Smart Characterization

Studies have been performed to address site characterization for geological CO₂ storage

State of the art

Studies have been performed to address site characterization for geological CO2 storage (CO2CRC, 2008; DNV, 2009; NETL, 2010; SiteChar, 2013; Bachu, 2015; Heidug, 2013). This is a data, cost and time intensive process.

The initial steps is to collect all available data followed by screening, and detailed modeling using both static and dynamic models. For a depleted hydrocarbon site, large amount of data is usually available to the field owner including well logs, seismic surveys and historical production data. On the contrary, deep saline storage sites characterization will involve active data collection by drilling wells or performing new seismic surveys.

Site characterisation heavily relies on building a static model describing the structural and stratigraphic geology of the reservoir and the cap rock as well as populating this model with properties from available data. Decision on actual project development is based on assessing storage site uncertainties and reducing these uncertainties corresponds to active collection of new data.

Progress beyond the state of the art

This task will develop methodologies to optimise the data gathering and interpretation process to balance the acquisition costs with the insights this data provides. Statistical experimental design provides rules for resource allocation for information gathering. An experimental design approach for agile site characterization, where the process is continuously assessed and revised as the data being collected will be tested. For the highly nonlinear process of CO2 injection, optimal designs generally depend on the true values of the model parameters and since the data has not been collected yet, the model parameter values are not known. Bayesian methodologies for optimal experimental design (Huan & Marzouk, 2013) provide an elegant solution to this problem where prior distributions of the unknown parameters are postulated during the formulation of the experimental design problem. Solving the Bayesian optimal design involved maximizing the expected utility function over the different options of data collection with respect to the future observed data and model parameters. Two approaches are proposed:

1. Nested Sampling algorithm (Skilling 2006, Elsheikh et al. 2014) for performing the high dimensional integration of
   the expected utility function for solving the optimal experimental design problem.
2. meta-modeling techniques (Elsheikh et al. 2014, Petvipusit et al. 2014, Rohmer, 2014) for efficient solution of the
   optimal experimental design problem.

Outcomes

Methods and a modelling strategy (framework) on how to prioritize data collection, summarized in a report detailing the general procedure for smart site characterization and accompanied by a set of computer script (i.e. wrappers) utilizing commercial simulation tool. The streamlining framework of data use and acquisition will contribute to cost efficient characterization.