Stochastic approach to data uncertainties

We encountered a common problem while constructing models for resource analysis of Enhanced Geothermal Systems. As reservoirs are typically in great depth (i.e. 3-5 km) and of lower financial yield compared to oil reservoirs, raw data is sparse and expensive. The quality of the data might vary significantly. Furthermore, data is usually not equally distributed in the whole modeled volume space. Thus, after constructing the model, the question remains: where is it accurate, what is fact, what is fiction?
The quality of geological interpolation techniques has already been intensely studied. In our approach, we identify the effect of uncertainties of the input data themselves on the model result. We show that these uncertainties can be tracked during the model construction and identified in the resulting 3-D model.
We address this problem using a stochastic modeling approach with the following workflow:
- Construct a geological model using the potential-field approach implemented in the software GeoModeller. This allows direct modeling of geological input data
- Identify input data quality and assign statistical values
- Use a statistical random-generator to disturb input data
- Construct new model based on randomly generated data set
- Iterate a number of times to reach desired significance
As a test of feasibility, we apply the method to the construction of a resource scale model for geothermal exploration. The results show how uncertainties of different data types interplay during model evolution. Finally, a geological model with identified accuracy is obtained.
Combined with the analysis of the interpolation technique, this leads to an integrated interpretation of the quality of a 3-D geological model for a large variety of possible applications.