On some issues regarding 3D‐gravity inversion

Summary Gravity surveys have been used in the investigations of oil and mineral explorations. The inversion of gravity data collected over a three dimensional earth provides meaningful construction of density contrast models and therefore enable to extract more useful information from the gravity data. However, a major difficulty encountered in the utilization of 3D-gravity is the non-uniqueness of the inverted models. If a model is found to fit the data, there maybe many other models that fit the data to the same degree. For example, an anomaly near the earth surface may have the same response of an deeper anomaly with higher density. To overcome this difficulty, we utilize the Euler deconvolution technique to generate the locations of the gravity anomalies, which introduces prior information into the inversion process. Our 3D-gravity inversion method is analogous to that of Li and Oldenburg (1998). Basically, we subdivide a 3D- volume directly beneath the survey area into rectangular cells each of which has constant but unknown density. We search for the optimum distribution of density in terms of minimizing objective function subject to fitting the observed data with a prescribed tolerance. The objective function includes terms that penalize the roughness in various spatial directions. We use a conjugate gradient technique to search for the optimum solutions while Oldenburg and Li (1994) utilize a linear subspace technique. The main advantage of the conjugate gradient technique is that it provides fast rate of convergence without storage of any matrices. We used our forward simulation algorithms to generate the vertical component of gravity field (Gz) on the ground surface and the airborne data Gzz which is the spatial vertical derivative of Gz. We applied our inversion technique to synthetic ground data and airborne data. The results of this work demonstrates that in some cases the Euler deconvolution technique plays important role in enhancing our 3D-gravity inversion.