Advanced Deep Learning Methods for CO₂ Leakage Detection and Environmental Risk Assessment

Abstract

Carbon capture, utilization, and storage (CCUS) has emerged as a cornerstone technology in global decarbonization strategies, yet its long-term viability hinges on the ability to reliably detect CO₂ leakage and quantify associated risks across heterogeneous subsurface environments. Traditional monitoring, verification, and accounting (MVA) approaches—reliant on sparse well logs, periodic seismic surveys, and physics-based reactive transport simulations—struggle to keep pace with the spatial and temporal complexity of injected CO₂ plumes. Deep learning (DL) has recently emerged as a transformative tool capable of fusing multimodal geophysical, geomechanical, and petrophysical data streams to detect anomalies, forecast plume migration, and assess wellbore and caprock integrity in near real time. This review synthesizes recent advances in deep learning applications for CO₂ leakage detection and risk assessment, drawing on case studies from depleted gas reservoirs, geothermal analogues, and offshore geomechanical systems. It further examines cross-domain methodological parallels with large-scale predictive analytics frameworks developed for public health and healthcare economics, which offer transferable insights into national-scale risk stratification and cost-benefit modeling. The review concludes by identifying persistent challenges—data scarcity, model interpretability, and regulatory validation—and charts a research agenda for physics-informed, uncertainty-aware deep learning systems capable of supporting safe, scalable geologic carbon storage.