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

Accurately capturing the complex interaction between CO2 and water in porous media at the pore scale is essential for various geoscience applications, including carbon capture and storage (CCS). We introduce a comprehensive dataset generated from high-fidelity numerical simulations to capture the intricate interaction between CO2 and water at the pore scale. The dataset consists of 624 2D samples, each of size 512x512 with a resolution of 35 μm, covering 100 time steps under a constant CO2 injection rate. It includes various levels of heterogeneity, represented by different grain sizes with random variation in spacing, offering a robust testbed for developing predictive models. This dataset provides high-resolution temporal and spatial information crucial for benchmarking machine learning models.

💡 Research Summary

The paper presents a high‑resolution benchmark dataset designed to support machine‑learning surrogates for pore‑scale CO₂‑water multiphase flow. A total of 624 two‑dimensional samples are generated, each consisting of a 512 × 512 pixel image with a physical resolution of 35 µm. For every sample, 100 equally spaced temporal snapshots are provided, containing water saturation (α_water), pressure (p), capillary pressure (p_c), horizontal and vertical velocity components (U_x, U_y), and a binary mask of the pore geometry (img). The dataset spans five levels of heterogeneity, realized by random perturbations of grain radius and position on a triangular lattice; the perturbation amplitudes increase from Level 1 (well‑sorted) to Level 5 (highly heterogeneous). Grain radii of 70, 80, and 90 µm and porosities from 0.20 to 0.45 are combined, yielding 78 base geometries after visual screening; each base is cropped into four quadrants and vertically mirrored, producing the final 624 geometries.

The flow simulations are performed with GeoChemFoam, an OpenFOAM‑based solver that employs the algebraic Volume‑of‑Fluid (VOF) method together with the Continuous Surface Force (CSF) model. CO₂ is injected from the left boundary at a constant rate of 1 × 10⁻⁸ m³ s⁻¹ (capillary number ≈ 5 × 10⁻⁶). Fluid properties are set to realistic subsurface values: μ_CO₂ = 7.37 × 10⁻⁸ m² s⁻¹, ρ_CO₂ = 3.84 × 10² kg m⁻³; μ_water = 1 × 10⁻⁶ m² s⁻¹, ρ_water = 1 × 10³ kg m⁻³; interfacial tension σ = 0.03 N m⁻¹, contact angle θ = 45°. Simulations run for 1 s with a 0.01 s output interval and a convergence tolerance of 1 × 10⁻⁸. The resulting fields are stored in HDF5 files; water saturation values lie in