3-Dimensional (3D) geological models are created by modelling a series of data points depicting the location and elevation of the formation, member, or group tops in the subsurface. Surfaces are modelled using various geostatistical and geomodeling techniques to represent the often complex geological features that form each surface. Because the geological complexity for each surface is often unique, the uncertainty results generated for each model result are often unique as well.
Figure 1: The Bluesky Formation surface in 2-D (birds-eye view) and, b) 2.5-D (oblique view) in West Central Alberta.
Why do we Characterize Model Uncertainty?
Uncertainty analysis provides spatial information about the reliability of the 3D geological model. The prediction uncertainty can be measured by geostatistical tools to define the accuracy of the surface model at specific locations across the model area, and how it honours the geological features in the observed data. Uncertainty analysis is done for each formation separately; since the uncertainty varies from one formation to another due to unique geological complexities related to each surface. The uncertainty in each surface can be expressed both globally and locally. Global uncertainty provides a single numerical value characterizing the average uncertainty for the entire surface. Local uncertainty provides a measure of the uncertainty for numerous small zones across the model area. Uncertainty often varies dramatically across each study area; therefore providing uncertainty maps showing the local uncertainty gives more information to the user about specific locations of high and low uncertainty across the model area.
Uncertainty maps are visualizations of the kriging estimation standard errors across the model area. The estimation error is the difference between the observed data and the predicted data. The uncertainty map represents a useful visualization tool that allows the geostatistician and geomodeler to identify areas within a formation that have high uncertainty. Regions with high uncertainty values are isolated and sent back to geologists to confirm the accuracy of the available data. Minimizing uncertainty is critical to building geologically accurate surfaces which honor the observed data.
Figure 2: Uncertainty map of the Swan Hills Formation showing the highest uncertainty in red, corresponding to areas with no data. The lowest uncertainty (blue) represents zones with lower estimation error, and are often associated with areas of abundant data.
Areas of high uncertainty are often the result of sparse (insufficient) data. Some formations do not have enough data to sufficiently characterize the geology across the entire model area. Uncertainty associated with sparse data can be reduced by using different geostatistical parameters designed to expanding the search radius in the process of estimation, thus including more data.
Another cause of high uncertainty can be extreme high and low values, also known as outliers. Once an outlier has been identified, it is sent to a geologist to be verified and either corrected (changed) or retained (not changed) as the data in question may represent ‘real’ geological complexity in that area.
Uncertainty associated with geological modelling is inevitable, and as such it is essential that we continue to strive to find better ways to communicate model uncertainty to users and decision makers.