Machine learning approaches for responsible ore prospectivity modelling
Researcher
Fereshteh Khammar, Doctoral candidate, University of Helsinki
Supervisors
Prof. Christoph Beier, Department of Geosciences and Geography, University of Helsinki, and Research Prof. Vesa Nykänen at Geological Survey of Finland (GTK) Adjunct Professor at University of Helsinki
Description of the project:
An increased demand for base metals, dependency on high-tech elements, and diversity in application of critical raw materials have motivated researchers to put priority on exploration of potential deposits. However, exploration of hidden and complex ore deposits without outcrops is a challenge that geologists are facing especially in Finland. Therefore, more efforts are rquired to apply new techniques to explore new prospects in areas with thick/ a vast vegetation and/or snow cover and/or to improve the circumstantial evidence of deposits, to reduce the cost of the exploration process and reduce the negative impacts of mining. The Exploration Information System (EIS) project, funded by European Onion and lead by the GTK, to which the present PhD study contributes, aims to develop new data analysis methods using artificial intelligence (AL), machine learning (ML), deep learning into mineral prospectivity mapping (MPM) according to mineral systems modelling. MPM from regional to local scales has been developed to integrate various geoscience data to identify exploration target areas. In a mineral prospectivity model, a set of different data including geology, geochemistry, geophysics, etc. is considered as an input and mineralization potential areas as a desirable output from an integration function.
Aims of the project:
The main objective of this research is to map favorable exploration target areas/mineral targeting maps using the mineral potential tool (MPM) with the application of machine learning (ML) algorithms and deep learning for three main target areas/IOCG style deposits: Hannukainen, Kuervitikko, and Cu-Rautuvaara in the Kolari region, Central Lapland Greenston Belt, north of Finland. To achieve the objective, the study has been divided in a geological and a mathematical framework. The geological framework focusses on ore-forming process and critical geological processes based on mineral system analysis. At the first stage, I will concentrate on defining a set of targeting criteria/mappable criteria (proxies) according to their spatial or genetic associations with the targeted mineral deposits. e.g., element (Fe-Au-Cu ± As, Bi, Co, K, Li, LREE, Mo, Se, Te, U), mineral (magnetite-chalcopyrite-pyrite-gold) association, and host rock type (e.g., skarn, monzonite) in Fennoscandian IOCG deposits. Key is to take into account geochronological, and geochemical data as indicators for the ore-formation processes to evaluate the main geological criteria as initial input. Moreover, artificial intelligence (AI) and machine learning (ML) methods have been recently used to enhance statistical methods applied in the Earth sciences. ML methods can potentially generate models of complex and nonlinear systems such as multistage geological events. Therefore, we will apply diverse knowledge- and data-based approaches and compared the results based on geomodels, representing how reliable the outcomes are as the second stage of the project. The reliable output would be able to demonstrate promising areas where they have obvious similarities with well-known IOCG-style deposits in the mentioned region.
