The successful storage of CO2 in saline aquifers requires careful selection of injection scenarios that maximize storage capacity whilst minimizing leakage risks. The limited data on these fields means the challenge lies in optimizing development strategies whilst accounting for uncertainty. This study proposes an automated workflow to generate an ensemble of hydrodynamic-geomechanical models and optimize potential injection locations for efficient carbon storage.

The workflow was applied to the Smeaheia open dataset, creating multiple plausible models by varying key reservoir and geomechanical parameters. A neural network algorithm was then run across this ensemble to identify injection strategies that maximise CO₂ storage whilst maintaining geomechanical integrity. 

Results show that effective well location and perforation schemes can be identified across a wide range of possible scenarios. Compared to deterministic optimization, this uncertainty approach reduces the likelihood of selecting injection locations that appear optimal in one case but are high risk under alternative scenarios. 

This work demonstrates the value of combining automated optimization with uncertainty analysis in coupled fluid-geomechanical models. The methodology highlighted in this study offers a transferable framework for designing injection strategies that remain effective under geological and geomechanical uncertainty.