Interpretation of Airborne Electromagnetic Data with Neural Networks

Abstract

Artificial Neural Networks (ANNs) are used for the interpretation of multi-frequency airborne electromagnetic (AEM) data independently of the sensor height, with one-dimensional (1-D) horizontally layered homogeneous earth structures. A divide-and-conquer strategy is applied. One ANN is trained to interpret data, which are best described by homogeneous half-space (HHS) models. A second ANN inverts data from horizontally layered half-space models with two layers (2LHS). Tests have shown that when the 2LHS ANN is applied to data, which are best, described with a HHS-like structure, interpretation errors can become large. Therefore, a third ANN is trained, which classifies the best interpretation of measurements as a HHS model or a 2LHS model. This modular ANN approach shows a good performance on synthetic data. Finally, the interpretation of data from an AEM survey over a tertiary basin structure, shows good accordance with known geological data.