Back to list of UTIG presentations at Fall Agu

Lopamudra Roy, Mrinal K. Sen, D. Blankenship, P. L. Stoffa, and T. Richter

Inversion of gravity data can result in several non-unique solutions. Therefore it becomes important to assign uncertainties in the resulting model parameters along with their covariances. Uncertainties arise due to several reasons, namely, noise in the data, approximation in mathematical formulation of forward problem, inappropriate inverse technique etc. Though one cannot remove these factors completely, uncertainty can be reduced considerably by using sufficient a priori information, which helps to restrict the inverted models to be within realistic geological setup. We have used simulated annealing (SA) based inversion technique to invert the gravity data in two and three dimensions. In our implementation of SA, we generated several thousand models with same control parameters but different starting models, selected from a prior distribution of model. Using a Bayesian framework, we use all these sampled models to construct marginal posterior probability density function (PPD), and several orders of moments. The correlation matrix shows the dependence of one parameter with other. Several correlation plots are used to study the effect of a-priori information in reducing the uncertainty in the solutions. We applied the technique to a few synthetic data and 3D field data over lake Vostok, East Antarctica. A priori constraints were developed from available seismic data and radar information. The inversion results produced a map of the structure of the basin along with their associated uncertainties.