Investigating the confidence in lithologies recovered from geophysical inversion results

We propose and test a methodology allowing the calculation of confidence levels in lithologies recovered a posteriori from constrained joint geophysical inversion results. Our technique relies on the self-organizing maps to classify the different parts of the recovered models accordingly with the values obtained from the different attributes characterizing inverse modelling results. This, in turn, allows us to estimate the percentage of error in terms of recovered lithologies for each class across the model. We test our approach using a realistic synthetic case study integrating geological, petrophysical and potential field geophysical data. Results show that we can estimate the confidence in recovered lithologies consistently with areas where they have been wrongly assigned. Our results also indicate that model cells characterized by high estimates of confidence values have a very low rate of false positives.