In CGRE, we develop novel methods to address the complex geometries encountered in structural geology. Our methods are specifically tuned to deal with scarce data in a formal probabilistic manner.
We investigate different mathematical models to construct complex geometries, e.g. based on gaussian processes, NURBS and subdivision surfaces and combine them with innovative probabilistic methods, such as Probabilistic ML, gradient based samplers, as well as the visualisation and interface of these methods, e.g. VR/ AR.
In order to achieve the sustainable developments goals of the UN, werequire predictions of the processes in the earth’s subsurface.Unfortunately, we do not have direct access to the subsurface. Hence, werequire numerical simulations based on physical models to obtain thesepredictions, nonetheless. To account for complex model geometries andthe correct in-situ conditions, as well as area wide information we canneither rely on analytical nor laboratory analyses. Therefore, we requirenumerical simulations to characterize the complex coupled physicalprocesses of the earth’s subsurface. The processes include, for instance,fluid and heat transport, chemical species transport, and mechanicalprocesses.
We develop and use diverse machine learning methods to address spatial geoscientific problems.
Recent developments include a combined Hidden Markov Random Field (HMRF) - Gaussian Mixture Model (GMM) method to perform integrated spatial analyses of geophysical EMI measurements and Remote Sensing information.
In the research project "MINERALS", we are developing addressing the lack of open training sets for machine learning methods with a compilation of fully segmented thin section images.
CGRE is involved in several research projects, focusing on quantification of uncertainties in 3D-geological models and their influence on process-based simulations, as well as methods to make these computational simulations more efficient.