Harnessing the Power of Artificial Intelligence and Machine Learning in Mineral Exploration—Opportunities and Cautionary Notes

Editor’s note: The aim of the Geology and Mining series is to introduce early-career professionals and students to various aspects of mineral exploration, development, and mining in order to share the experiences and insight of each author on the myriad of topics involved with the mineral industry and the ways in which geoscientists contribute to each. Abstract Artificial intelligence (AI), and machine learning (ML) have emerged in the last few years from relative obscurity in the mineral exploration sector and they now attract significant attention from people in both industry and academia. However, due to the novelty of AI and ML applications, their practical use and potential remain enigmatic to many beyond a relatively few expert practitioners. We introduce this subject for the nonexpert and review some of the current applications and evolving uses. For the most traditionally minded geologist, we argue that ML can be an invaluable new tool, contributing to topics that range from exploratory data analysis to automated core logging and mineral prospectivity mapping, such that it will have a substantial impact on how exploration is conducted in the future. However, ML algorithms perform best with a large amount of homogeneously distributed clean data for a well-constrained objective. For this reason, the application to exploration strategy, especially for optimizing target selection, will be a challenge where data are heterogeneous, multiscale, amorphous, and discontinuous. For the more tech-savvy geologist and data scientist, we provide notes of caution regarding the limitations of ML applied to geoscience data, and reasons to temper expectations. Nonetheless, we project that such technologies, if used in an appropriate manner, will eventually be part of the full range of exploration tasks, allowing explorers to do more with their data in less time. However, whether this will tip the scales in favor of higher discovery rates remains to be demonstrated.

Volcaniclastic deposits and sedimentation processes around volcanic ocean islands: the central Azores

Geological histories of volcanic ocean islands can be revealed by the sediments shed by them. Hence there is an interest in studying cores of volcaniclastic sediments that are particularly preserved in the many flat-floored basins lying close to the Azores islands. We analyse four gravity cores collected around the central group of the islands. Three sedimentary facies (F1-F2a, F2b) are recognized based on visual core logging, particle morphometric and geochemical analyses. F1 is clay-rich hemipelagite comprising homogeneous mud with mottled structures from bioturbation. F2a and F2b are both clay-poor volcaniclastic deposits, which are carbonate-rich and carbonate-poor, respectively. More biogenic carbonate in F2a reflects the incorporation of unconsolidated calcareous material from island shelves or bioturbation. Within F2a and F2b we identify deposits emplaced by pyroclastic fallout, primary or secondary turbidity currents by combining multiple information from lithological composition, sedimentary structures, chemical composition of volcanic glass shards and morphometric characteristics of volcanic particles. Primary volcaniclastic sediments were found in all four cores, echoing activity known to have occurred up to historical times on the adjacent islands. These preliminary results suggest that greater details of geological events could be inferred for other volcanic islands by adopting a similar approach to core analysis.Supplementary material at https://doi.org/10.6084/m9.figshare.c.5602176

Differentiated Archean Dolerites: Igneous and Emplacement Processes that Enhance Prospectivity for Orogenic Gold

Magnetite-bearing granophyre and quartz dolerite are the evolved fractions of differentiated dolerite (diabase) sills and are an important host to Archean gold deposits because they are chemical traps for orogenic fluids. Despite their economic importance, there is a poor understanding of how melt composition, crystal fractionation, sill geometry, and depth of emplacement increase the volume of host rock that is most favorable for gold precipitation during orogenesis. We use drill core logging, whole-rock geochemistry, magnetic susceptibility, gold assay, and thermodynamic modeling data from 11 mineralized and unmineralized ca. 2.7 Ga differentiated dolerites in the Eastern Goldfields superterrane (Yilgarn craton, Western Australia) to better understand the influence of igneous and emplacement processes on gold prospectivity. Orogenic gold favors differentiated dolerites, derived from iron-rich parental magmas, that crystallize large volumes of magnetite-bearing quartz dolerite (>25% total thickness). Mineralized sills are commonly >150 m thick and hosted by thick and broadly coeval sedimentary sequences. Sill thickness is an important predictor for gold prospectivity, as it largely controls cooling rate and hence fractionation. The parental melts of gold mineralized sills fractionated large amounts of clinopyroxene and plagioclase (possibly up to 50%) at depth before emplacement in the shallow crust. A second fractionation event at shallow levels (<3 km) operated both vertically and laterally, resulting in an antithetic relationship between quartz (magnetite) dolerite and cumulates (pyroxenites and peridotites). By comparison with younger mafic sills emplaced in synsedimentary basins, we argue that the geometry of these high-level sills was more irregular than the often-assumed tabular form. Any irregularities in the lower sill margin act as traps for early formed (dense) ferromagnesian minerals, now represented by pyroxene and peridotite cumulates. In contrast, irregularities in the upper sill margin trap the buoyant fractionated liquids when the sill is more crystalline, through magma flow on the scale of <1 km. Sills derived from iron-poor melts are rarely mineralized and, all else being equal, probably have to be thicker than Fe-rich sills to be similarly prospective for orogenic gold. Finally, we provide a list of quantifiable parameters that can be incorporated into an exploration program targeting differentiated dolerites that host orogenic gold.

METHODOLOGICAL APPROACHES TO PREDICTING THE MINERAL-COMPONENT COMPOSITION OF THE BAZHENOV FORMATION BASED ON «CORE-LOGGING» INTERCONNECTIONS

The article presents a methodology developed by the authors for calculating the lithological composition of the Bazhenov Formation in Western Siberia. It is based on the identified “core-logging” interconnections between the mineral-component composition of rocks and the physical properties of the section. The convergence of experimental data and calculated values is shown. The proposed technique was tested. The conditions of its applicability have been substantiated.

Characterization of altered mafic and ultramafic rocks using portable XRF geochemistry and portable Vis-NIR spectrometry

The accurate characterization of mafic and ultramafic rocks is a challenging but necessary task given the spatial and genetic relationship of mineralization with specific lithologies (e.g. komatiite hosted nickel-sulfides preferentially associated with cumulate-rich ultramafic rocks). Rock classification is further complicated as most mafic and ultramafic rocks have undergone varying degrees of alteration. The accuracy and reproducibility of characterization can be significantly improved by using portable energy dispersive X-ray Fluorescence (pXRF) chemical data with portable Visible and Near-Infrared (pVis-NIR) mineralogical data.A new workflow using pXRF and pVis-NIR is presented and used to reliably characterize mafic and ultramafic rocks from the Yilgarn Craton, Western Australia. The workflow involves 6 steps: Mitigate and identify compound processing and closure issues. For example, we used a pXRF with helium flush to reliably and rapidly measure light elements and mitigate closure, i.e. problems related to data failing to sum to 100%.Identify and exclude geochemically heterogeneous samples. Heterogeneity may be unrelated to alteration and caused by veining or small-scale structure interleaving of different rock types. Geochemical heterogeneity was evaluated using skewness and kurtosis of SiO2 data.Relate rocks from similar magmatic, weathering and alteration events. This was achieved by interpreting data grouping of Vis-NIR ferric and ferrous iron data via a 852 nm/982 nm reflectance v. 651 nm/982 nm reflectance plot and the Ferrous Abundance Index. Unrepresentative data were omitted.Correct XRF iron data, and characterize lithology and alteration. Values ascribed to regions in the TAS (Total Alkali Silica) diagram were used to approximate FeO and Fe2O3. Subsequently, geochemical indices (e.g. Mg#) were used to characterize the alteration box plot.Characterize fractionation in detail. Fractionation variation diagrams were used to interpret fractionation, e.g. MgO v. Al2O3, Ca/Al v. Al2O3, Ni/Cr v. Ni/Ti, and MgO v. Cr.Identify and quantify talc alteration and serpentinization. This included the use of a new alteration plot (Mg# v. 1410 nmRAD/Albedo) to estimate serpentinization and identify relationships between serpentine, carbonate, chlorite and talc abundances. The results and observations contained in this contribution have important implications for progressive technologies such as core logging platforms that are equipped with pXRF and pVis-NIR instruments.

A Comparative Study of Porphyry-Type Copper Deposit Mineralogies by Portable X-ray Fluorescence and Optical Petrography

Porphyry-type deposits are crucial reserves of Cu and Mo. They are associated with large haloes of hydrothermal alteration that host particular mineral assemblages. Portable X-ray fluorescence analysis (pXRF) is an increasingly common tool used by mineral prospectors to make judgments in the field during mapping or core logging. A total of 31 samples from 13 porphyry copper deposits of the Western Cordillera were examined. Whole-rock composition was estimated over three points of analysis by pXRF. This approach attempts to capture the rapid and sometimes haphazard application of pXRF in mineral exploration. Modes determined by optical petrography were converted into bulk rock compositions and compared with those determined by pXRF. The elements S, Si, Ca, and K all were underestimated by optical mineralogy, and the elements Cu, Mo, Al, Fe, Mg, and Ti were overestimated by optical mineralogy when compared with pXRF results. Most of these porphyry samples occur in veined porphyritic quartz monzonite that is characteristic of these deposits. Sulfide and silicate vein stockworks are pervasive in most of the samples as well as dissemination of sulfides outwards from veinlets. Ore minerals present include chalcopyrite and molybdenite with lesser bornite. Chalcocite, digenite, and covellite are secondary. Potential sources of analytical bias are discussed.

Possible ore-controlling structures for the Woxi Au/Sb/W deposit in Tanghuping Mining Section, Hunan Province, South China

The Woxi Mining District in Hunan Province, South China, hosts numerous high tonnage gold, stibnite, and tungsten strata-bound deposits. Our research focuses on the geological characteristics of the Tanghuping and Xintianwan faults in the Tanghuping Mining Section, within the Woxi Mining District. The Woxi Mining District exposes mainly slate, with interbedded quartz sandstone and clastic rocks, cut by three faults: the Tanghuping, Xintianwan, and Woxi faults. We test the hypothesis that the mineralization in this district is structurally controlled, related to motion along these faults. While the presence of Tanghuping and Xintianwan faults in the Tanghuping Mining Section has not been confirmed, the proposed faults are interpreted to be intimately related to mineralization. The results of previous field works have corroborated the location of the Tanghuping and Xintianwan faults and suggest that zones of high strain coincide with these faults. However, recent core logging data do not show evidence of fracture zones at the expected depth. It is therefore still unclear if the high strain is related to the presence of the faults or if it is the result of fold-related interlayer flexural slip. Microstructural analysis is used to further characterize strain-induced recrystallization textures and sense of shear near and within the interpreted fault zones. Understanding this relationship is important to assist exploration in detecting ore systems on a regional scale and identify the most profitable areas.