Optimizing an Existing Inversion Code

Airborne ElectroMagnetic methods (AEM) have been used for geophysical surveys for several decades, in particular due to its ability to cover large areas on a limited budget and in relatively little time. Ever since the method was first deployed the tendency has been for the typical survey size to be growing at a very rapid rate, in particular during the last decade. AEM surveys of today are often in the order of tens of thousands of flight line kilometres, producing huge datasets that needs to be inverted. Performing a full non-linear inversion of such datasets is a tremendous computational burden and consequently, these datasets are very often inverted using approximate inversion schemes. Several full non-linear inversion codes for AEM data is reported in the literature, however, most of these codes operate on single soundings only and are not optimized for large datasets. We discuss how the existing constrained inversion code em1dinv has been optimized for modern parallel computing and show performance results. The parallel optimized code is capable of handling very large datasets and provides a significant reduction in the total inversion time. Furthermore, we show how the optimized code eliminates inversion artefacts inherent of the previous implementation.