Multi-source data fusion for geological boundary modelling via Gaussian Processes

Multiple data sources can be fused to improve boundary modelling in stratified deposits such as the banded iron formation (BIF) hosted deposits in the Pilbara Region of Western Australia. Accurate boundaries are essential to produce high quality geological models. Initial models are based on the widely spaced exploration hole data. This creates a high level of uncertainty in areas between the holes, with different interpretations of the boundary equally probable. Including directional information collected in these holes in the models can reduce the number of possibilities. Including data collected during mining, such as blastholes and bench measurements, can further guide the model. This research demonstrates a Gaussian Processes based method of fusing these disparate data sources to better inform the boundaries. Our results demonstrate that when only exploration points are used the model creates a smooth boundary with minimal variation between the holes. Each additional data source refined the model and added additional detail to the boundary. For the blast data, the new detail projects down to the bench below the current bench to allow for a better prediction before mining commences on that bench.