Geoscientific data, which is analysed and integrated to develop a clear understanding of subsurface composition, is being produced in an ever-increasing variety of modalities. However, characteristics inherent to these data can complicate analysis b traditional statistical methods. These characteristics include high levels of measurement and label noise, high dimensionality, and spatial paucity of ground truth. This thesis reports the development and evaluation of three kernel-based machine learning approaches for analysing geological logs, multi-element geochemical assays of drill core, and 3D geophysical inversion models all sourced from a common case study: the Kevitsa Ni-Cu-PGE deposit (Lapland, Finland).
|Qualification||Doctor of Philosophy|
|Award date||22 Jan 2019|
|Publication status||Unpublished - 2019|