Comprehensive Uncertainty Sampling for Optimised CO2 Storage Site Design

Over recent decades, CO2‐storage design has relied on deterministic models and one-at-a-time sensitivity analyses, which fail to capture the full spectrum of geological, petrophysical, and operational uncertainties. This study introduces an ensemble-based framework that systematically samples these uncertainties – using hundreds of realisations of a saline‐aquifer reservoir model – to generate probabilistic forecasts of CO2‐plume behaviour. We perturb layer geometries and petrophysical properties with spatial random fields and simulate 15 years of injection plus 285 years of post-injection plume migration. The resulting CO2-plume shapes vary significantly across the 272 realisations and yield large differences in the minimum distance to legacy wells – a key risk metric for assessing leakage potential. A small ensemble of 11 realisations, mimicking one-at-a-time parameter perturbations of a base model, fails to capture the full range of outcomes. Our findings highlight the necessity of comprehensive uncertainty sampling to avoid underestimating risks and to enable more robust, cost-effective storage designs. Future work will extend the framework to additional risk metrics, integrate further sources of uncertainty, and deploy multi-fidelity proxy models for even broader and more efficient uncertainty assessments.