Boundary condition fidelity for bottom-hole pressure and CO2 plume prediction in geological carbon storage

Technical Summary: Boundary Condition Fidelity for Bottom-Hole Pressure and CO2 Plume Prediction in Geological Carbon Storage

Problem Statement
Accurate prediction of bottom-hole pressure (BHP) and CO2 plume migration is critical for the safe and economic deployment of geological carbon storage (GCS). However, practical simulations often rely on truncated domains to manage computational costs, introducing artificial boundaries that reflect pressure back toward the injection well rather than allowing dissipation into the far field. This reflection artificially elevates simulated BHP and distorts plume trajectories by constraining the pressure gradients driving lateral migration. While various modifier-based boundary treatments (e.g., pore-volume multipliers and transmissibility modifiers) have been proposed to mitigate these artifacts, existing studies have generally assessed pressure or plume behavior in isolation. Furthermore, the specific impact of corner cells—where two domain edges meet and represent a disproportionately large chunk of missing reservoir volume—has not been systematically pinned down across both homogeneous and heterogeneous reservoir architectures.

Methodology
The study evaluates ten distinct boundary-condition configurations against a full-domain "truth" reference simulation (501 × 501 grid) using two reservoir models: a homogeneous model and a heterogeneous model generated via sequential Gaussian simulation. Both models share identical geometry, injection schedules (2,500 tonnes/day for 10 years, followed by 40 years of monitoring), and physical properties, ensuring that differences in BHP and plume behavior are attributable solely to boundary treatments and reservoir heterogeneity.

The ten scenarios tested include:

1. Uniform treatments: Uniform pore-volume multipliers (PVM) alone (V) and with transmissibility modifiers (VT).
2. Corner-adjusted treatments: Uniform PVM/TM with additional scaling for corner cells to recover missing volume (VA, VAT).
3. Layered treatments: PVM and TM distributed across four concentric cell layers (GhV, GhVT), adapted from Ghomian et al. (2024), with corner adjustments.
4. Gradual treatments: Power-law distributed PVM and TM that vary continuously from the edge midpoint to the corners (VAG, VAGT), based on the framework of Ramadhan & Hosseini (2026).

Performance was quantified using BHP peak pressure deviation, Root-Mean-Square Error (RMSE), Normalized RMSE (NRMSE), and the Intersection over Union (IoU) metric for CO2 plume extent (gas saturation > 0.001).

Key Contributions and Findings

• Corner-Volume Conservation is Paramount: The most significant finding is that conserving the pore volume of corner cells is the primary requirement for accurate truncated-domain modeling. Uniform treatments that neglect corner storage (V and VT) generate substantial errors. In the homogeneous model, these schemes produced BHP RMSE values of 362–382 psi and plume IoU values of ~0.80 (indicating 16–20% of the plume area is misrepresented). In the heterogeneous model, errors were similarly large (250–304 psi RMSE; IoU 0.82–0.84).
• The Dual Nature of Transmissibility Modifiers: The study demonstrates that transmissibility correction is not universally beneficial.
  • In homogeneous reservoirs, adding a transmissibility modifier to a corner-adjusted PVM (VAT) slightly improves BHP fidelity (reducing RMSE from 139 to 108 psi) and plume IoU (0.978 to 0.991).
  • In heterogeneous reservoirs, applying a uniform transmissibility modifier (VAT) degrades performance significantly. It over-restricts flow across high-permeability boundary faces and under-restricts it across low-permeability faces, leading to a net throttling effect. This increased BHP error (NRMSE rose from 4.83% to 11.00%) and contracted the predicted plume (IoU dropped from 0.9468 to 0.9408).
• Superiority of Distributed Corrections: The gradual power-law modifier with transmissibility correction (VAGT) and the layered scheme with transmissibility correction (GhVT) provided the most consistent performance across both reservoir types. By distributing corrections smoothly or across multiple layers, these methods avoid the abrupt property contrasts that amplify errors in heterogeneous settings.
  • Homogeneous: VAGT achieved BHP NRMSE < 3.6% and plume IoU > 0.99.
  • Heterogeneous: VAGT achieved BHP NRMSE < 3.7% and plume IoU > 0.97.
• Grid Resolution Effects: In homogeneous reservoirs, increasing the grid size from 101×101 to 251×251 offered negligible accuracy gains for the gradual modifier, as the pressure diffusion was already well-represented. In heterogeneous reservoirs, the larger grid provided a modest but measurable improvement, as the increased physical separation allowed local pressure fluctuations to dissipate before reaching the modified boundary.

Significance and Practical Implications
The paper establishes that boundary condition fidelity is a coupled problem affecting both pressure and spatial dimensions of GCS risk. The authors conclude that for practical GCS workflows:

1. Corner adjustment is non-negotiable: Any truncated model must account for the missing corner volume to avoid large systematic errors.
2. Context-dependent modifier selection: While uniform transmissibility modifiers may suffice for homogeneous screening, they are detrimental in heterogeneous reservoirs.
3. Recommended approach: The authors propose the corner-adjusted PVM (VA) as a robust, low-complexity baseline for early screening. For detailed characterization and regulatory compliance modeling—where both BHP and CO2 plume fidelity are critical—they recommend the gradual power-law modifier with transmissibility correction (VAGT) as the default strategy.

These findings provide a direct path to improving injection design, Area-of-Review (AoR) delineation, and regulatory compliance demonstrations by ensuring that simulation artifacts do not trigger unnecessary injection rate adjustments or regulatory reviews.