AI-Based Risk-Oriented Assessment of CO₂ Storage Capacity and Leakage Using Global Open Datasets
Paper ID : 1010-UGESR
Authors
Mohaddeseh Nazari, Fatemeh Daraee *
Farzanegan Campus of Semnan University, Semnan, Iran
Abstract
Global CO₂ emissions continue to rise, making Carbon Capture, Utilization, and Storage (CCUS) essential for achieving climate targets. However, assessing CO₂ storage projects is challenging due to uncertainties in capacity, operational performance, and leakage risk. In this paper, an AI-based, risk-oriented framework is proposed for large-scale assessment of CO₂ storage projects using the IEA CCUS Projects Database as a global open-access data source. A transparent rule-based scoring approach is employed to generate binary risk labels (High/Low) based on storage capacity, project type, and operational status, enabling supervised machine learning without reliance on subsurface simulations. Data preprocessing, feature encoding, and model training are implemented within a reproducible machine learning pipeline. Three algorithms XGBoost, Random Forest, and Logistic Regression are evaluated using a stratified train–test split and 5-fold cross-validation. The results demonstrate that XGBoost achieves superior predictive performance, with an F1-score of 0.97, outperforming the other models. Feature importance analysis consistently identifies project type and estimated storage capacity as the most influential predictors of risk classification. Overall, the findings confirm the potential of integrating AI with global open datasets to enable scalable, data-driven screening of CO₂ storage projects, supporting early-stage decision-making, investment prioritization, and risk mitigation in CCUS deployment.
Keywords
CO₂ Storage, Risk Assessment, Machine Learning, XGBoost, IEA CCUS Database
Status: Accepted (Poster Presentation)