Journal of Conference Abstracts

Analysis of Gravity Images with Specialized Wavelets

Peter Hornby (peterho@ned.dem.csiro.au), Fabio Boschetti (f.boschetti@ned.dem.csiro.au), Frank Horowitz (frank@ned.dem.csiro.au) & Louis Moresi (louis@ned.dem.csiro.au)

CSIRO Exploration & Mining, PO Box 437, Nedlands, WA 6009, Australia

We analyze gravity images by deriving a wavelet from the Green's function for 1/r potential fields. Uniting potential fields and wavelets provides new insight into both, e.g. we reinterpret upward continuation as a continuous scale change. Following Mallat and Zhong (1992), we skeletonize the gravity image by defining multiscale edges as the collection of gradient extrema at many scales. At the finest scale (i.e. without upward continuation) such multiscale edges are basically the features commonly drawn when interpreting a gravity map. The main lineaments, like faults and geological contacts, form an approximate 2-D geological map from the gravity image. However, the skeletonization extracts more information from the image because the position and scaling behaviour of the multiscale edges contain information about the position and geometry of the underlying density contrasts. The multiscale edges can be a visual aid in interpreting the underground geology, for example in estimating the dip of a fault or discriminating between bodies of different density.

Potential field inversion is non-unique. Correlating the multiscale edges with density contrasts sets a priori constraints on an inversion. The resulting piecewise constant sources represent a style of source variation with widespread geological applicability. We will show early results from such inversions.

The skeletonization contains the same information as the original potential field image, which can be reconstructed from the edges. However, the edges have the advantage of being local. We can exploit this fact in many ways, such as building new classes of image processing operations on gravity images, or by building inversions with fewer parameters and a huge reduction in computational complexity.

As an exercise in demonstrating how this technique can be used to interpret buried geological structures, we analyse the gravity fields resulting from numerical models of granitic intrusions within the crust. As the density field is known exactly for these models but also geologically realistic, we can test the ability of the edges to located interesting structures, and the ability of our inversions to recover the salient features of deep density variations.