AMC presents at Geological Society Conference

AMC Consultants (AMC) is pleased to contribute the Mine Smarter perspective at the upcoming Mineral Resource Estimation: Recent Advances and Current Best Practice Conference hosted by the Geological Society of London. The conference will take place in London on 22 October from 9.30am and will finish with a drinks reception. AMC principal Geologist, Mark Burnett, will present on the application of machine learning in palaeoplacer exploration during session one.

The Geological Society of London is a not-for-profit organisation, and a registered charity that aims are to improve knowledge and understanding of the Earth, promote Earth science education and awareness, and promote professional excellence and ethical standards in the work of Earth scientists, for the public good. AMC’s office in the UK looks forward to participating in this important event.

The application of machine learning in palaeoplacer exploration

Monday 22 October, 2018 | 11.15-11.30 AM | The Geological Society (Burlington House)

Machine learning, a branch of artificial intelligence, is based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention. Current estimates indicate that more than 11 million geospatial sample data points (gold and uranium assay values and sedimentological attributes) have been collected from Witwatersrand gold reefs, of which ~35 % are partially digitized. With very few exceptions, the distribution and concentration of gold and uranium in palaeoplacers reflects the changes in the depositional environment and mineral source through space and time.

Determining the sedimentological characteristics of palaeoplacer gold deposits is fundamental to understanding the nature of metal distribution and grade variability. Metal grade and thickness of Witwatersrand reef horizons have traditionally been estimated using assay data and conglomerate thickness measurements, with sedimentological attributes being used in a qualitative manner to assist in the definition of areas of geological homogeneity known as geozones for mineral resource estimation.

A Microsoft Access ™based application has recently been developed at the University of the Witwatersrand that is being used to capture sedimentological parameters (e.g. conglomerate thickness, % conglomerate, % pyrite mineralisation, pebble assemblage and sorting of the Witwatersrand Supergroup reefs), collected by mine geologists and samplers. The development of this software has created the opportunity to interrogate sedimentological parameters in a quantitative manner, using machine learning algorithms, to build predictive metal grade models.

Unlike conventional kriging of gold and uranium assay grades into resource block models of different support sizes, the use of sedimentological attributes to predict gold grades through training of big data has the potential to improve gold grade estimates in areas that are sparsely sampled. By using a combination of sedimentological parameters, metal assay data and experimental semi-variograms of such, it is possible to get an estimation of grade, enhancing predictive exploration strategies at the mine to target development scales.

Authors: Glen T. Nwaila, Ergong S. Zhang, Leon Tolmay, Musa S.D. Manzi, Christina Dohm, Mark Burnett, Raymond J. Durrheim