CO2 Dissolution Trapping: Can Geologic Framework Models be used to Capture such Storage?
Abstract
An efficient simulation methodology is developed to investigate fundamental complexity in modeling geological carbon sequestration, whereby upscaled hydrostratigraphic models (HSMs) with reduced characterization cost can be used to accurately model CO2 flow and storage. Based on a three-dimensional experimental stratigraphy which exhibits permeability (k) heterogeneity at multiple scales, a fully heterogeneous model (FHM) with 3.2 million grid cells is created. Using an image processing algorithm that can capture large-scale facies connectivity, three HSMs of decreasing heterogeneity resolutions are created with 8, 3, and 1 stratigraphic unit. To overcome the computation challenge of simulating these large models, a parallel flow simulator was written and verified. Increasing system ln(k) variances - 0.1, 1.0, 4.5 - are tested, leading to a suite of 12 conceptual flow models. Equivalent k tensors are then computed for each unit of the HSMs using a numerical upscaling technique. For all the HSMs, at all the variances tested, significant accuracy is achieved with the upscaled ks in terms of capturing both the FHM fluid head and flow connectivity. Using PFLOTRAN, a state-of-the-art massively parallel subsurface flow and reactive transport code [1], CO2 injection is simulated with all models at 2 km depth for 20 years, followed by 1000 years of monitoring. Compared to the FHM which provides the reference solution, when the variance of ln(k) is low, all HSMs yield similar reservoir fluid pressure, plume footprint, and dissolution fingering (and therefore the total predicted dissolution storage at the end of the simulation time) as the FHM. When the variance of ln(k) is high, the HSMs are still able to accurately capture the fluid pressure of the FHM, but they predict more dissolution fingering due to their increasingly homogenized representation of the reservoir permeability. The higher the level of homogenization, the stronger the predicted fingering is. On the other hand, fingering/dissolution is not apparent in the high-variance FHM because strong heterogeneity contributes to enhanced lateral flow, which leads to reduced dissolution fingering and less total aqueous CO2 storage. As a result of the reduced fingering, it takes longer for the supercritical CO2 phase to dissolve into the brine, or less dissolution is predicted by this model per unit time. Clearly, whether the geologic framework models, which are widely used for large- to regional-scale CO2 modeling, can accurately capture dissolution storage depends strongly on the reservoir permeability variability. To further screen for the factors which can exert significant influence on CO2 storage predictions over increasingly longer time scales, the computationally efficient design of experiments (DoE) sensitivity methodology is adopted. For the low ln(k) variance system, a set of preliminary DoE analysis suggests that at all time scales, brine salinity has the most important impact on the prediction of the dissolved CO2. Research is ongoing exploring the high variance system as well as in developing DoE-based proxy models to estimate uncertainties of different prediction outcomes. Reference: [1] PFLOTRAN, http://ees.lanl.gov/pflotran/
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H23B1269L
- Keywords:
-
- 1829 HYDROLOGY Groundwater hydrology;
- 1847 HYDROLOGY Modeling;
- 1873 HYDROLOGY Uncertainty assessment;
- 1805 HYDROLOGY Computational hydrology