Near-Surface Monitoring of Plugged and Abandoned Wells for Real-Time Leakage Detection in Geologic Carbon Storage

The risk of CO 2 and brine leakage to environmental receptors is one of the main concerns in geologic CO 2 storage. Legacy wells from past oil and gas activities may be located within the area of review, necessitating continuous monitoring to ensure they are properly sealed to prevent fluid migration. Deployment of an efficient monitoring system for early detection of leakage from failed wells is imperative to mitigate environmental and financial risks. This study proposes a cost-effective near-surface monitoring package capable of real-time surveillance of plugged and abandoned (P&A) wells. Controlled pilot-scale CO 2 and water release experiments were conducted to identify soil properties that are most sensitive to leakage in the near-surface vadose zone above P&A well stubs. Multiple release scenarios with different rates and durations were implemented, and machine learning techniques were applied to identify anomalous data patterns caused by leakage. Among measured parameters, soil electrical conductivity (EC) was the most sensitive indicator of leakage. Several machine learning models, including Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Random Forest, XGBoost, and LightGBM, were evaluated for anomaly detection in EC data. Tree-based models outperformed traditional classifiers, with Random Forest achieving the lowest false alarm rate and XGBoost yielding the highest detection rate. Uncertainty quantification using Conformal Prediction showed that LightGBM had the highest confidence in anomaly prediction. Although the experiments were performed under controlled conditions, the approach demonstrates a relatively promising, low-cost monitoring technique for leakage detection for near-surface monitoring of legacy wells. • A low-cost sensor network was offered for monitoring of leakage from P&A wells. • Soil electrical conductivity was the most sensitive indicator of CO 2 leakage. • Tree-based ML models enabled reliable anomaly detection in leakage data.