Optimal selection of regularization parameter in magnetotelluric data inversion

Abstract

Inversion of magnetotelluric data is known as a nonlinear and ill-posed problem. To obtain meaningful and unique results, Tikhonov's regularization method is commonly used to solve it. The optimal selection of the regularization parameter is another important factor for achieving an ideal inverse modeling. The aim of the present study is to find the optimal value for the regularization parameter in a two-dimensional inversion of magnetotelluric data by introducing a novel method. Furthermore, the Lanczos bidiagonalization method has been used to speed up the inversion process. For this purpose, three common methods including L-Curve, Generalized Cross-Validation, and Discrepancy Principle were investigated and then compared with the Adaptive Regularization as a novel optimal method in the inversion of 2D magnetotelluric data. All methods were provided as the Matlab code by authors. A 2D synthetic MT data with 3% random noise and Bushli (Nir) geothermal field MT data in Ardabil province, in the NW of Iran, was used by the introduced method for demonstrating its efficiency. The obtained results affirm that despite the capability of all methods in selecting the regularization parameter, the introduced method is more efficient than other conventional methods in terms of required memory, elapsed time, convergence to the desired model in fewer iterations, and modeling accuracy. Morever, applying this method on real data demonstrates its ability to generate a realistic inverted model.