Explainable AI for Risk Assessment in Geological CO₂ Storage

Abstract

Geological CO₂ storage (GCS) is a cornerstone technology for large-scale decarbonization, but its safe and durable deployment depends on rigorous, defensible risk assessment across containment, injectivity, and geomechanical stability. Machine learning and deep learning models are increasingly used to accelerate this risk assessment, offering predictive power that often exceeds conventional statistical or deterministic methods. However, the opacity of many high-performing models—particularly deep neural networks—creates a fundamental barrier to regulatory acceptance, operator trust, and defensible decision-making in a domain where errors carry substantial environmental and financial consequences. Explainable artificial intelligence (XAI) has emerged as a critical complement to predictive modeling in GCS, providing tools that decompose model outputs into interpretable, feature-level contributions and thereby bridge the gap between black-box predictive accuracy and the transparency demanded by engineers and regulators. This review examines the theoretical foundations and applications of XAI—including SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and inherently interpretable model architectures—for risk assessment across the CO₂ storage lifecycle, drawing on recent case studies in heterogeneous depleted gas reservoir optimization, geothermal-analogue reservoir characterization, and geomechanically informed wellbore stability assessment. It further considers cross-domain parallels with explainability-driven AI frameworks developed for public health and healthcare administration, which offer transferable lessons on deploying auditable AI systems in regulated, high-consequence domains. The review concludes by identifying persistent challenges in explanation fidelity, computational cost, and the translation of feature attributions into physically meaningful engineering insight, and proposes a research agenda for advancing XAI toward routine use in commercial-scale CO₂ storage risk management.