3-D NUMERICAL MODELING OF MAGNETOTELLURIC BASED ON EDGE FINITE ELEMENT METHOD WITH RESTARTED BICGSTAB MATRIX SOLVER TECHNIQUE

• Main Author: Susilawati, Anggie
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84908
• Institution: Institut Teknologi Bandung

Three-dimensional magnetotelluric forward modeling has been successfully developed using the Galerkin finite element method with a residual-based approach. The modeling algorithm is based on the Helmholtz equation for electric field vectors and uses interpolation functions based on the edges of hexahedral elements. The linear system is solved using the iterative restarted BiCGSTAB method with a Jacobian preconditioner. Several models were simulated to test and validate the forward modeling algorithm developed in this study. The homogeneous earth model was applied to validate the accuracy of the forward modeling algorithm in this study by comparing numerical results with analytical solutions. The results show good agreement with small relative error values for apparent resistivity and phase impedance. Then, the second test model used the COMMEMI 3D-1A model. The results of the COMMEMI 3D-1A test model using the algorithm in this study show good agreement with the response of the COMMEMI 3D-1A model for both components at a frequency of 10 Hz. The third test model using DTM1 produced a response that agreed with the response of Mackie (FD), Miensopust (FD), and Han and Lee (FE) at high frequencies. Finally, several simple Earth models were simulated in this study to determine the response of MT data under these conditions and the stability and effectiveness of the forward modeling program. The forward modeling results were able to provide MT responses that agreed with the simulated models, as seen from the distribution of apparent resistivity and phase impedance. The use of matrix solution techniques with the iterative restarted BiCGSTAB method with a Jacobian preconditioner provided fast convergence at high frequencies with a residual error of 10-7 . However, at low frequencies, the calculation was difficult to achieve convergence and stopped at a certain iteration. Therefore, the confidence level of the calculation at low frequencies was not very high, but still had a small residual error within the range of 10-6 – 10-5 .