Model complexity in carbon sequestration: A design of experiment and response surface uncertainty analysis

Abstract

Geologic carbon sequestration (GCS) is considered a promising means of reducing atmospheric carbon dioxide (CO2). In Wyoming, GCS is proposed for the Nugget Sandstone in Moxa Arch, a deep, regional-scale saline aquifer with a large CO2 storage potential. For a proposed storage site, this study builds a suite of increasingly complex conceptual geologic model families, using subsets of the site characterization data: a homogeneous model family (FAM1), a stationary petrophysical model family (FAM2), a stationary facies model family with sub-facies petrophysical variability (FAM3), and a non-stationary facies model family (with sub-facies variability) conditioned to soft data (FAM4). These families, representing alternative conceptual site models built with increasing data, were simulated with the same CO2 injection test (50 years at 1/10Mt (1.0×108kg) per year), followed by 2950 years of monitoring. Using the design of experiment, an efficient sensitivity analysis (SA) is conducted for all families, systematically varying uncertain aquifer parameters, while assuming identical well configuration, injection rate, bottomhole pressure constraint, and boundary conditions, i.e., the model is considered a part of a larger, semi-infinite system, where both the injected CO2 and the formation brine can flow out. The SA results are compared among the families to identify parameters that have 1st order impact on predicting CO2 storage ratio (SR) at two different time scales, i.e., end of injection and end of monitoring. This comparison indicates that, for this deep aquifer with a gentle incline, geologic modeling factors do not significantly influence the short-term prediction of the CO2 storage ratio. However, these factors become more important over the monitoring time, but only for those families where such factors are accounted for (in other words, their long-term importance cannot be revealed by the relatively simple conceptual models). Based on the SA results, a response surface analysis is conducted to generate prediction envelopes of the storage ratio, which are also compared among the families, and at both time scales. Results suggest a large uncertainty in the predicted storage ratio, given the uncertainties in model parameters and modeling choices: the SR varies from 5–60% (end of injection) to 18–100% (end of monitoring), although its variation among the model families due to different modeling choices is relatively minor. Moreover, long-term leakage risk is considered small at the proposed site. This is because, in the lowest-SR scenarios, all model families predict gravity-stable supercritical CO2 migrating toward the bottom of the aquifer. In the highest-SR scenarios, supercritical CO2 footprints are relatively insignificant by the end of monitoring.