3D inversion of gravity data using Lp-norm sparse optimization

Three-dimensional inversion for density distribution has become a common tool for quantitative interpretation of gravity data in recent years. However, as gravity data lacks depth resolution, such inversion has severe non-uniqueness. To reduce this problem, the most common approach is to introduce extra prior information. To further perfect this tool, we propose a 3D sparse gravity inversion method which minimizes an L-p-norm (0 <= p <= 1) of the density model subject to bound constraints. Compared with the traditional L-2-norm inversion approach, our method permits to use known physical property information more effectively and produce solutions characterized by sharp boundaries and high depth resolution. To better understand our method, we analyze the equivalence between our method and binary or ternary inversion. Moreover, we also point out several issues of this method that need to be noticed. The validity of our method is tested by both synthetic- and real-data examples.