Mats Lundh Gulbrandsen
A thesis submitted January 2017 for the degree of Doctor of Philosophy and defended January 20, 2017.
The PhD School of Science, Theoretical Quantum Optics, Faculty of Science, Niels Bohr Institute, University of Copenhagen
Quantitative geological modeling based on probabilistic integration of geological and geophysical data
In order to obtain an adequate geological model of any kind, proper integration of geophysical data, borehole logs and geological expert knowledge is important. Geophysical data provide indirect information about geology, borehole logs provide sparse point wise direct information about geology, and the geologist’s job is to combine these sources of information with his or her own knowledge about lithology and geological structures and develop geological models. Large and data-rich geophysical surveys make this job extremely difficult. With a manual interpretation approach it is extremely time demanding and practically impossible to develop geological models that are consistent with all available data in an objective fashion. This thesis addresses these issues, and presents new methodologies and workflows, which are developed to assist the geologists in their work on developing plausible and reliable geological models. The work is manifested in two main directions. One direction focuses on how to fast and reliably be able to map geological boundary layers that uses all available geophysical data, treat all data consistently and at the same time treasure geological knowledge. For this purpose a methodology entitled Smart Interpretation is developed. This semi-automatic method learns the relation between a set of data attributes extracted from deterministically inverted airborne electromagnetic data and a set of interpretations of a geological layer that is manually picked by a geological expert. This relation can then be used to predict the interpreted geological layer, throughout the whole geophysical survey. Two applications of this method are presented. In one study, the distribution of permafrost in the Yukon Flats, Alaska is mapped, and in the other study, Smart Interpretation is using well-log data to automatically interpret the base of aquifer in Morrill, Nebraska. The other direction of the thesis is related to seismic inversion. The aspects of a probabilistic inversion of a seismic trace are presented, with the focus on how to properly integrate the geological information when defining the prior distribution. Finally a study addressing the problems of how to interpret and visualize results from a probabilistically defined inverse problem in a way that is meaningful in a geological point of view is presented.