Benjamin Maveau

Regularization methods improve the stability of ill-posed inverse problems by introducing some a priori characteristics for the solution such as smoothness or sharpness. In this contribution, we propose a multidimensional, scale-dependent wavelet-based ℓ₁-regularization term to cure the ill-posedness of the airborne (time-domain) electromagnetic induction inverse problem. The regularization term is flexible, as it can recover blocky, smooth and tunable in-between inversion models, based on a suitable wavelet basis function. For each orientation, a different wavelet basis function can be used, introducing an additional relative regularization parameter. We propose a calibration method to determine (an educated initial guess for) this relative regularization parameter, which reduces the need to optimize for this parameter, and, consequently, the overall computation time is under control. We apply our novel scheme to a time-domain airborne electromagnetic data set in Belgian saltwater intrusion context, but the scheme could equally apply to any other 2D or 3D geophysical inverse problem.

If the magnetic field caused by a magnetic dipole is measured, the electrical conductivity of the subsurface can be determined by solving the inverse problem. For this problem a form of regularisation is required as the forward model is badly conditioned. Commonly, Tikhonov regularisation is used which adds the $\ell_2$-norm of the model parameters to the objective function. As a result, a smooth conductivity profile is preferred and these types of inversions are very stable. However, it can cause problems when the true profile has discontinuities causing oscillations in the obtained model parameters. To circumvent this problem, $\ell_0$-approximating norms can be used to allow discontinuous model parameters. Two of these norms are considered in this paper, the Minimum Gradient Support and the Cauchy norm. However, both norms contain a parameter which transforms the function from the $\ell_2$- to the $\ell_0$-norm. To find the optimal value of this parameter, a new method is suggested. It is based on the $L$-curve method and finds a good balance between a continuous and discontinuous profile. The method is tested on synthetic data and is able to produce a conductivity profile similar to the true profile. Furthermore, the strategy is applied to newly acquired real-life measurements and the obtained profiles are in agreement with the results of other surveys at the same location. Finally, despite the fact that the Cauchy norm is only occasionally used to the best of our knowledge, we find that it performs at least as good as the Minimum Gradient Support norm.

SUMMARY: The inversion of electromagnetic induction data to a conductivity profile is an ill-posed problem. Regularization improves the stability of the inversion and a smoothing constraint is typically used. However, the conductivity profiles are not always expected to be smooth. Here, we develop a new inversion scheme in which we transform the model to the wavelet space and impose a sparsity constraint. This sparsity constrained inversion scheme will minimize an objective function with a least-squares data misfit and a sparsity measure of the model in the wavelet domain. A model transform to the wavelet domain allows to investigate the temporal resolution (periodicities at different frequencies) and spatial resolution (location of the peaks) characteristics of the model, and penalizing small-scale coefficients effectively reduces the complexity of the model. The novel scale-dependent regularization term can be used to favour either blocky or smooth structures, as well as high-amplitude models in globally smooth structures in the inversion. Depending on the expected conductivity profile, a suitable wavelet basis function can be chosen. The scheme supports multiple types of regularization with the same algorithm and is thus flexible. Finally, we apply this new scheme on a frequency domain electromagnetic sounding data set, but the scheme could equally apply to any other 1-D geophysical method.

If a magnetic dipole is placed above the surface of the earth, the electromagnetic induction (EMI) effect, encoded in Maxwell's equations, causes eddy currents in the soil which, on their turn, induce response electromagnetic fields. The magnetic field can be measured in geophysical surveys to determine the conductivity profile of the ground in a non-destructive manner. The forward model used in the inversion of experimental data usually consists of a set of horizontal homogeneous layers. A frequently used analytical model, proposed by McNeill, does not include the interaction between the eddy currents and therefore fails for larger conductivities. In this paper, we construct a new forward, analytical, model to estimate the magnetic field caused by a horizontally stratified earth which approximates the interaction between eddy currents. This makes it valid for a broader range of parameters than the current state of the art. Furthermore, the error with the (numerically obtainable) exact result is substantially decreased. We also calculate the vertical sensitivity (“depth of exploration”) of the model and observe that it is in good agreement with the values obtained from the exact model.

If a magnetic dipole is placed above the surface of the earth, the Electromagnetic Induction (EMI) effect, encoded in Maxwell's equations, causes eddy currents in the soil which, on their turn, induce response electromagnetic fields. The magnetic field can be measured in geophysical surveys to determine the conductivity profile of the ground in a non-destructive manner. The forward model used in the inversion of experimental data usually consists of a set of horizontal homogeneous layers. A frequently used analytical model, proposed by McNeill, does not include the interaction between the eddy currents, and therefore fails for larger conductivities. In this paper we construct a new forward, analytical, model to estimate the magnetic field caused by a horizontally stratified earth but which approximates the interaction between eddy currents. This makes it valid for a broader range of parameters than the current state of the art. Furthermore, the error with the (numerically obtainable) exact result is substantially decreased. We also calculate the vertical sensitivity ("depth of exploration") of the model and observe that it is in good agreement with the values obtained from the exact model.