Spatially constrained spectral clustering on AEM-based resistivity models

Abstract

AEM (Airborne Electro-Magnetic) data is data collected by inducing electromagnetic currents in the earth with helicopter-flown equipment, and measuring the magnetic response. Since collecting AEM data is relatively efficient and inexpensive, The Geological Survey of the Netherlands has a lot of this data available. AEM measurements can be interpolated to construct a 3D resistivity model that consists of a grid containing the electrical resistivity of the terrain at each point. Converting these resistivity models to useful geological models is a difficult task, mainly because of the complex relationship between electrical resistivity and geology. The conversion can be done manually by a domain expert, but this is not feasible on a large scale since it would be too time-consuming. In this thesis we present a clustering method to automatically detect geologically meaningful regions in resistivity models. We describe geologically significant clusters as clusters that are both spatially contiguous and have a coherent range of resistivity values. We combine the resistivity data with existing borehole data and perform experiments with a spatially constrained spectral clustering method. The results show that some existing geological models can be approximated with this method.