Advanced Inversion Techniques

The predictive power of our models depends on our ability to calibrate them against observations. Objective and robust calibration is thus an essential aspect of models for the imaging, conceptualisation and prediction of water, energy and mineral resources in the Deep Earth. This program seeks to develop the next generation of physics-based probabilistic inverse methods in combination with advances in data processing, machine learning, model parametrisation, data collection, forward modelling, and predictive applications.

The primary motivation for inference is to gain knowledge about a geological parameter or process of relevance to an explorer, for example, the depth-to-base of a paleochannel, or the geochemical processes related to the formation of mineral deposits. Geophysical inversions often have a preference for mathematically motivated parameterisations and geological parameters have to be derived through a qualitative interpretation of the inversion result. In contrast to this, a core aspect of this program is the development of targeted inversion methods that will allow direct inference for a parameter of relevance.

The true value of a Bayesian perspective is around predictive applications and the parametrisation employed by the predictive application defines the type of model which is sought from the data. If we, for example, seek to predict the water availability from an unconfined aquifer formed by a paleochannel, then near-surface geophysical data ought to be inverted for the structure and not for a smooth distribution of geophysical properties. The recovered structure and associated uncertainties can then be used to make predictions of water table uncertainty in a groundwater model.