Rock mechanics–driven interpretable machine learning framework for preliminary seal integrity screening in geological CO2 storage
Muhammad Hammad Rasool , A.K.M. Eahsanul Haque , Numair Ahmed Siddiqui , Hassan Soleimani , Maqsood Ahmad , Nisar Ahmed , Muhammad Mubashir , Venkateshwaran Baskaran , Md Nasirul Islam , Lawal Opeyemi Olajid
Green Energy and Resources ›› 2026, Vol. 4 ›› Issue (3) : 100203
Seal integrity assessment is essential for ensuring the long-term security of geological CO2 storage; however, early-stage evaluation is often challenged by limited and heterogeneous subsurface datasets. This study proposes an interpretable machine learning framework for preliminary seal integrity screening using limited rock mechanics data from the 31/5–7 Eos well-core of the Northern Lights CO2 storage project. An unsupervised learning workflow was first employed to reconstruct Safe, Transitional, and Risk integrity regimes through cluster validation and ambiguity-based boundary analysis. Multiple supervised learning algorithms, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and XGBoost, were subsequently developed and evaluated under stratified train/test split, Leave-One-Out Cross Validation (LOOCV), and repeated stratified cross-validation strategies. Random Forest demonstrated the most stable predictive behaviour across validation schemes, achieving balanced accuracies exceeding 0.86. Explainability analyses using Random Forest feature importance and SHAP consistently identified porosity and failure related effective axial stress as dominant screening controls, suggesting coupled stress–structure interactions govern integrity behaviour within the studied dataset. Based on these observations, a Stress–Porosity Screening Score (SPSS) was developed as a simplified post hoc screening descriptor for rapid integrity evaluation. Statistical analyses demonstrated significant differentiation of SPSS across reconstructed integrity regimes (p < 0.001). Decision-tree interpretation further enabled extraction of preliminary data-emergent screening pathways and threshold behaviour. Conclusively, the proposed methodology provides an interpretable and data-adaptive screening approach intended to support early-stage caprock assessment and guide subsequent detailed geomechanical investigations.
Caprock integrity / CO2 geological storage / Stress-controlled behaviour / Interpretable machine learning / Random forest / Decision tree
| [1] |
|
| [2] |
|
| [3] |
Bhagat, M., Bakariya, B., 2022. Implementation of logistic regression on diabetic dataset using train-test-split, K-Fold and stratified K-Fold approach: M. Bhagat, B. Bakariya,. Natl. Acad. Sci. Lett. 45 (5), 401-404. |
| [4] |
|
| [5] |
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
|
| [16] |
|
| [17] |
|
| [18] |
|
| [19] |
|
| [20] |
|
| [21] |
|
| [22] |
|
| [23] |
|
| [24] |
|
| [25] |
|
/
| 〈 |
|
〉 |