Enhancements to deterministic AEM inversion through better geometry constraints and a bunch-by-bunch algorithm

We have enhanced the existing program for the 1D deterministic inversion of airborne electromagnetic data called GALEISBSTDEM. The new source code will be publicly released in 2023. The updates include a number of improvements to the usability of the code including additional input and output options, null value handling, data culling, parameter bounds and a better build system. A more conservative line-search has been employed for determining the regularisation parameter at each iteration and for the parameter-change vector step-lengths. This has improved the robustness of the convergence of the algorithm and substantially reduced the number of narrow vertical striation artefacts in conductivity sections. The new release will also include the functionality to perform an inversion on the combined amplitude of the response in the XZ-plane rather than on the X- and Z-vector component data. This removes the dependence of the receiver bird pitch on the modelled response and allows the data to be fitted more easily, resulting in more laterally coherent conductivity sections with fewer artefacts. We have also introduced the concept of a bunch-by-bunch inversion in which several soundings are inverted in a single minimization problem. The bunch, or window of soundings, is then moved along the flight line until the full data set is inverted. The approach allows along-line constraints to be placed upon the layer conductivities as well as any elements of the system geometry that are being inverted for. This is a feature that cannot be attained with conventional sounding- by-sounding approach. In addition to the usual constraints on the smoothness of the vertical conductivity profile, various types of along-line constraints may optionally be applied. These include conventional linear smoothness constraints, plus we also add similarity constraints and what we call a cable-length constraint that we can apply to the inversion of the transmitter-receiver separation for fixed-wing AEM systems. The inclusion of the along-line constraints are designed to make the inversion of the elements of AEM system geometry more stable and realistic as they are constrained to follow a smooth trajectory that is dictated by the dynamics of a high-drag towed receiver bird. We find that the bunch-by-bunch approach has so far not delivered the degree of improvements in the level of data fitting for the X- and Z-vector component inversions that we had originally hoped for. However we are encouraged by some reduction in the level of data misfit and a corresponding improvement in the along-line smoothness of the inverted geometry parameters.