Trans-dimensional Bayesian inversion of frequency-domain airborne EM data

Traditional gradient inversion has been widely used in airborne EM data processing. However, the inversions are seriously affected by the starting model and can be easily trapped in local minima. To solve these problems, we apply the trans-dimensional Bayesian inversion in our airborne EM data inversion. In trans-dimensional Bayesian inversion, the candidate models are randomly sampled from the proposed distribution and screened by the probability of acceptance. The probability distribution for the inversion models are finally obtained that contains parameter uncertainties. We further introduce the weighting coefficients to adjust the constraints on inverse models, so that the inversions for deep conductive layer are largely improved. In addition, based on the original sampling and accepting criteria, by improving the statistic method, in which only those models that fit the data are brought into statistics, the disturbances from unreasonable models are weakened. Finally, we prove the validity of the method by contrasting the inversion results of the synthetic data contaminated with gauss noise and field data with Occam method.