TL;DR
This paper presents a novel approach to monitoring CO₂ plumes underground by combining fluid flow simulations with time-lapse seismic data using an Ensemble Kalman Filter (EnKF). This method offers more accurate CO₂ saturation estimates than using either fluid physics or seismic data alone, paving the way for safer and more efficient carbon capture and storage.
What Problem Does This Paper Address?
Carbon capture and storage (CCS) is a crucial technology for mitigating climate change, but safely storing CO₂ underground requires robust monitoring. Traditional methods are limited by accessibility, making it difficult to track the CO₂ plume's spread and detect potential leaks. Seismic data offers a non-intrusive way to image these subsurface plumes, but integrating this data with complex fluid dynamics simulations in a computationally efficient manner has been a significant challenge. Existing methods often struggle with the high dimensionality of reservoir models and the nonlinear nature of both fluid flow and seismic wave propagation.
Methods
The authors employ the Ensemble Kalman Filter (EnKF), a sequential Bayesian data assimilation technique. The EnKF is well-suited for this problem because it can handle nonlinear dynamics and large datasets by representing the system's uncertainty with an ensemble of possible states.
Here's a breakdown of their approach:
- Two-Phase Flow Simulation: They use a standard model for CO₂ and brine flow in porous rock, incorporating Darcy's law and a modified Brooks-Corey model for relative permeability. The key challenge here is the unknown and heterogeneous permeability field, which is treated as a source of uncertainty.
- Seismic Forward Modeling: They simulate time-lapse seismic data by solving the acoustic wave equation. This involves a rock physics model (patchy-saturation model) that relates CO₂ saturation to seismic properties like density and squared slowness.
- EnKF Integration: The EnKF iteratively updates an ensemble of CO₂ saturation estimates. In the "predict" step, the fluid flow model advances the ensemble forward in time. In the "update" step, the seismic observations are used to correct the ensemble members, bringing them closer to the true state. The EnKF's strength lies in its ability to represent and update the covariance matrix implicitly through the ensemble, avoiding the computational burden of storing a full covariance matrix.
Datasets
The study uses a synthetic benchmark dataset based on the "Compass model," a 3D model designed to capture geological complexities relevant to seismic full waveform inversion. For computational efficiency, they use a 2D slice of this model. The simulation involves 5 years of CO₂ injection, with new seismic observations generated annually.
Key Insights and Ideas
- Synergy of Physics and Data: The EnKF effectively combines the physics of fluid flow with seismic observations, leading to significantly better CO₂ saturation estimates than methods relying on only one of these components.
- Handling Uncertainty: The EnKF's ensemble-based approach is crucial for managing the uncertainty in the subsurface permeability field, which is a major challenge in reservoir monitoring.
- Computational Efficiency: While high-resolution simulations are computationally expensive, the EnKF offers a scalable solution for data assimilation, making real-time monitoring more feasible.
- Robustness to Noise: The study investigates the EnKF's sensitivity to noise parameters, finding that it can be robust to mis-calibrated noise estimates, especially when noise is simulated.
Key Results
- The EnKF consistently achieved lower root-mean-squared error (RMSE) in CO₂ saturation estimates compared to baseline methods that either ignored seismic data (NoObs) or ignored fluid physics (JustObs).
- The EnKF analysis (after assimilation) produced sharper plume estimates than the NoObs prediction, indicating that data assimilation helps to resolve finer details of the CO₂ distribution.
- The study demonstrated that the EnKF can recover features of the CO₂ plume, such as pockets in low-permeability zones, which are missed by methods that do not assimilate seismic data.
- The EnKF showed resiliency to noise parameter variations, with a wide range of regularization parameters yielding similar results when noise was simulated.
Limitations and Future Work
- Permeability Update: The current study updates only the CO₂ saturation, while the permeability field remains fixed for each ensemble member. The authors acknowledge that to achieve truly accurate predictions, the permeability field itself needs to be updated.
- Future Work: The authors plan to extend their work by applying the EnKF to update permeability and porosity fields along with CO₂ saturation, building on previous research that has explored similar data assimilation strategies. They also aim to investigate the impact of nonlinear seismic observations and explore end-to-end inversion for permeability.
Summary updated