GlobalTomo: A global dataset for physics-ML seismic wavefield modeling and full-waveform inversion

Abstract#

GlobalTomo provides a global dataset for physics-ML seismic wavefield modeling and full-waveform inversion. This comprehensive dataset enables researchers to apply machine learning techniques to seismic data analysis and Earth’s structure modeling.

Dataset Overview#

Scale and Coverage#

• Global Coverage: Worldwide seismic data collection
• Temporal Range: Multi-year data spanning various seismic events
• Data Volume: Terabytes of processed seismic waveforms
• Resolution: High-resolution spatial and temporal sampling

Data Types#

1. Seismic Waveforms: Raw and processed seismic signals
2. Velocity Models: 3D Earth structure models
3. Event Catalogs: Earthquake and other seismic event metadata
4. Station Information: Global seismic station network data

Applications#

Physics-ML Integration#

• Combining physical models with machine learning
• Data-driven velocity model construction
• Automated event detection and characterization

Full-Waveform Inversion#

• Enhanced inversion algorithms using ML
• Improved computational efficiency
• Better handling of complex Earth structures

Technical Details#

Data Processing Pipeline#

Raw Seismic Data → Quality Control → Feature Extraction → ML-Ready Format

Machine Learning Applications#

• Neural network-based wavefield modeling
• Deep learning for velocity estimation
• Automated data quality assessment

Impact#

This dataset enables:

• Advanced seismic imaging techniques
• Better understanding of Earth’s internal structure
• Improved earthquake hazard assessment
• Development of next-generation seismic analysis tools

Access and Usage#

The dataset is publicly available through our project website and includes:

• Comprehensive documentation
• Example usage scripts
• Benchmark tasks for ML evaluation