Published September 15, 2021 | Version v1
Conference paper Open

Inversion-based automatic processing of AEM data

  • 1. Universita Statale di Milano, Milan, IT, gianluca.fiandacaunimi.it

Description

Data processing is a mandatory step before inversion for any geophysical inversion, because data outliers significantly affect the inversion process, often forbidding to reach reasonable inversion models and misfits. In the processing of Airborne Electromagnetic (AEM) data, the specificity consists in the necessity to cull out capacitive and galvanic coupled data, with the latter more difficult to recognize in data space alone. In this study I propose to use a generalization of the minimum support norm, namely the asymmetric generalized minimum support (AGMS) norm, for defining the data misfit in the objective function of an iterative reweighted least squared (IRLS) gauss-newton inversion. The AGMS norm in the data misfit puts a cap on the weight of non-fitting data points, allowing for the inversion to focus on the data points that can be fitted. Outliers can be identified after the AGMS inversion computing a classic L2 misfit from the final inversion model. Inversions on AEM data with and without manual processing are compared, with the AGMS inversion able to recognize outliers in the same areas in which data are manually culled out because of coupling, with comparable final inversion models. Moreover, the processing scheme can recognize not only data which are affected by noise, but also data that are not modelled correctly, for instance because of the dimensionality of the forward response: in this case, it can be used for identifying the appropriateness of the modelling within the inversion area. This inversion-based automatic processing scheme is very robust and works well also with a significant number of outliers; furthermore, it is fully general and can be applied not only to AEM data, but to any geophysical problem simply using the appropriate forward modelling.

Notes

Open-Access Online Publication: March 03, 2023

Files

ID260.pdf

Files (1.3 MB)

Name Size Download all
md5:b446ec6c072855cbc5c8763955b3fe60
1.3 MB Preview Download