A Data Science Approach to Uncertainty Evaluation in Fault-Seal Analysis for Exploration Applications in Siliciclastic Reservoirs

Abstract

Summary Fault-seal analysis is used to understand the risk of hydrocarbons leaking out of a trap across bounding faults. There is a broad standard industry workflow for this analysis in clastic rocks, involving building a structural framework of the key seismic horizons and faults, estimating the clay content of the stratigraphy, predicting fault clay content using an algorithm such as shale gouge ratio (SGR), converting the predicted clay content to capillary threshold pressure, and calculating sealing capacity using the reservoir fluid properties. The inputs to the analysis are typically subject to considerable uncertainty that is difficult to understand and evaluate. One approach is to create “end member” scenarios to evaluate the impact of the key uncertainties, but this gives an incomplete picture and only a limited number of scenarios can be run and analyzed in the time typically available. The approach outlined here instead uses an automated workflow that runs hundreds of scenarios with stochastically varying input parameters. Data-science techniques combining domain expertise, mathematics and statistics, and computer coding are used to analyze and build models of the results. The advantages of the approach over a traditional fault-seal analysis workflow are that a wider range of input uncertainties can be considered, and that the results of a large number of realizations can be consolidated and visualized in ways that are specific to the problem being addressed and useful to the geologist undertaking the analysis. A key advantage is that the uncertainty inherent in this type of analysis can be explicitly incorporated, and the impact of the uncertainty clearly understood and communicated to decision-makers. Similar techniques could be applied to other geological analyses where uncertainty is an issue.