Inversion of Magnetotelluric Data into 1D Resistivity Models Via Cumulative Resistance Mapping

Inverting magnetotelluric (MT) surface data to obtain subsurface resistivity models is a complex and non-linear process where the solution is inherently non-unique. We present a fully data-driven method that enables the direct transformation of MT data into 1D resistivity models using cumulative resistance models. Our approach introduces a cumulative representation of a layered resistivity model, which, at each depth, integrates the effect of overlying layers into the subsurface model. We then establish a relationship between the real part of the TE mode of the MT data and its corresponding cumulative resistance model. Subsequently, we use this relationship to train a neural network that rescales MT data directly into cumulative resistance models. Once the resistance model is retrieved, a numerical derivative is applied to obtain the interval resistivity model, without any prior assumptions about the subsurface structure or resistivity distribution. This approach was validated using both synthetic and real MT datasets, introducing a new perspective for tackling the inversion problem.