Spatially and Conductivity Log Constrained AEM Inversion

Abstract

SUMMARYWe have developed an algorithm and released open-source code for 1D inversion of airborne electromagnetic data incorporating spatial and conductivity log constraints. The deterministic gradient based inversion algorithm uses an all-atonce approach, in which whole datasets or flight lines are inverted simultaneously. This allows spatial constraints to be imposed while also ensuring the inversion model closely matches any downhole conductivity logs that are near to the flight lines. The intent of the algorithm is to improve consistency along and across flight lines by taking advantage of the assumed coherency of the geology.Instead of roughness constraints, 'sameness' constraints are used. To implement these the regularization penalizes differences between the conductivity of 1D model/layer pairs and the weighted average conductivity of every other neighbouring 1D model within a user selected radius of their position. The neighbour averages are computed with inverse distance to a power weighting. The comparisons can be made over equivalent elevations. Downhole conductivity log constraints are imposed in a similar fashion, by penalizing the differences between conductivity logs, averaged over selected intervals, with their respective neighbouring 1D models. Overall the regularization encourages the final 1D conductivity models to be as similar as possible to their neighbours and to conductivity logs.It is demonstrated with an example that the method enhances geological interpretation by improving the model's continuity along and between flight lines, and its match to conductivity logs.

Applied geophysics for cover thickness mapping in the southern Thomson Orogen

“…The 2016 AEM survey helped to spatially constrain the basement highs located in 2014, and mapped new shallow basement under cover of the Eromanga Basin in the Louth and Enngonia areas of NSW. Interpretation products were produced using the Geoscience Australia reversible-jump Markov chain Monte Carlo (GARJMCMC) inversion (Brodie, 2015;Brodie & Reid, 2013;Brodie & Sambridge, 2012), and a new algorithm in development, the Geoscience Australia sample-by-sample all-at-once algorithm, which is due to be released to the public in 2018 (Brodie & Ley-Cooper, 2018).…”