Synthetic geology – Exploring the "what if?" in geology

Abstract:

The spatial and temporal extent of geological phenomena makes experiments in geology difficult to conduct, if not entirely impossible and collection of data is laborious and expensive – so expensive that most of the time we cannot test a hypothesis. The aim, in many cases, is to gather enough data to build a predictive geological model. Even in a mine, where data are abundant, a model remains incomplete because the information at the level of a blasting block is two orders of magnitude larger than the sample from a drill core, and we have to take measurement errors into account. So, what confidence can we have in a model based on sparse data, uncertainties and measurement error?

Synthetic geology does not attempt to model the real world in terms of geological processes with all their uncertainties, rather it offers an artificial geological data source with fully known properties. On the basis of this artificial geology, we can simulate geological sampling by established or future technologies to study the resulting dataset. Conducting these experiments in silico removes the constraints of testing in the field or in production, and provides us with a known ground-truth against which the steps in a data analysis and integration workflow can be validated.

Real-time simulation of data sources can be used to investigate crucial questions such as the potential information gain from future sensing capabilities, or from new sampling strategies, or the combination of both, and it enables us to test many "what if?" questions, both in geology and in data engineering. What would we be able to see if we could obtain data at higher resolution? How would real-time data analysis change sampling strategies? Does our data infrastructure handle many new real-time data streams? What feature engineering can be deducted for machine learning approaches? By providing a 'data sandbox' able to scale to realistic geological scenarios we hope to start answering some of these questions.