One–dimensional magnetotelluric inversion by modified differential evolution algorithm

Traditional methods of solving geophysical inverse problems often used an initial structural model and an iterative process to find solutions. During each iteration, the model was linearized and updated, then the error between the model's response and the observed data were calculated and minimized. The inverse results solved by this method were often strongly dependent on the initial model choice, and the minimum of the objective function found could be the local value but not the global minimum. Global search methods such as genetic algorithm (GA) overcame these limitations of traditional solutions. This paper presented the application of modified differential evolution (MDE) algorithm which was developed from the genetic algorithm, to one-dimension (1D) magnetotelluric inversion. The algorithm had been used to invert two 3-layer geoelectrical models. The results showed that when using a 3-layer model for inversion, the obtained parameters of the subsurface almost coincided with those of the theoretical models. In the case the number of layers used for inversion had more than three layers (four layers and five layers), the obtained results showed that there was still a good agreement between the inverted structure and the theoretical model. Therefore, the algorithm was applied to one-dimensional (1D) inversion for data of one magnetotelluric station measured at Cu Chi district, Ho Chi Minh city. The results showed that the subsurface structure down to a depth of 16 km consisted of 3 layers: the top layer had a resistivity of 119 Ω.m with a thickness of about 760 m, the middle resistive layer with a resistivity of 4353 Ω.m with about 10 km thickness, and the lower half-space with a relatively low resistivity of 41.4 Ω.m. This result was consistent with the geoelectric structure information obtained from other magnetotelluric studies carried out in the studied area, demonstrating the practical applicability of the improved differential evolution algorithm.