Bayesian inversion of controlled source electromagnetic data for a laterally variable subsurface

Solving for subsurface models in geosciences often requires sophisticated mathematical inversion algorithms. A rigorous way of approaching ill-posed problems (due to, for example, limited physical data) are probabilistic (Bayesian) methods, where data is combined with physically meaningful prior information through sampling algorithms such as Markov chain Monte Carlo (MCMC). Bayesian techniques have been shown to rigorously quantify model parameters and their uncertainties, an advantage over standard schemes.

This project aims to develop an inversion algorithm for a depth and laterally-variable (two-dimensional, 2D) sub-seafloor electrical conductivity distribution using marine controlled source electromagnetic (CSEM) data. The CSEM method is a popular tool to detect conductivity contrasts in the seabed that may relate to, e.g., hydrocarbon reservoirs. We successfully implemented a Bayesian inversion for CSEM data for a model parametrisation where conductivity varies only with depth (one-dimensional, 1D). However, many areas cannot be sufficiently described in 1D. State-of-the-art algorithms and faster processing capabilities enable progress towards 2D algorithms. Hence, this project aims to develop a 2D algorithm that allows rigorous interpretation of CSEM data from geological targets for which a 1D approach is inadequate. It will be applied to real data sets, e.g., in the Black Sea and offshore Svalbard.

This project will evaluate, extend and develop state-of-the-art Bayesian approaches to geophysical data. It will build on existing 1D probabilistic (Bayesian) inversion techniques and CSEM forward simulators. Bayesian inversion has advantages over classical inversion techniques, because it only requires the forward simulator to sample the model space (usually via MCMC, a memoryless random walk). It does not need linearization or regularization, but allows inclusion of appropriate prior information and constraints, and does not get trapped in local minima. Most importantly, it can provide rigorous uncertainty quantification.

The student will develop new methods that extend current capabilities to address 2D inversion by involving (depending on the student’s interest) techniques such as efficient sampling algorithms, computation optimisation, statistical emulation methodology, data selection and automated adaptation of model complexity.

The algorithm will be applied to available CSEM data to estimate upper and lower bounds for the conductivity distribution. A combined geological interpretation will be obtained including other geophysical data (such as seismic reflection and in-situ petrophysical data) and collaborating with international researchers.

All doctoral candidates will enroll in the Graduate School of NOCS (GSNOCS), where they will receive specialist training in oral and written presentation skills, have the opportunity to participate in teaching activities, and have access to a full range of research and generic training opportunities. GSNOCS attracts students from all over the world and from all science and engineering backgrounds. There are currently around 200 full- and part-time PhD students enrolled (~60% UK and 40% EU & overseas).

The student will be registered at the University of Southampton. The student will join the UK’s most active marine geophysics group and the Southampton Statistical Sciences Research Institute. The student will have the opportunity to participate in marine science at sea, and will have access to a range of relevant high-level courses, and state-of-the-art modelling hardware and software. The student will receive training in electromagnetic methods, inversion theory, Bayesian methods, and will attend the national “Academy for PhD Training in Statistics”. The student will be introduced to international experts on Bayesian inversion of geophysical data and develop strong communication skills and build contacts for future employment in research or industry.