From a single geological interpretation to history matching: A SinGAN-ES-MDA framework for CO2 storage in channelized aquifers

Geological carbon storage (GCS) is a critical technology for mitigating anthropogenic CO 2 emissions, yet its widespread deployment is hindered by subsurface uncertainty and the scarcity of reliable geological models. To address these challenges, this paper introduces a novel SinGAN-ES-MDA framework that integrates a Single Image Generative Adversarial Network (SinGAN) with the Ensemble Smoother with Multiple Data Assimilation (ES-MDA). Unlike conventional generative models that typically require many training realizations, our framework can effectively capture multi-scale sedimentary patterns from a single, conceptually reliable interpretation and samples geologically plausible ensembles for automated history matching and uncertainty quantification. We evaluate the framework on a synthetic 3D channelized aquifer representative of GCS operations. Assimilating bottom-hole pressure observations from injection wells, SinGAN–ES-MDA reduces data–model misfit, improves structural similarity of updated models to the reference, and maintains credible ensemble spread for reliable forecasts of reservoir performance. This study highlights the potential of integrating deep learning-based generative models with ensemble-based data assimilation techniques to enhance the accuracy and reliability of subsurface flow simulations, thereby contributing to the de-risking of GCS projects. The framework is designed to correct geometric misalignments within the parameterized range rather than to compensate for fundamental misinterpretations of the depositional system and is particularly suited for data- and interpretation-limited GCS settings where only a single high-confidence geological model is available.