Exploring magnetotelluric model space

Abstract

Magnetotelluric (MT) inversions are inherently non-unique. Due to the large computational requirement of 3D MT inversion, there is always a trade-off between exploring model space and the amount of time invested in the inversion process. A standard approach is a two-stage inversion, where coarse features are resolved first, and the outputs from the coarse model are used as the starting model for a finely discretized inversion. Inversion may be followed by a combination of sensitivity analysis and forward modelling to test the robustness of features. This approach quickly leads to a low RMS model but has a limited capacity to explore the potential range of acceptable models. In this paper, we propose a revised two-stage process which enables fast comparison of inversion results from a series of inversion parameters. Three inversions were run using a starting half space with only the input data varied: one inversion using un-rotated full tensor and tipper data; one inversion using only tipper data; and the third inversion using rotated tensor and tipper data. In addition to these half-space models, an inversion was run using starting models based on geological constraints. It used three domain boundaries with roughness penalties turned off between units. The modified two stage inversion process suggested in this paper offers a balance between arriving at final inversion within a reasonable time frame while still allowing the user to explore a larger range of possible models than the original methodology.