Early-stage saline aquifer CO2 sequestration projects are filled with uncertainty and often pose a challenge for decision makers. The industry-standard saline aquifer suitability screening checklists provide limited insight into sequestration opportunity high-grading and development planning. Detailed full-field geological and numerical simulation modelling can be prohibitively expensive in the initial project scoping stage. Here we present an alternative data-driven approach to assist in early-stage CO2 sequestration project scoping. A developed machine learning model ties geological variables (predictors) with CO2 sequestration key variables (responses), such as cumulative CO2 injected, number of injection wells, injection rate, and CO2 plume diameter. The dataset underpinning the model consists of over 200 numerical simulation sensitivity runs. The Bayesian regression approach has been utilized allowing to express the model parameters with probability distributions. The novel machine learning model predicts key project variables, advancing the understanding of future saline aquifer development plans. The estimates are expressed as probability distributions, which allow for an uncertainty assessment that is tied to critical project decision-making variables. These results can be further used to determine the impact of geological variables on project key variables. The demonstrated novel application of Bayesian regression allows to assess uncertainty of project key variables and can be used to reduce the risk of CO2 sequestration projects prior to project decision-making. The solution provides the basis for economic modelling and accelerates the understanding of future saline aquifer development plans and the impact of key uncertainties.

You can access this article if you purchase or spend a download.