Speaker
Description
Faults are common geologic structures in sedimentary basins that may host industrial-scale geologic CO$_2$ sequestration (GCS). However, their three-dimensional architecture and heterogeneous material distribution are typically poorly characterized, which poses significant challenges for assessing the risk of fluid migration. To support the safe scale-up of GCS, decision-support methods must quantify and reduce uncertainty in the fluid-flow properties of fault zones and their impact on CO$_2$ migration. Achieving this requires close collaboration between geologists, reservoir engineers, and uncertainty-quantification researchers.
In this contribution, we briefly introduce PREDICT, an open-source methodology to quantify the directional components of the fault permeability tensor and the fault capillary pressure in siliciclastic settings. We then demonstrate how PREDICT can be integrated in an uncertainty quantification workflow to forecast pore pressure and fluid migration within faults. The workflow, which couples flow and geomechanics, leverages a time-marching surrogate model with a deep neural network architecture to reduce the computational cost of quantifying uncertainty in the quantities of interest (QoIs). We show that the surrogate model successfully captures the multimodal nature of the QoI probability distributions, and identifies the dominant parameters for each QoI using variance-based global sensitivity analysis. The resulting ensemble statistics for the QoIs provide critical information to guide decision-making in CO$_2$ storage projects.
| Country | United States |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
