An investigation into tensor based magnetic field forward modelling and source detection

Abstract

The modelling of potential field data is often both time consuming and ambiguous. With the use of tensor datasets becoming more commonplace, efficient techniques to model these data are necessary. The two components of modelling, namely forward modelling and inversion are addressed. The derivation of tensor forward modelling equations is illustrated and it is shown that the forward modelling of voxel data, both in conventional and tensor form, is not only viable but also has efficiencies which are as good, if not better (when taking editing into account) than non-voxel techniques. The theoretical basis for this modelling is presented here. Source distance calculations is a form of inversion that provides a good starting model in an efficient way. New tensor equations utilising analytic signals were derived for both source distance and susceptibility calculations. The tensor forms offer the possibility of lower noise in the calculations when dealing with tensor data. The synthesis of results from these techniques into a final model is semi-automated and includes using cluster analysis methods such as DBSCAN. This allows for automatic determination of relevant features from the depth calculations. The presence of remanent magnetisation in data often presents problems in forward and inverse modelling. By deriving novel equations to directly calculate magnetic field direction cosines from the tensor magnetic components, it is possible to get an indication of the presence of remanence, direction of the remanent field and the complexity of remanence within bodies. Tests on real tensor data over the Tallawang deposit in Australia showed both strengths and limitations. In spite of not being a perfect dyke, calculations for depth and width produced solutions in the expected range. Direction cosine solutions over the body show a degree of complexity in the remanence, possibly due to the presence of magnetite in lenses, thereby suggesting a complex composition. The low Q-ratio and uncertainty in susceptibility for the area contributed to non-optimal solutions for total magnetisation, remanent magnetisation, inclination and declination. Synthetic modelling demonstrates that should the total magnetisation and susceptibility be accurately known, it is possible to accurately derive remanent magnetisation, inclination and declination. The use of actual tensor data as well as the derivation of tensor datasets from total magnetic intensity data showed that the process derived in this project not only is viable, but also achieves good results. The extraction of valid source distance solutions from raster data is straightforward and allows fast creation of the 3D starter model for the area, from which improvements can be made through further forward modelling.