Towards plausible lithological classification from geophysical inversion: honouring geological principles in subsurface imaging
We propose a methodology for the recovery of lithologies from geological and geophysical modelling results and apply it to field data. Our technique relies on classification using self-organizing maps (SOMs) paired with geoscientific consistency checks and uncertainty analysis. In the procedure we develop, the SOM is trained using prior geological information in the form of geological uncertainty, the expected spatial distribution of petrophysical properties and constrained geophysical inversion results. We ensure local geological plausibility in the lithological model recovered from classification by enforcing basic topological rules through a process called "post-regularization". This prevents the three-dimensional recovered lithological model from violating elementary geological principles while maintaining geophysical consistency. Interpretation of the resulting lithologies is complemented by the estimation of the uncertainty associated with the different nodes of the trained SOM. The application case we investigate uses data and models from the Yerrida Basin (Western Australia). Our results generally corroborate previous models of the region but they also suggest that the structural setting in some areas needs to be updated. In particular, our results suggest the thinning of one of the greenstone belts in the area may be related to a deep structure not sampled by surface geological measurements and which was absent in previous geological models.