A Benchmark Dataset for Machine Learning Surrogates of Pore-Scale CO₂-Water Interaction

Area of Science:

• Geosciences
• Pore-scale physics
• Computational modeling

Background:

• Understanding CO2-water interactions in porous media is critical for geoscience applications.
• Carbon capture and storage (CCS) requires accurate pore-scale models.
• Existing datasets may lack the resolution or heterogeneity needed for robust model development.

Purpose of the Study:

• To generate a comprehensive dataset of CO2-water interactions at the pore scale.
• To provide high-resolution temporal and spatial data for model benchmarking.
• To facilitate the development of predictive models for geoscience applications.

Main Methods:

• High-fidelity numerical simulations were employed.
• A dataset of 624 2D samples (512x512 resolution, 35μm) was generated.
• Simulations covered 100 time steps with constant CO2 injection and varied heterogeneity.

Main Results:

• A detailed dataset capturing CO2-water dynamics at the pore scale was created.
• The dataset includes varying levels of porous media heterogeneity.
• High-resolution data suitable for machine learning model benchmarking is provided.

Conclusions:

• The generated dataset offers a robust testbed for developing predictive models.
• This resource is valuable for advancing research in carbon capture and storage.
• The dataset enhances the ability to study complex fluid interactions in porous geological formations.