Probabilistic analysis of EM data sensitivity and inversion accuracy

Abstract

Towed streamer EM data offer a possibility to generate a resistivity model of the earth within the sensitivity range of the EM survey, using procedures such as inverse modelling. There is, however, an inherent non-uniqueness in the problem due to noise, uncertainties, finite number of measurement positions and field components. To assess the model uncertainty, and analyse the data information content, we propose the Bayesian method of calculating probabilities for model parameters within a given set of models. Real towed streamer EM noise applied to synthetic EM data in a 3D model, similar to Barents Sea conditions, was used for the evaluation. The result of this work is a formulation of the posterior probability distribution for a set of sub surface resistivity model parameters. By analysing these probability functions we find that we are able to evaluate how a change in the data, e.g. different frequencies, sensor positions, noise levels or complexity of background, affect the probability of finding a model close to the true model.