A Trace Element Classification Tree for Chalcopyrite from Oktyabrsk Deposit, Norilsk–Talnakh Ore District, Russia: LA-ICPMS Study

The Oktyabrsk PGE-Cu-Ni deposit is one of the largest resources in the Norilsk–Talnakh ore district, Russia, and it is viewed as an ore giant on a global scale. It contains three types of ores: massive, disseminated and veinlet-disseminated. The two former ore types were formed by a liquation process, whereas the latter was associated with fluid-induced selective metasomatic replacement of metamorphosed wall rocks. One of the major ore minerals in all ore types is chalcopyrite. In this study, we determined concentrations of trace elements in this mineral using laser ablation inductively coupled plasma mass spectrometry. It appeared that standard geochemical tools, such as plotting the data in the form of diagrams of normalized concentrations, binary and ternary plots, do not allow one to distinguish chalcopyrite from visually and genetically different ore types. In contrast, more advanced statistical methods such as cluster analysis show different groupings of elements for each ore type. Based on the element clustering, a classification tree was suggested, which allowed for the differentiation of massive, disseminated and veinlet-disseminated ore types of the Oktyabrsk deposit by Se, Te, Cd and Pb concentrations in chalcopyrite with a success rate of 86%. The general feature is that chalcopyrite of veinlet-disseminated ore is poorer in these elements compared to chalcopyrite of the two other ore types. Chalcopyrite of massive ore is poorer in Se and Te when compared to chalcopyrite of disseminated ore.

The disseminated type of ore is localized sometimes in contact gabbrodolerites and olivine-free gabbrodolerites (Figure 2d,e), but more often in picrites and taxtic gabbrodolerites (Figure 2f). Sulfides with a size of 0.1-2 mm are found in interstitial of silicates and as separate blobs up to 40-50 mm. Disseminated ores also occur in taxites where sulfide blobs are generally larger in size.
The disseminated type of ore is localized sometimes in contact gabbrodolerites and olivine-free gabbrodolerites (Figure 2d,e), but more often in picrites and taxtic gabbrodolerites (Figure 2f). Sulfides with a size of 0.1-2 mm are found in interstitial of silicates and as separate blobs up to 40-50 mm. Disseminated ores also occur in taxites where sulfide blobs are generally larger in size.
Veinlet-disseminated ores are found in the frontal part of the intrusion and above the intrusion within the contact metamorphic and metasomatic aureole of the Kharaelakh intrusion. Ores, which contain up to 30-50% of sulfides by volume, are located within hornfels and metasomatically altered Devonian sediments. By sulfide minerals, the veinlet-disseminated ores are (1) pyrrhotite, (2) chalcopyrite-pyrrhotite, (d-f)-disseminated ores (d-in contact gabbrodolerites, e-in olivine-free gabbrodolerites, f-in picrites); (g-j)-veinlet-disseminated ores (pyrrhotitechalcopyrite subtype) in carbonate skarns. In a-c: pyrrhotite in groundmass with the emulsion phase of chalcopyrite is indicative of coeval precipitation (massive texture) at the bottom part of the intrusion. In d-e: pyrrhotite-pentlandite-chalcopyrite drops are typical for the upper contact of the intrusion. In f: pyrrhotite-pentlandite-chalcopyrite drops of the intermediate part of the intrusion. In g-j: skarn-like rocks.

SEM Analyses
Major element analyses and back scattered electron (BSE) images were taken using a TescanVega 3 SBU scanning electron microscope (SEM) equipped with an energy dispersive spectrometer (EDS) of Oxford Instruments (Abingdon, UK) with an Aztec-based system of microanalysis. The operating conditions were in the high-vacuum mode (<9 × 10 −3 Pa) at an accelerating voltage of 20 kV with a high resolution and a distance of 15 mm.
SEM EDS analyses and BSE images were used for the two reasons: first, for selection and a visual control of minerals for the LA-ICPMS study and, second, for determination of iron content used as an internal standard in LA-ICPMS. An example of representative BSE images of chalcopyrite from different types of ores is shown in Figure 3.

SEM Analyses
Major element analyses and back scattered electron (BSE) images were taken using a TescanVega 3 SBU scanning electron microscope (SEM) equipped with an energy dispersive spectrometer (EDS) of Oxford Instruments (Abingdon, UK) with an Aztec-based system of microanalysis. The operating conditions were in the high-vacuum mode (<9 × 10 −3 Pa) at an accelerating voltage of 20 kV with a high resolution and a distance of 15 mm.
SEM EDS analyses and BSE images were used for the two reasons: first, for selection and a visual control of minerals for the LA-ICPMS study and, second, for determination of iron content used as an internal standard in LA-ICPMS. An example of representative BSE images of chalcopyrite from different types of ores is shown in Figure 3.

LA-ICPMS Analysis
The concentrations of 29 209 Bi in sulfides were determined by LA-ICPMS at the Laboratory of Mineral Analysis of Institute of Geology of Ore Deposits Mineralogy, Petrography, and Geochemistry, Russian Academy of Sciences. The LA-ICPMS system consists of a quadrupole mass spectrometer Thermo Xseries2 equipped with a New Wave 213 laser instrument. Selected isotopes do not contain a significant isobaric and molecular interference.
For external calibration, the standard reference materials (SRM) UQAC FeS-1 (University of Quebec in Chicoutimi, Canada), GSD-1G and STD GL-3, produced from natural sulfide powder and doped with trace elements, were used. MASS-1 polymetal sulfide (USGS) was used for verification of the results.
LA-ICPMS analyses were conducted using an 80 μm beam diameter for spots and 40-65 μm for lines, a laser frequency of 15 Hz, 5-7 J/cm 2 energy density and a 7 μm/s ablation speed for lines. The carrier gas consists of a mixture of helium (0.7 L/min) and argon (0.85 L/min). Acquisition time in a spot mode was 30 s for the backgrounds, 60 s for the mineral analysis and 30 s of washout. Each line profile preceded 30 s of background. The material was then analyzed using the ICPMS operating in a time-resolved mode using peak jumping and a dwell time of 10 ms/peak per element.
Signal quantification was carried out by Iolite software using iron content in each sulfide determined by the SEM-EDS as an internal standard. Minerals in the profiles through intergrowths were calculated separately, avoiding areas of the contacts. In the case of detection of microinclusions in the LA-ICPMS spectra, the corresponding part of a spectrum was cut out off-line. The detection

LA-ICPMS Analysis
The concentrations of 29 209 Bi in sulfides were determined by LA-ICPMS at the Laboratory of Mineral Analysis of Institute of Geology of Ore Deposits Mineralogy, Petrography, and Geochemistry, Russian Academy of Sciences. The LA-ICPMS system consists of a quadrupole mass spectrometer Thermo Xseries2 equipped with a New Wave 213 laser instrument. Selected isotopes do not contain a significant isobaric and molecular interference.
For external calibration, the standard reference materials (SRM) UQAC FeS-1 (University of Quebec in Chicoutimi, Canada), GSD-1G and STD GL-3, produced from natural sulfide powder and doped with trace elements, were used. MASS-1 polymetal sulfide (USGS) was used for verification of the results.
LA-ICPMS analyses were conducted using an 80 µm beam diameter for spots and 40-65 µm for lines, a laser frequency of 15 Hz, 5-7 J/cm 2 energy density and a 7 µm/s ablation speed for lines. The carrier gas consists of a mixture of helium (0.7 L/min) and argon (0.85 L/min). Acquisition time in a spot mode was 30 s for the backgrounds, 60 s for the mineral analysis and 30 s of washout. Each line profile preceded 30 s of background. The material was then analyzed using the ICPMS operating in a time-resolved mode using peak jumping and a dwell time of 10 ms/peak per element.
Signal quantification was carried out by Iolite software using iron content in each sulfide determined by the SEM-EDS as an internal standard. Minerals in the profiles through intergrowths were calculated separately, avoiding areas of the contacts. In the case of detection of microinclusions in the LA-ICPMS spectra, the corresponding part of a spectrum was cut out off-line. The detection limits for LA-ICPMS analyses were calculated as three sigma times the background counts for the gas blank. Silicon was monitored to exclude silicate inclusions. Figure 4 shows representative lines through sulfide minerals with the background of BSE images. It is seen that chalcopyrite (Ccp), pentlandite (Pn) and pyrrhotite (Po) are easily recognized by their corresponding element spectra. Furthermore, we focus on chalcopyrite data only, because this mineral occurs in all ore types in a significant amount.
Minerals 2020, 10, x FOR PEER REVIEW 6 of 16 limits for LA-ICPMS analyses were calculated as three sigma times the background counts for the gas blank. Silicon was monitored to exclude silicate inclusions. Figure 4 shows representative lines through sulfide minerals with the background of BSE images. It is seen that chalcopyrite (Ccp), pentlandite (Pn) and pyrrhotite (Po) are easily recognized by their corresponding element spectra. Furthermore, we focus on chalcopyrite data only, because this mineral occurs in all ore types in a significant amount.

3.3.R Package
We used free RStudio software as a medium for R language. The used libraries were: "ggplot2" for making the violin plots, "cluster" for cluster analysis, "pvclust" for construction of classification trees, and "psych" for constructing correlation tables and binary plots https://cran.r-project.org (R for Windows v.R-4.0.2).

Results
Concentrations of 14 trace elements were determined in chalcopyrite of veinlet-disseminated (46 analyses), disseminated (25 analyses) and massive (16 analyses) ores (Table S1). Chalcopyrite of massive ores (32 analyses) was previously analyzed by LA-ICPMS for trace elements by [16]. Our new data are in good agreement with previously published results. However, for further internal consistency, we used only our new data. For each element of every ore type, violin plots were constructed ( Figure 5, Figure S1). Element concentrations of chalcopyrite of different ore types fully overlap. This is an unexpected result, because massive and disseminated ores are related to the process of silicatesulfide immiscibility [20][21][22], whereas veinlet-disseminated ores are associated with the fluid-induced selective metasomatic replacement of metamorphosed wall rocks [23]. Thus, the source of metals of the two ore types is expected within the magma itself, whereas veinlet-disseminated ores should be affected by host rocks.

R Package
We used free RStudio software as a medium for R language. The used libraries were: "ggplot2" for making the violin plots, "cluster" for cluster analysis, "pvclust" for construction of classification trees, and "psych" for constructing correlation tables and binary plots https://cran.r-project.org (R for Windows v.R-4.0.2).

Results
Concentrations of 14 trace elements were determined in chalcopyrite of veinlet-disseminated (46 analyses), disseminated (25 analyses) and massive (16 analyses) ores (Table S1). Chalcopyrite of massive ores (32 analyses) was previously analyzed by LA-ICPMS for trace elements by [16]. Our new data are in good agreement with previously published results. However, for further internal consistency, we used only our new data. For each element of every ore type, violin plots were constructed ( Figure 5, Figure S1). Element concentrations of chalcopyrite of different ore types fully overlap. This is an unexpected result, because massive and disseminated ores are related to the process of silicate-sulfide immiscibility [20][21][22], whereas veinlet-disseminated ores are associated with the fluid-induced selective metasomatic replacement of metamorphosed wall rocks [23]. Thus, the source of metals of the two ore types is expected within the magma itself, whereas veinlet-disseminated ores should be affected by host rocks.  Figure S1. Colors are the same as in Figure 1b for the ore types.
Different elements may show coherent behavior depending on the process of crystallization of chalcopyrite. To test this, we applied cluster analysis, whose outcome is shown in Figure 6. Chalcopyrite of veinlet-disseminated, disseminated and massive ores is characterized by four, two and three clusters of elements with high similarity, respectively. Some element clustering can be easily explained by co-crystallization of particular mineral phases, though some spurious clustering could also occur ( Table 1). For example, it is difficult to explain Ti-Sn, Ti-Mn-Mo and Mn-Ti-Zn-Ag clusters for chalcopyrite of veinlet-disseminated, disseminated and massive ores, respectively. Since, cluster analysis is merely a statistical tool, it inevitably produces some clusters that may have no geological meaning. Other clusters, such as Bi-Se-Te, Zn-Cd and Co-Ni in chalcopyrite of veinlet-disseminated ores, Ag-Co-Ni-Sn-Bi-Tl-Pb-Zn-Cd-Se-Te in chalcopyrite of disseminated ores and Co-Ni-Mo-Pb in chalcopyrite of massive ores, are explained by the presence of mineral phases observed in these ore types, which co-crystallized with chalcopyrite (Table 1). Se and Te form clusters in all three ore types because these two elements have similar geochemical behavior.  Figure S1. Colors are the same as in Figure 1b for the ore types.
Different elements may show coherent behavior depending on the process of crystallization of chalcopyrite. To test this, we applied cluster analysis, whose outcome is shown in Figure 6. Chalcopyrite of veinlet-disseminated, disseminated and massive ores is characterized by four, two and three clusters of elements with high similarity, respectively. Some element clustering can be easily explained by co-crystallization of particular mineral phases, though some spurious clustering could also occur (Table 1). For example, it is difficult to explain Ti-Sn, Ti-Mn-Mo and Mn-Ti-Zn-Ag clusters for chalcopyrite of veinlet-disseminated, disseminated and massive ores, respectively. Since, cluster analysis is merely a statistical tool, it inevitably produces some clusters that may have no geological meaning. Other clusters, such as Bi-Se-Te, Zn-Cd and Co-Ni in chalcopyrite of veinlet-disseminated ores, Ag-Co-Ni-Sn-Bi-Tl-Pb-Zn-Cd-Se-Te in chalcopyrite of disseminated ores and Co-Ni-Mo-Pb in chalcopyrite of massive ores, are explained by the presence of mineral phases observed in these ore types, which co-crystallized with chalcopyrite (Table 1). Se and Te form clusters in all three ore types because these two elements have similar geochemical behavior. Minerals 2020, 10, x FOR PEER REVIEW 8 of 16 For constructing a classification tree based on element clustering, we picked up several elements manually, which are the most common elements in all ore types. It appeared that using a limited number of elements, namely, Se, Te, Cd and Pb, allows satisfactory discrimination of the ore types with a success rate of 86%. The suggested classification tree is shown in Figure 7. For constructing a classification tree based on element clustering, we picked up several elements manually, which are the most common elements in all ore types. It appeared that using a limited number of elements, namely, Se, Te, Cd and Pb, allows satisfactory discrimination of the ore types with a success rate of 86%. The suggested classification tree is shown in Figure 7. Minerals 2020, 10, x FOR PEER REVIEW 9 of 16

Type of Ores Groups of Elements Interpretation
Veinlet-disseminated ore

Discussion
A standard approach for the interpretation of multielement data includes (a) plotting of element concentrations normalized to a reference composition (chondrite, primitive mantle, mid-oceanic ridge basalt (MORB), etc.) (e.g., [1,50]) and (b) using binary and ternary plots (e.g., [51]). Such diagrams are used to visualize differences and similarities between different datasets and to draw genetic conclusions. However, in the case of our study, none of these diagrams show differences between chalcopyrite of the three different ore types of the Oktyabrsk deposit. As an example, we show a MORB sulfide-normalized diagram ( Figure 8) and a set of binary plots with calculated Pearson correlation coefficients (Figure 9). It was already obvious from the violin plots ( Figure 5, Figure S1) that such standard plots may not work.

Type of Ores Groups of Elements Interpretation
Veinlet-disseminated ore

Discussion
A standard approach for the interpretation of multielement data includes (a) plotting of element concentrations normalized to a reference composition (chondrite, primitive mantle, mid-oceanic ridge basalt (MORB), etc.) (e.g., [1,50]) and (b) using binary and ternary plots (e.g., [51]). Such diagrams are used to visualize differences and similarities between different datasets and to draw genetic conclusions. However, in the case of our study, none of these diagrams show differences between chalcopyrite of the three different ore types of the Oktyabrsk deposit. As an example, we show a MORB sulfide-normalized diagram ( Figure 8) and a set of binary plots with calculated Pearson correlation coefficients (Figure 9). It was already obvious from the violin plots ( Figure 5, Figure S1) that such standard plots may not work.  [52].

Figure 9.
A fragment of Pearson correlation matrix and corresponding binary plots for chalcopyrite of the three different ore types of the Oktyabrsk deposit. The full correlation matrix is not provided due to its size. However, other elements show the same feature that the three different ore types cannot be distinguished from each other by using binary plots, though some elements show stronger correlations for chalcopyrite of disseminated ore (e.g., Zn-Cd, Zn-Pb) when compared to chalcopyrite of the other ore types. Colors in the caption (red, green, yellow) correspond to symbols of the same colors.
Previously, George et al. [1] suggested a genetic classification of chalcopyrite based on Zn and Cd concentrations (Figure 10). A stronger correlation coefficient for chalcopyrite of disseminated ore compared to the other two ore types gives hope to the idea that a Cd vs. Zn diagram could distinguish the disseminated ore type at least. However, all three ore types overlap on this diagram. Our data spread over most of the fields with the majority of analyses located in the field of high crystallization temperature (Figure 10).  [52].

Figure 9.
A fragment of Pearson correlation matrix and corresponding binary plots for chalcopyrite of the three different ore types of the Oktyabrsk deposit. The full correlation matrix is not provided due to its size. However, other elements show the same feature that the three different ore types cannot be distinguished from each other by using binary plots, though some elements show stronger correlations for chalcopyrite of disseminated ore (e.g., Zn-Cd, Zn-Pb) when compared to chalcopyrite of the other ore types. Colors in the caption (red, green, yellow) correspond to symbols of the same colors.
Previously, George et al. [1] suggested a genetic classification of chalcopyrite based on Zn and Cd concentrations (Figure 10). A stronger correlation coefficient for chalcopyrite of disseminated ore compared to the other two ore types gives hope to the idea that a Cd vs. Zn diagram could distinguish the disseminated ore type at least. However, all three ore types overlap on this diagram. Our data spread over most of the fields with the majority of analyses located in the field of high crystallization temperature (Figure 10). Figure 9. A fragment of Pearson correlation matrix and corresponding binary plots for chalcopyrite of the three different ore types of the Oktyabrsk deposit. The full correlation matrix is not provided due to its size. However, other elements show the same feature that the three different ore types cannot be distinguished from each other by using binary plots, though some elements show stronger correlations for chalcopyrite of disseminated ore (e.g., Zn-Cd, Zn-Pb) when compared to chalcopyrite of the other ore types. Colors in the caption (red, green, yellow) correspond to symbols of the same colors.
Previously, George et al. [1] suggested a genetic classification of chalcopyrite based on Zn and Cd concentrations (Figure 10). A stronger correlation coefficient for chalcopyrite of disseminated ore compared to the other two ore types gives hope to the idea that a Cd vs. Zn diagram could distinguish the disseminated ore type at least. However, all three ore types overlap on this diagram. Our data spread over most of the fields with the majority of analyses located in the field of high crystallization temperature ( Figure 10).
We also tried using the Se-Ni-Cd ternary genetic classification diagram of Duran et al. [13], suggested for discrimination between magmatic and hydrothermal chalcopyrite ( Figure 11). In this diagram, half of analyses of chalcopyrite from the Oktyabrsk deposit are classified as magmatic and the other half as hydrothermal, irrespective of the ore type. Similarly, half of the analyses of chalcopyrite from massive ore of the Oktyabrsk deposit [16] fall into the hydrothermal type, whereas they should be within the magmatic field ( Figure 11).
Using our classification tree (Figure 7), 90% of analyses of chalcopyrite from the Mansur et al. [16] database are correctly classified as belonging to the massive type of ore. This shows that the classification suggested in this study works, at least for the Oktyabrsk deposit, and is not dependent on the analytical set-up used for the analysis of trace element concentrations in chalcopyrite. However, our classification tree for the dataset of chalcopyrite analyses from Duran et al. [13] does not work. This shows that classifications such as those previously proposed by Duran et al. [13] or suggested in this study are dependent on the type of particular deposits. Our classification tree will likely work for the so-called Norilsk type of deposits, but may not be applicable for other types of deposits. The problem of using chalcopyrite for genetic classification is in the limited amount of empirical data. Our study also shows that interpretation of complex trace element data in chalcopyrite should be conducted using advanced statistical methods.
[16] database are correctly classified as belonging to the massive type of ore. This shows that the classification suggested in this study works, at least for the Oktyabrsk deposit, and is not dependent on the analytical set-up used for the analysis of trace element concentrations in chalcopyrite. However, our classification tree for the dataset of chalcopyrite analyses from Duran et al. [13] does not work. This shows that classifications such as those previously proposed by Duran et al. [13] or suggested in this study are dependent on the type of particular deposits. Our classification tree will likely work for the so-called Norilsk type of deposits, but may not be applicable for other types of deposits. The problem of using chalcopyrite for genetic classification is in the limited amount of empirical data. Our study also shows that interpretation of complex trace element data in chalcopyrite should be conducted using advanced statistical methods.
Based on the suggested classification tree and known mineral associations in the different ore types, some ideas regarding the behavior of elements can be suggested. For example, chalcopyrite of disseminated ore is richer on average in Cd and Pb than chalcopyrite of the other two ore types. This is probably because the initial sulfide melt was rich in these elements and chalcopyrite crystallized before galena. Accumulation of sulfide drops to form massive ore, whose chalcopyrite is poorer on average in Cd and Pb when compared to chalcopyrite of disseminated ore, occurred after the beginning of galena crystallization. It was shown experimentally that cooling of sulfide melt from 700 o C to 400 o C leads to increase in Cu/Pb ratio in melt by a factor of 3, suggesting galena crystallization at higher temperatures than chalcopyrite [53]. As for the distribution of Se and Te in chalcopyrite, it is considered that Se is homogeneously distributed between monosulfide solid solution (MSS) and intermediate solid solution (ISS), whereas Te is favorably distributed to ISS [52]. Due to this, there is no difference between chalcopyrite of disseminated and massive ores in terms of these elements because these ores are the product of the evolution of MSS. Chalcopyrite of veinlet-disseminated ores is poorer on average in Te than chalcopyrite of the other two types. This may show that veinlet-disseminated ore is not a product of the evolution of MSS and is related to another process, for example, metamorphism, as suggested by the authors of [49].  Based on the suggested classification tree and known mineral associations in the different ore types, some ideas regarding the behavior of elements can be suggested. For example, chalcopyrite of disseminated ore is richer on average in Cd and Pb than chalcopyrite of the other two ore types. This is probably because the initial sulfide melt was rich in these elements and chalcopyrite crystallized before galena. Accumulation of sulfide drops to form massive ore, whose chalcopyrite is poorer on average in Cd and Pb when compared to chalcopyrite of disseminated ore, occurred after the beginning of galena crystallization. It was shown experimentally that cooling of sulfide melt from 700 • C to 400 • C leads to increase in Cu/Pb ratio in melt by a factor of 3, suggesting galena crystallization at higher temperatures than chalcopyrite [53]. As for the distribution of Se and Te in chalcopyrite, it is considered that Se is homogeneously distributed between monosulfide solid solution (MSS) and intermediate solid solution (ISS), whereas Te is favorably distributed to ISS [52]. Due to this, there is no difference between chalcopyrite of disseminated and massive ores in terms of these elements because these ores are the product of the evolution of MSS. Chalcopyrite of veinlet-disseminated ores is poorer on average in Te than chalcopyrite of the other two types. This may show that veinlet-disseminated ore is not a product of the evolution of MSS and is related to another process, for example, metamorphism, as suggested by the authors of [49].
Minerals 2020, 10, x FOR PEER REVIEW 12 of 16 Figure 11. Ternary classification diagram based on concentrations of Cd, Se and Ni, according to the authors of [13]. The major idea of this classification is that the apexes trending to Ni and Cd are characteristics of magmatic and hydrothermal deposits, respectively. The apex trending towards Se was not given a genetic explanation by Duran et al. [13]. The thin dotted line is for the composition of chalcopyrite from massive ore of the Oktyabrsk deposit according to [16]. The bold dotted line discriminates hydrothermal and magmatic chalcopyrite according to [13]. Different symbols (circle, triangle and rhomb) are used for different ore types, respectively, disseminated, massive and veinlet-disseminated. (a) disseminated ore; (b) massive ore (c) veinlet-disseminated ore.

Conclusions
Trace element concentrations in chalcopyrite of massive, disseminated and veinlet-disseminated ores of the Oktyabrsk deposit of the Norilsk-Talnakh ore district are characterized by a large range of values. Conventional methods of representation of the trace element data such as plotting normalized concentrations or constructing binary and ternary diagrams do not allow for the separation of different ore types by trace element data in chalcopyrite. Thus, we suggested using more advanced statistics based on cluster analysis and a classification tree to distinguish chalcopyrite of different types of ores. Using the suggested classification tree allows separation of chalcopyrite from three types of ores with a success rate of 86%. A test on an independently published trace element dataset for chalcopyrite from massive ore of the same deposit shows a 90% of success rate for identifying the correct ore type. Overall, our study shows that (1) chalcopyrite is potentially useful for genetic classification and (2) advanced statistical methods are superior to using simple diagrams, especially when complex analytical data are interpreted. Figure 11. Ternary classification diagram based on concentrations of Cd, Se and Ni, according to the authors of [13]. The major idea of this classification is that the apexes trending to Ni and Cd are characteristics of magmatic and hydrothermal deposits, respectively. The apex trending towards Se was not given a genetic explanation by Duran et al. [13]. The thin dotted line is for the composition of chalcopyrite from massive ore of the Oktyabrsk deposit according to [16]. The bold dotted line discriminates hydrothermal and magmatic chalcopyrite according to [13]. Different symbols (circle, triangle and rhomb) are used for different ore types, respectively, disseminated, massive and veinlet-disseminated. (a) disseminated ore; (b) massive ore (c) veinlet-disseminated ore.

Conclusions
Trace element concentrations in chalcopyrite of massive, disseminated and veinlet-disseminated ores of the Oktyabrsk deposit of the Norilsk-Talnakh ore district are characterized by a large range of values. Conventional methods of representation of the trace element data such as plotting normalized concentrations or constructing binary and ternary diagrams do not allow for the separation of different ore types by trace element data in chalcopyrite. Thus, we suggested using more advanced statistics based on cluster analysis and a classification tree to distinguish chalcopyrite of different types of ores. Using the suggested classification tree allows separation of chalcopyrite from three types of ores with a success rate of 86%. A test on an independently published trace element dataset for chalcopyrite from massive ore of the same deposit shows a 90% of success rate for identifying the correct ore type. Overall, our study shows that (1) chalcopyrite is potentially useful for genetic classification and (2) advanced statistical methods are superior to using simple diagrams, especially when complex analytical data are interpreted.