Two-dimensional MT inversion method based on an improved preconditioned nonlinear conjugate gradient algorithm

Key words:  magnetotelluric (MT)  preconditioned nonlinear conjugate gradient (PNLCG)  regularization parameter  self-adapted algorithm  inversion

Abstract:

Accuracy and computational efficiency are conflicting to each other in MT inversion methods. Gauss-Newton methods have high accuracy but poor efficiency, while nonlinear conjugate gradient methods are computationally efficient but less accurate. A new preconditioned nonlinear conjugate gradient method is proposed to improve both accuracy and computational cost by using a precondition operator whose performance is approximated by Gauss-Newton Hessian matrix. A self-adapted algorithm updating the regularization parameter is used to guarantee the balance between the stability and the accuracy of the inversion. Through the model tests and comparisons with other methods, the proposed method is proved achieving the goal. The method is applied to invert the MT data measured from a region in west China. The inverted resistivity profile is conjointly interpreted with seismic migration profile, and illustrates its significance in complicated structural interpretation.