Ensuring the integrity of CO₂ storage sites requires timely detection of fault, leaks and fracture activation that could compromise containment. While deformation data including displacements, tilts and strains offers a sensitive indicator of subsurface pressure changes, conventional numerical inversion methods remain too computationally intensive for real-time monitoring.

In this work, we trained a surrogate model to predict tilt signals from CO₂ injection scenarios. The model is trained using high-resolution finite element simulations and captures the relationship between pressure plume properties and geological deformation of the layers. It significantly accelerates inversion workflows without compromising accuracy. The model has been tested with thousands of new samples with variable depths, reservoir properties and plume sizes. Furthermore, the model was successfully tested against the field data from In Salah CCS project where both single and double plumes existed. The inversion model accurately predicted the location and the size of each plume.

Our approach enables rapid, field-scale interpretation of surface tilt for both conformance monitoring and anomaly detection. These results demonstrate how learned surrogate models can support efficient Measurement, Monitoring and Verification (MMV) strategies and unlock scalable digital workflows for large-scale CCS deployment.