Repository navigation
OpenSWI-v1.0
Overview
We are excited to release OpenSWI, a benchmark dataset designed to support data-driven deep learning methods for surface wave dispersion curve inversion. OpenSWI addresses challenges in geophysical exploration and geological studies by providing diverse datasets for both shallow and deep earth models, along with real-world data for model generalization.
Datasets
- OpenSWI-shallow: 22 million 1-D velocity profiles derived from shallow 2-D geological models, covering structures like flat layers, faults, and folds.
- OpenSWI-deep: 1.26 million 1-D velocity-dispersion pairs from 14 global and regional 3-D models for deep earth studies.
- OpenSWI-real: Real-world observed dispersion curves and reference models for validating deep learning model performance.
Features
- Comprehensive Data: Synthetic and real-world datasets for both shallow and deep geological applications.
- Pre-trained Models: Deep learning models trained on OpenSWI datasets and evaluated on real-world data.
- SWIDP Toolbox: Tools for preparing datasets and running inversion experiments.
Results
Models trained on OpenSWI-shallow and OpenSWI-deep show strong performance on OpenSWI-real, confirming the utility of the dataset for advancing surface wave inversion.
Get Started
Download the datasets, pre-trained models, and toolbox from the OpenSWI GitHub repository.