Academic Benchmarks
4 curated benchmark datasets for reservoir simulation ML evaluation. Each dataset is evaluated with standardized arena metrics (rel-L2, MRE, MAE in consistent units) alongside dataset-specific metrics for direct comparison across model families.
CO2 Nested
3D CO2 storage benchmark on a nested 100x100x5 grid with 24 prediction timesteps. Models predict pressure change from permeability, porosity, and injection parameters.
CO2 Radial
2D CO2-water multiphase flow benchmark for carbon capture and storage. Radial-symmetric domain (96x200 grid) with 24 prediction timesteps.
Two-Phase 3D Channels
3D two-phase flow benchmark on a 40x40x20 channel model. Models predict pressure and saturation at 10 selected timesteps from permeability and initial conditions.
Two-Phase Oil-Water
2D Cartesian two-phase oil-water flow benchmark. Models predict pressure and water saturation over 40 timesteps from permeability, well configuration, and initial state on a 40x40 grid.