Bootstrapping as a Means of Solution Ensemble Based Uncertainty Analysis in Geophysical Inversion Modelling

Abstract

Many geophysical models are created without satisfactory uncertainty analysis. Most geophysicists are aware of their modelâ??s limitations, but if the model is passed on to a third party, this information is lost and the risk of misinterpretation arises. This project develops multi-solution inversion techniques to improve inversion and joint inversion modelling of geophysical data in mineral exploration. The main focus is the advancement of the probability and uncertainty analysis of inversion models to increase their reliability. To create solution ensembles, a bootstrapping resampling approach is taken, which produces reduced data sets from a base data set by random omission of data points. Each of these new data sets is run through a conventional inversion process to produce a variety of solutions with minor variations. In the appraisal stage the solution ensemble is statistically analysed to infer model uncertainties, which are then visualised to allow easy communication of the results. The process yields a clear and easy to interpret uncertainty map for the connected model and we demonstrate its effectiveness with several case studies. Furthermore, we are currently investigating swarm intelligence based global search algorithms as a second approach to solution ensemble creation.