testament to the power of interdisciplinary collaboration between geophysicists and computer scientists. By combining their expertise, they have created a novel approach that has the potential to revolutionize surface wave tomography.
Cite this article: “Transformer-Based Neural Network Advances Surface Wave Tomography”, The Science Archive, 2025.
Geophysics, Surface Wave Tomography, Transformer-Based Neural Network, Dispformer, Zero-Shot Learning, Few-Shot Learning, Natural Language Processing, Earth Sciences, Artificial Intelligence, Machine Learning
