Airborne electromagnetic 3D modelling and inversion

Abstract

Numerical EM 3D modelling and inversion are computationally challenging problems that can be solved by traditional methods only for models with a rather limited number of parameters. This severely restricts their application to realistic geophysical problems. In an effort to ease this restriction a number of levels of approximate forward modelling have been implemented including the 1D approximation, the 3D Born approximation, the finite element method, and the direct hybrid solution. A comparison of these modelling methods for the simulation of airborne EM data over a 3D conductivity earth model shows that the 1D approximation is very fast and gives a reasonable qualitative response, the Born approximation is moderately rapid but gives an inaccurate response, and the finite element and direct hybrid methods give similar responses with the finite element method requiring significantly less computer resources. The corresponding inverse problem is more challenging than the forward problem because each method of solution for the forward problem may require a different optimisation method for the inverse problem. This occurs because of the trade-off between the CPU time and accuracy of the forward modelling and the CPU time and number of forward modellings required by the optimisation method. Inversion tests based on the same airborne EM simulation show that for moderately large models the preferred optimisation method is quasi-Newton however for smaller models or when the finite element or hybrid modelling methods are used then the preferred method is Gauss-Newton.