A sensitivity analysis of geological uncertainty to data and algorithmic uncertainty

Abstract:

How do cakes end up being unique when baked by different persons following a similar simplified recipe? Actually, people do not use the same tools, do not apply the same rigor in measuring ingredients and have a unique interpretation of the recipe. Similarly, modellers using the same initial geological data set will produce a wide variety of geological models. However, in practice, a very limited number of persons will generate geological models. In order to avoid uncertainty underestimation when the purpose of modelling is decision-making, uncertainties related to observations, algorithms and conceptual representations should be propagated in the generation of stochastic geological realization ensembles.

Here, we focus on the sensitivity of data and algorithmic uncertainties on the resulting geological uncertainty. Indeed, it might not make sense to compare a pie with a cake or a mousse. This is why we leave conceptual uncertainty aside and in the hands of model selection techniques. While data errors can be estimated by repeating some measurements, algorithmic uncertainties might be more complex to define and are not always accessible. To handle that, we propose to rely on the use of pilot-stick perturbations, which consists in adding fictive drill-holes complying with the assumed stratigraphy and the presence or absence of surface geological information.

The sensitivity analysis is performed on a synthetic case, based on a Precambrian basin setting, with three different geological modelling engines. The resulting geological uncertainty is analysed with different indicators based on the cardinality, entropy, connectivity, topology and geostatistics of both lithological formations and their underlying scalar-fields. Preliminary results show the pre-dominant importance of pilot-stick perturbations and their ability to mitigate the smoothing resulting from implicit modelling, in particular at locations where no surface data is available.

Acknowledgement

This work is supported by the ARC-funded Loop: Enabling Stochastic 3D Geological Modelling consortia (LP170100985) and DECRA (DE190100431) and by the Mineral Exploration Cooperative Research Centre whose activities are funded by the Australian Government's Cooperative Research Centre Programme. This is MinEx CRC Document 2020/39.