Geological interpretation of potential field inverse models using automated classification

Abstract

Conventional interpretation of three dimensional potential field inverse models usually involves thresholding the model volume to create isosurfaces that hopefully outline the subsurface distribution of geological units. Unfortunately models are inherently smooth and the choice of the most appropriate value for isosurface generation is seldom clear-cut. However if both density and susceptibility models are available for an area then the combined dataset can be interpreted by classification using techniques developed for multi-band image data. Classification can be conducted using unsupervised or supervised techniques using either hard or soft classification algorithms. Supervised classification can be based on measured petrophysical data or on model values in areas where the surface or subsurface geology is well established. Soft classifiers are generally more appropriate than hard classifiers for this purpose since they better reflect the inherent geological ambiguity associated with often overlapping physical property distributions. Geological classification of potential field inverse models is illustrated with examples from the Northern Territory.