Parallel inversion of drone-based electromagnetic data for near-surface geophysical prospecting

The rapid development of geophysical systems utilizing drones facilitates mineral exploration with more efficient and economical data collection. To align the progress of hardware advancement and meet the model complexity needs for exploration, we aim to develop an efficient and robust 3D inversion code to interpret the drone-based EM data. Here, we present the framework of the implementation and show some preliminary results of the development.

We use a total electric field formulation with curl-conforming Nédélec elements to solve Maxwell equations in the frequency domain. Octree grids are used to accommodate the meshing of even large models at adequate resolution, separately in forward and inverse domains. A direct solver (MUMPS) is applied to solve the linear system of equations of the forward problem. The code is implemented in C++ and allows for easy adaptation for various sources and data types.

Currently, to solve the inverse problem, we minimize the misfit using a Gauss-Newton scheme with explicit computation of the Jacobian. The implementation was built on the deal.II library, where the interface wrappers allow to use MUMPS and PETSc for numerically intensive computations, such as system equation solving (MUMPS) and inversion model update (PETSC Conjugate Gradient solver). Currently, the code is parallelized using MPI throughout for both forward and inverse modeling and additionally OpenMP for MUMPS only.

The code is planned to be a reliable and competent imaging tool, that can be applied for both commercial and educational use. Currently, the code is under the development and testing stage. The preliminary results will be shown on-site.