A Comparison of different methods for choosing regularization parameter in regularized MT inversion

Abstract

Geophysical inversion is ill-posed. We can get a stable result not only from high resolution observed data, but also using the regularization methods to add stabilizing functional to increase the stability of the solution. The conjugate gradient method is used in Occam's inversion to improve the efficiency of inversion operator. By establishing a layered electric model, we use L-curve, GCV (Generalized Cross Validation) and UPRE (Unbiased Predictive Risk Estimator) to select the optimal regularized parameter. Through analyzing the characteristic of each ways, L-curve method is very stable and the result is appreciable, the result of GCV or UPRE is also well and tends to overfit the data slightly.