We have developed a mineral exploration method for the joint inversion of 2D gravity gradiometry and magnetotelluric (MT) data based on data-space and normalized cross-gradient constraints. To accurately explore the underground structure of complex mineral deposits and solve the problems such as the non-uniqueness and inconsistency of the single parameter inversion model, it is now common practice to perform collocated MT and gravity surveys that complement each other in the search. Although conventional joint inversion of MT and gravity using model-space can be diagnostic, we posit that better results can be derived from the joint inversion of the MT and gravity gradiometry data using data-space. Gravity gradiometry data contains more abundant component information than traditional gravity data and can be used to classify the spatial structure and location of underground structures and field sources more accurately and finely, and the data-space method consumes less memory and has a shorter computation time for our particular inversion iteration algorithm. We verify our proposed method with synthetic models. The experimental results prove that our proposed method leads to models with remarkable structural resemblance and improved estimates of electrical resistivity and density and requires shorter computation time and less memory. We also apply the method to field data to test its potential use for subsurface lithofacies discrimination or structural classification. Our results suggest that the imaging method leads to improved characterization of geological targets, which is more conducive to geological interpretation and the exploration of mineral resources.

We present a joint 2D inversion approach for magnetotelluric (MT) and gravity data with elastic-net regularization and cross-gradient constraints. We describe the main features of the approach and verify the inversion results against a synthetic model. The results indicate that the best fit solution using the L2 is overly smooth, while the best fit solution for the L1 norm is too sparse. However, the elastic-net regularization method, a convex combination term of L2 norm and L1 norm, can not only enforce the stability to preserve local smoothness, but can also enforce the sparsity to preserve sharp boundaries. Cross-gradient constraints lead to models with close structural resemblance and improve the estimates of the resistivity and density of the synthetic dataset. We apply the novel approach to field datasets from a copper mining area in the northeast of China. Our results show that the method can generate much more detail and a sharper boundary as well as better depth resolution. Relative to the existing solution, the large area divergence phenomenon under the anomalous bodies is eliminated, and the fine anomalous bodies boundary appeared in the smooth region. This method can provide important technical support for detecting deep concealed deposits.