Two-dimensional Bayesian inversion of magnetotelluric data using trans-dimensional Gaussian processes.

Area of Science:

• Geophysics
• Electromagnetism
• Computational Science

Background:

• Bayesian inversion of electromagnetic data is vital for subsurface resistivity uncertainty.
• High computational costs limit Bayesian methods to 1-D models.
• 2-D and 3-D inversions remain computationally challenging.

Purpose of the Study:

• To demonstrate the first fully 2-D, trans-dimensional Bayesian inversion of magnetotelluric (MT) data.
• To make computationally expensive Bayesian inversions tractable for 2-D resistivity models.
• To quantify subsurface electrical resistivity uncertainty in 2-D.

Main Methods:

• Utilized a Gaussian process (GP) for parsimonious model parametrization.
• Employed a trans-dimensional, parallel tempered Markov chain Monte Carlo sampler.
• Integrated GP with the MARE2DEM finite element forward solver for dense mesh computations.

Main Results:

• Successfully inverted synthetic and field MT data for 2-D electrical resistivity models.
• Achieved a parameter reduction of over 32× compared to traditional methods.
• Generated resistivity probability distributions quantifying 2-D model uncertainty.

Conclusions:

• The novel trans-dimensional GP sampler enables computationally tractable 2-D Bayesian MT inversion.
• The method provides quantitative uncertainty information crucial for geophysical interpretation.
• This technique can inform estimations of subsurface physical properties like porosity and fluid content.