No abstract

SUMMARY The posterior distribution of earth models that fit observed geophysical data convey information on the uncertainty with which they are resolved. From another perspective, the non‐uniqueness inherent in most geophysical inverse problems of interest can be quantified by examining the posterior model distribution converged upon by a Bayesian inversion. In this work we apply a reversible jump Markov chain Monte Carlo method to sample the posterior model distribution for the anisotropic 1‐D seafloor conductivity constrained by marine controlled source electromagnetic data. Unlike conventional gradient based inversion approaches, our algorithm does not require any subjective choice of regularization parameter, and it is self parametrizing and trans‐dimensional in that the number of interfaces with a resistivity contrast at depth is variable, as are their positions. A synthetic example demonstrates how the algorithm can be used to appraise the resolution capabilities of various electromagnetic field components for mapping a thin resistive reservoir buried beneath anisotropic conductive sediments. A second example applies the method to survey data collected over the Pluto gas field on the Northwest Australian shelf. A benefit of our Bayesian approach is that subsets of the posterior model probabilities can be selected to test various hypotheses about the model structure, without requiring further inversions. As examples, the subset of model probabilities can be viewed for models only containing a certain number of layers, or for models where resistive layers are present between a certain interval as suggested by other geological constraints such as seismic stratigraphy or nearby well logs.

Bayesian methods can quantify the model uncertainty that is inherent in inversion of highly nonlinear geophysical problems. In this approach, a model likelihood function based on knowledge of the data noise statistics is used to sample the posterior model distribution, which conveys information on the resolvability of the model parameters. Because these distributions are multidimensional and nonlinear, we used Markov chain Monte Carlo methods for highly efficient sampling. Because a single Markov chain can become stuck in a local probability mode, we run various randomized Markov chains independently. To some extent, this problem can be mitigated by running independent Markov chains, but unless a very large number of chains are run, biased results may be obtained. We got around these limitations by running parallel, interacting Markov chains with “annealed” or “tempered” likelihoods, which enable the whole system of chains to effectively escape local probability maxima. We tested this approach using a transdimensional algorithm, where the number of model parameters as well as the parameters themselves were treated as unknowns during the inversion. This gave us a measure of uncertainty that was independent of any particular parameterization. We then subset the ensemble of inversion models to either reduce uncertainty based on a priori constraints or to examine the probability of various geologic scenarios. We demonstrated our algorithms’ fast convergence to the posterior model distribution with a synthetic 1D marine controlled-source electromagnetic data example. The speed up gained from this new approach will facilitate the practical implementation of future 2D and 3D Bayesian inversions, where the cost of each forward evaluation is significantly more expensive than for the 1D case.

S U M M A R YWe apply a reversible-jump Markov chain Monte Carlo method to sample the Bayesian posterior model probability density function of 2-D seafloor resistivity as constrained by marine controlled source electromagnetic data. This density function of earth models conveys information on which parts of the model space are illuminated by the data. Whereas conventional gradient-based inversion approaches require subjective regularization choices to stabilize this highly non-linear and non-unique inverse problem and provide only a single solution with no model uncertainty information, the method we use entirely avoids model regularization. The result of our approach is an ensemble of models that can be visualized and queried to provide meaningful information about the sensitivity of the data to the subsurface, and the level of resolution of model parameters. We represent models in 2-D using a Voronoi cell parametrization. To make the 2-D problem practical, we use a source-receiver common midpoint approximation with 1-D forward modelling. Our algorithm is transdimensional and self-parametrizing where the number of resistivity cells within a 2-D depth section is variable, as are their positions and geometries. Two synthetic studies demonstrate the algorithm's use in the appraisal of a thin, segmented, resistive reservoir which makes for a challenging exploration target. As a demonstration example, we apply our method to survey data collected over the Scarborough gas field on the Northwest Australian shelf.