Next Article in Journal
Hydrothermal Karstification of the Pre-Messinian Eonile Canyon: Geomorphological and Geochemical Evidences for Hypogene Speleogenesis in the Middle Nile Valley of Egypt
Next Article in Special Issue
Genesis of Xinjiazui Gold Deposit: In Situ Geochemical Constraints from Arsenopyrite
Previous Article in Journal
Study on the Effects of Microwave Heating Time and Power on the Mechanical Properties of Cemented Tailings Backfill
Previous Article in Special Issue
Hydrothermal Alteration Processes of Xincheng Gold Deposit Jiaodong Peninsula, China: Constraints from Composition of Hydrothermal Rutile
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Minerals
Volume 14
Issue 9
10.3390/min14090945
minerals-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Application of Machine Learning to Characterize Metallogenic Potential Based on Trace Elements of Zircon: A Case Study of the Tethyan Domain

by
Jin Guo
1 and
Wen-Yan He
2,*
1
School of Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China
2
Frontiers Science Center for Deep-Time Digital Earth, State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences, Beijing 100083, China
*
Author to whom correspondence should be addressed.
Minerals 2024, 14(9), 945; https://doi.org/10.3390/min14090945
Submission received: 28 May 2024 / Revised: 11 September 2024 / Accepted: 12 September 2024 / Published: 16 September 2024
(This article belongs to the Special Issue The Formation and Evolution of Gold Deposits in China)
Browse Figures
Versions Notes

Abstract

:
Amidst the rapid advancement of artificial intelligence and information technology, the emergence of big data and machine learning provides a new research paradigm for mineral exploration. Focusing on the Tethyan metallogenic domain, this paper conducted a series of research works based on machine learning methods to explore the critical geochemical element signals that affect the metallogenic potential of porphyry deposits and reveal the metallogenic regularity. Binary classifiers based on random forest, XGBoost, and deep neural network are established to distinguish zircon fertility, and these machine learning methods achieve higher accuracy, exceeding 90%, compared with the traditional geochemical methods. Based on the random forest and SHapley Additive exPlanations (SHAP) algorithms, key chemical element characteristics conducive to magmatic mineralization are revealed. In addition, a deposit classification model was constructed, and the t-SNE method was used to visualize the differences in zircon trace element characteristics between porphyry deposits of different mineralization types. The study highlights the promise of machine learning algorithms in metallogenic potential assessment and mineral exploration by comparing them with traditional chemical methods, providing insights into future mineral classification models utilizing sub-mineral geochemical data.

1. Introduction

Porphyry deposits provide approximately 75% of the world’s Cu, 50% of Mo, and 20% of Au, as well as significant Ag, Zn, Sn, Pb, and other metals [1,2]. These deposits originate from fluid exsolution processes linked to hydrous and oxidized magmas [3,4], which are distinguished by elevated Eu/Eu*, V/Sc, and Sr/Y ratios [5,6,7,8]. These geochemical signatures have become increasingly valuable in the exploration of porphyry deposits. However, weathering, hydrothermal alteration, and metamorphism can considerably modify these signatures, leading to unclear conclusions when they are employed as indicators of mineral prospectivity (e.g., [9,10]). As a stable and ubiquitous accessory mineral found in medium-acid rock bodies, zircon has become an effective indicator mineral for porphyry Cu deposits [11]. It captures the compositional traits and physicochemical parameters of the original magma, such as temperature, oxidation state, and water content [12,13,14,15,16]. Various indicators, including Eu/Eu* and Ce/Ce* ratios in zircons, are frequently employed to trace fertile magmas [11,17,18,19,20]. However, the binary classification diagrams of zircon trace elements used in different regions are not applicable to other areas, where the classification boundaries are unclear [10,20,21].
Machine learning algorithms offer distinctive advantages in managing vast datasets and high-dimensional feature data due to their superior interpretability and transparency [22,23,24]. These techniques hold considerable promise for porphyry mineral exploration as datasets continue to grow in size. Past studies have demonstrated the use of these methods across different types of data to forecast mineral potential (e.g., identifying geochemical anomalies related to Cu-Au mineralization and delineating ore-forming target zones [25,26]), distinguish alteration zones associated with porphyry Cu deposits [27], and establish metallogenic potential discrimination models [28,29]. However, previous studies have focused on the global scale or individual deposits, and region-specific studies are lacking. The ore-forming rock bodies developed under specific environmental conditions often have significant differences in geochemical characteristics [30]. Therefore, conducting relevant research tailored to specific regions is particularly crucial. In this study, we collected a total of 3751 zircon data from the Tethys domain. This paper’s principal contributions can be outlined as follows: Different machine learning algorithms were used to carry out machine learning training on the compiled zircon data. A classification model was established and visualized for the porphyry deposits of various mineralization types to explore the differences in trace element characteristics.

2. Geological Setting

The Tethyan metallogenic domain extends approximately 10,000 km from east to west. It has undergone the expansion, sedimentation, and closure uplift of the Neo-Tethys and Paleo-Tethys oceans, as well as two large-scale plate subduction collisions [31]. Dynamic geological processes within the Tethyan domain have led to the formation of world-renowned mineral belts [4,32]. Mineralization related to Paleo-Tethys Ocean subduction is primarily concentrated in the eastern Qinghai–Tibet Plateau. Under the backdrop of the westward subduction of the Ganzi–Litang terrane, the Yidun magmatic arc was formed. The Pulang super-large porphyry Cu-Au deposit is distributed in the eastern segment of the Yidun arc [33,34,35]. Mineralization related to the Paleo-Tethys Ocean occurred during the collision between the Qiangtang–Lhasa terrane in the northern part of the Bangong–Nujiang suture zone. The Duolong mineralization cluster in Tibet contains several large porphyry Cu-Au polymetallic deposits such as Bolong, Naruo, and Duobuza [36]. The Xiongcun porphyry Cu-Au deposit, located in the Gangdese magmatic arc, is considered to be related to the Neo-Tethys Ocean subduction mineralization. The Neo-Tethys Ocean closed later in Pakistan, and within the Chagai magmatic belt, there are Saindak and Reko Diq large porphyry Cu-Au deposits [32]. The collision and post-collisional stages of the Neo-Tethys Ocean witnessed extensive mineralization. Representative deposits during the main collision stage (65~41 Ma) include the Sharang porphyry Mo deposits [30]. Post-collisional mineralization refers to the deposits formed subsequent to the primary collision between the Indian and Eurasian continents, represented by the Gangdese belt, and it features a series of large porphyry Cu-Mo deposits, e.g., the Qulong, Jiru, Jiama, and Zhunuo deposits [37,38]. Additionally, the Sahand–Bazman belt in Iran and the Ailaoshan–Red River belt in China are large porphyry mineralization belts formed during the post-collisional stage of the Neo-Tethys. The Sahand–Bazman belt trends northwest–southeast, extending approximately 1500 km. Mineralization occurred from 16~9 Ma, and representative large-scale deposits include the Sar Cheshmeh and Sungun porphyry Cu-Mo deposits [39]. Representative deposits in the Ailaoshan–Red River belt include the Yulong super-large porphyry Cu deposit and the Beiya porphyry-skarn Au deposit [40,41,42]. The mineral belts and deposits mentioned above are shown in Figure 1 and Table 1. The physical map is sourced from https://ontheworldmap.com (accessed on 10 August 2024).

3. Methods

3.1. Data Resources and Preprocessing

Magmatic zircon data in this study were primarily collected from the literature related to porphyry deposits in the Tethyan domain, and the dataset also included some new zircon data collected by this study. Zircons containing cracks or inclusions were avoided by excluding data with concordances below 90%. The inherited zircons are removed, and the values below the minimum detection limit are screened out. The samples are from 29 porphyry deposits in six metallogenic belts in the Tethyan domain. The complete compilation of data and references is shown in Supplementary Table S1.
Data cleaning is required before data training and classification to avoid the impact of abnormal data on classification results. First, zircon data with La > 1 ppm and Ti > 50 ppm were deleted to exclude the pollution caused by apatite and Ti-Fe oxide [11]. An absence of values in certain instances could potentially result in false positive identifications [58]. In order to avoid affecting the classification effect, the missing values are deleted. Finally, 1203 fertile zircons and 1069 barren zircons were used to build machine learning models. The data underwent log transformation and Z-score standardization after data cleaning. In the preparation of the fertile zircon and the barren zircon classification model, Ti, Y, La, Ce, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, and Hf were selected as the classification indicators. The three subplots of Figure 2, respectively, show the original data distribution of zircon trace elements, the distribution after the log transformation, and the distribution after the Z-score standardization.

3.2. Machine Learning Algorithm

Random forest, proposed by Breiman in 2001, is an ensemble learning algorithm that combines the bagging ensemble learning theory and the random subspace method [59,60,61]. Its main principle is shown in Figure 3a: random forest uses the bootstrap resampling method to extract multiple sample data from the dataset and randomly selects some features of each bootstrap sample to establish a decision tree model [62]. These models learn, predict, and obtain results independently. Finally, the results of each decision tree are combined, and the final prediction is obtained by voting. The advantages of random forest are that it can process high-dimensional data with more features, has a high tolerance for noise and outliers, and is not prone to overfitting.
XGBoost (eXtreme Gradient Boosting), proposed by Tianqi Chen, is a more efficient system implementation of gradient boosting based on a gradient-boosting decision tree [63]. XGBoost improves prediction accuracy by constantly forming new decision trees to fit the residual of the prediction results, thereby narrowing the gap with the actual value. The objective function defined by the training XGBoost model includes the loss function and regular term. The loss function represents the degree to which the model fits the data, and the second-order Taylor expansion can better judge the gradient descent trend, resulting in higher accuracy and faster convergence. The regular term is used to control the complexity of the model [64]. The objective function and model formula of XGBoost can be expressed as Formulas (1) and (2).
Obj = + ∑ k = 1 K Ω f k
y i ^ = ∑ k = 1 K f k x i
where l y i , y i ^ represents the loss between the predicted value y i ^ and the true value y i of the i -th sample, and Ω f k represents the complexity of the k -th decision tree. f k ∈ F , where F = f x = w q x , each f k corresponds to an independent decision tree structure q and leaf weight w . w i represents the weight of the i -th node, and w q x is the scoring for sample x , i.e., the model prediction.
A deep neural network (DNN) consists of an input layer, hidden layers, and an output layer, which transmits data to the output layer through forward propagation [65,66]. The weights and bias parameters of each layer determine the transmission and conversion of data. In the process of backpropagation, the error is calculated by the loss function, and the error is backpropagated to each layer by the chain rule, and the loss function is reduced by updating the parameters. DNN uses the nonlinear expression of hidden layers and activation functions to simulate nonlinear relationships in complex data [67,68]. SHAP is a method based on game theory concepts that are often used to explain the results of DNN model predictions [69]. The principal diagram of the DNN can be seen in Figure 3b.

3.3. Model Performance Evaluation

Accuracy, precision, recall, and F1 score constitute four pivotal metrics employed in evaluating model performance. Formulas (3)–(6) encapsulate the calculation methodologies for the four metrics [70], with TP, TN, FP, and FN, respectively, representing the true positive, true negative, false positive, and false negative instances. TPR (true positive rate), also known as sensitivity or recall, represents the proportion of correctly predicted positive instances by the model. FPR (false positive rate) indicates the model’s error rate in predicting positives. These metrics are crucial for evaluating model performance, especially when dealing with imbalanced datasets in academic research standards.
a c c u r a c y = T P + T N T P + F P + T N + F N
p r e s i t i o n = T P F P + T P
r e c a l l = T P T P + F N
F 1   s c o r e = 2 ∗ p r e c i s o n ∗ r e c a l l p r e c i s o n + r e c a l l

3.4. t-SNE Visualization

The t-SNE (t-distributed Stochastic Neighbor Embedding), proposed by Laurens van der Maaten and Geoffrey Hinton, has become a popular tool for visualizing high-dimensional data by reducing it to two- or three-dimensional data [71]. Although traditional linear methods (e.g., PCA and LDA) have performed well in geochemical research, they are based on linear mapping transformations and may not effectively capture nonlinear relationships in complex data. The t-SNE method is based on the Gaussian distribution centered at each point, calculating the similarity of each logarithmic point in a high-dimensional space, creating a probability distribution, and then creating a similar probability distribution in a low-dimensional space. The algorithm minimizes the difference between two probability distributions using algorithms such as gradient descent [72].

4. Results and Discussion

4.1. Conventional Geochemical Indicators

Before applying machine learning methods for classification, we tried to utilize traditional chemical methods to discriminate zircon types. A spider diagram is a traditional geochemical analysis method used to distinguish the differences in element depletion and enrichment between different samples. Random samples were taken from six metallogenic zones and spider diagrams were drawn based on 16 elements, as shown in Figure 4, which illustrates the differences in zircon element content across different terranes and the variations between fertile and barren zircons. Compared with machine learning methods, the spider diagrams focus on the depletion and enrichment of trace elements (e.g., Hf, Ti, Ce, and Eu). On the whole, it can be seen that light rare earth elements (La, Ce, Nd, Sm, and Eu) in fertile zircons are relatively deficient, and heavy rare earth elements (Gd, Tb, Dy, Ho, Y, Er, Tm, Yb, and Lu) are relatively enriched. However, the barren zircons mostly share the same chemical characteristics. It is difficult to distinguish zircon only based on the depletion and enrichment state of trace elements. Machine learning makes full use of the 16 element indices to explore complex patterns and relationships, thus improving the accuracy of zircon classification.
Lu et al. and Pizarro et al. proposed methods based on zircon trace element indicators to assess the metallogenic potential of porphyry Cu deposits [11,20]. Their study identified the characteristic features of fertile magmatic zircons as Eu/Eu* > 0.3; 10,000*(Eu/Eu*)/Y > 1; (Ce/Nd)/Y > 0.01; Dy/Yb < 0.3; Ce/Ce* >100. We applied the samples associated with porphyry copper deposits from the compiled zircon dataset to these methods and plotted binary scatter plots, as shown in Figure 5, to explore the accuracy, FPR, TPR of the conventional methods. Figure 5 shows that the classification model based on traditional methods achieved an accuracy of 0.68~0.78 and a TPR of 0.52~0.91. The discrimination based on Eu/Eu* and (Ce/Nd)/Y achieved the highest accuracy, reaching 0.78. The discrimination method based on Dy/Yb and 10,000*(Eu/Eu*)/Y indicators, (Ce/Nd)/Y, and 10,000*(Eu/Eu*)/Y indicators achieved the highest TPR, reaching 0.91. The formation and evolution of the Tethys metallogenic domain are primarily controlled by multiple phases of the subduction and collision of the Tethys Ocean, with metallogenic belts forming at different times. Mineral deposits in this region mainly date back to the Triassic, Cretaceous, and Neogene periods [32]. To explore whether zircons from different periods affect our classification, zircon samples from the porphyry copper deposits of the Triassic, Cretaceous, and Neogene periods were randomly selected and analyzed using traditional plotting methods. Figure 6 shows that the classification remains unaffected by the age of the zircons.

4.2. Results from Machine Learning Models

To build machine learning models, we divided the preprocessed zircon trace element data into a training set, a validation set, and an independent test set according to the ratio of 8:1:1. To avoid significant discrepancies in sample sizes among the different mineralization belts, we use stratified sampling in Python’s scikit-learn. Before applying these machine learning algorithms, we utilized the RandomizedSearchCV class in Python’s scikit-learn library to perform a random search method for hyperparameter optimization [74]. The optimal parameters are then fine-tuned using 5-fold cross-validation, which provides a more stable and reliable evaluation of model performance through multiple rounds of training and testing. The parameters of the basic classifiers and their accuracy in the cross-validation set and test set are shown in Table 2. The values of the Area Under Curve (AUC) of the three models are all above 0.95, indicating that these classifiers have application values if they properly set thresholds [75]. Finally, the performance of the models for the classification of positive and negative samples was evaluated intuitively through the normalized confusion matrix, as shown in Figure 7. The confusion matrix shows that the TN rate of DNN is 0.913, which has a good performance in predicting barren samples. The TP rate of random forest is 0.975, which indicates that random forest has good performance in predicting fertile samples. With machine learning models, we were able to comprehensively consider the features of all 16 elements. In this way, the model can extract information from more comprehensive data, identifying subtle differences and underlying taxonomic rules. This method significantly improves the reliability and accuracy of the classification, and provides stronger support for the accurate identification of zircon.
Feature analysis is an important step after the generation of a classification model. It aids in comprehending the correlation between features and classification outcomes. It can be summarized from Figure 8 that as indicators can reflect magmatic oxygen fugacity; Ce and Eu play an important role in identifying zircons. Both are multivalent elements, and they can interconvert under certain redox conditions. The formation of porphyry copper deposits is more favorable with higher degrees of oxidation [76]. In hydrous melts, hornblende crystallizes early in magmatic evolution, reducing the Y value in the residual melt. The Y value lost in the residual melt and the Eu value not consumed by the early plagioclase crystallization result in high Eu outliers and low Y values in the fertile magmatic suites [77]. In the fertile magmatic suites, hornblende fractionation tends to preferentially incorporate the rare earth element Dy over the heavy rare earth element Yb. Therefore, in relatively fertile ore-forming magma systems, the Dy values are lower, and the Yb values are higher [11].
In previous studies, elements previously confirmed as indicators of fertile magmatic zircons have been reaffirmed through machine learning-based feature importance ranking, validating the scientific efficacy of our established model. In this study, the contents of Ti, Lu, Hf, and Ho are considered significantly related to distinguishing zircons. By creating scatter plots of characteristic elements, the distribution features of characteristic element contents in fertile zircons are observed, which is shown in Figure 9. It indicates that the Ti content in fertile magmatic zircon is mainly concentrated in 0–5 ppm, Eu content is concentrated in 0–2 ppm, and Ce content is mostly in the range of 10–60 ppm. Moreover, the Dy content of fertile magmatic zircons is concentrated in 10–100 ppm, Lu content in 20–100 ppm, Hf content in 8000–12,500 ppm, and Yb content in 200–500 ppm. However, there are currently no published conclusions by scholars regarding the roles and influences of these elements in mineralization mechanisms, necessitating further exploration and research in the future.

4.3. Application on Porphyry Deposits in Turkey

Machine learning classifiers were applied to the zircon samples from Turkey. The Turkish porphyry deposits are located in the middle of the Tethys and are divided into three main metallogenic belts from north to south: Pontides, Anatolides, and Border Folds [78,79,80,81,82,83,84]. Among them, the Tavşanli belt is characterized by a concentration of porphyry systems from the Eocene period, with its tectonic environment primarily associated with calc-alkaline magmatism resulting from plate collision and alkaline magmatism arising from post-collisional processes [85]. The zircon data from the rock mass hosting the Sarıçayırıyayla porphyry Cu-Mo deposit within the belt were utilized for the study. The Biga Peninsula mainly develops the porphyry metallogenic system from Eocene to Oligocene, and the formation of porphyry deposits is related to the calc-alkaline volcanic-intrusive rocks. The test samples encompass volcanic assemblages from the region hosting the Halilağa porphyry Cu-Mo deposit within the mineralization system, the contemporaneous volcanic assemblages from the mineralization epoch of the porphyry Au deposit at the Kışladağ deposit, and the mineralizing magmatic rock assemblages from the Pınarbaşı porphyry Mo-Cu deposit. The locations of the deposits are marked in Figure 1. These samples are sourced from different types of mineralized porphyry deposits, including those from fertile suites as well as from barren suites. In this study, the trace elements of zircon underwent preprocessing steps identical to those used during model establishment. These preprocessed data were then used as input for prediction on the classifiers, and the resulting outcomes are presented in Table 3.
The classification results of random forest and XGBoost are quite consistent with no significant deviation. All three algorithms, based on different principles, can accurately identify barren samples, e.g., streatted lava and mafic enclave. Based on the results of the random forest, we selected the higher-ranked elements Ce, Eu, Ti, and Dy—according to the feature importance ranking from the random forest classifier—as the horizontal and vertical coordinate axes to draw binary graphs. It can be seen from Figure 10 that the range of Ti and Eu values of fertile zircons is concentrated, while Ce and Dy values are scattered. Whether it is fertile zircon or barren zircon, the distribution of these important elements is not regular. It also indicates that although these elements have an important impact on random forest decisions, random forest classification is not based on individual elements for classification but learns from all features, making it more effective than the traditional binary plotting classification methods.

4.4. Classification of Porphyry Deposits of Different Metallogenic Types

As early as 1993, based on the geochemical data of 32 porphyry Cu-Mo mineralized rock masses in China, scholars have proposed a classification and quantitative discrimination model for the different mineralization types of porphyry Cu (Mo) deposits [86]. However, the traditional classification models relying solely on single-element approaches have shown lower accuracy in practical applications. Based on abundant magma zircon data from the study area, this paper employs machine learning methods to investigate whether there are systematic differences in the trace elements of zircons among the different mineralization types in porphyry deposits. The study categorizes deposits primarily into six types: Mo-rich, Au-rich, Cu, Cu-Mo, Cu-Au, and Cu-Mo-Au. It includes 648 zircon data points. To provide a clearer interpretation of the classification results, this study incorporates other element ratios of geological significance, such as Th/U, Lu/Hf, and Dy/Yb ratios. The Th/U ratio serves as an indicator of the extent of metamorphism within zircon [87]. The Lu/Hf isotope system is instrumental in establishing the timing and duration of geological events [88]. Furthermore, the Dy/Yb ratio is utilized to evaluate the mineralization potential of porphyry copper deposits [11].
We chose the random forest model to classify the types of deposits. Firstly, the training set, validation set, and test set were split in an 8:1:1 ratio. Due to significant variations in the number of zircons from the individual deposits in the dataset, we implemented selective sampling to balance the dataset during partitioning, ensuring the proportions of samples from the different deposit types are similar for training and testing. Recursive feature elimination (RFE) was employed for feature selection to prevent interference between feature elements. Finally, 15 indicators were selected for classifier establishment: Y, Ce, Sm, Eu, Gd, Tb, Ho, Er, Tm, Lu, Hf, U, Dy/Yb, Lu/Hf, and Th/U. The classification report is presented in Table 4, and the ranking of feature importance is shown in Figure 11.
Based on the t-SNE method, the chemical data of zircon from the different mineralization types of porphyry deposits were reduced and visualized, and the final results are shown in Figure 12. It can be seen that the distribution of element data gradually transitions from porphyry Cu to Au-rich to Mo-rich deposits. Porphyry Cu-Mo and Cu-Mo-Au deposits are widely distributed in the middle of the Cu and Mo types. Porphyry Cu-Au deposits are located near the boundary between the Cu and Au deposits.
Figure 11 indicates that Ce, Th/U, U, Eu, Lu, and Tm contribute significantly to classifying deposit types. By leveraging the t-SNE method, we generate scatter plots where color denotes the concentrations of elements. This visualization is depicted in Figure 13. By analyzing the visual data of element content, we can observe that the different mineral deposit types may have different characteristics in terms of trace element content. The Ce content in zircons from the porphyry Cu, Mo, and Cu-Au deposits is notably lower than that in the porphyry Au and Cu-Mo-Au deposits. Furthermore, the Th/U ratio in zircons from various deposit types exhibits trends: as one transitions from porphyry Cu deposits to Au-rich deposits to Mo-rich deposits, the Th/U ratio progressively diminishes, and the concentration of U correspondingly rises. The analysis of the Eu and Lu concentrations reveals that the porphyry Au-rich deposits have relatively high Eu content and lower Lu content. Compared to the top five elements, the concentration of Tm shows no significant variation across the different types of deposits. These findings may be associated with these deposits’ geological characteristics and formation mechanisms.

4.5. Discussion

Compared to the traditional chemical methods, machine learning methods demonstrate superior performance in establishing zircon classifiers and exploring different mineralization types in porphyry deposits. The traditional geochemical methods typically rely on laboratory analyses, which may be constrained by time, cost, and data volume. Moreover, relying on 2–3 elements for discrimination can lead to unclear classification boundaries. Machine learning methods, on the other hand, can handle large volumes of complex data, extracting latent patterns and features to automate classification and prediction in an efficient manner, and achieving higher accuracy and true positive rates.
However, in our research, we have also identified limitations in the application of machine learning. Some geological events or types of mineralization may be rarer or harder to obtain sufficient sample data for, resulting in imbalanced training datasets. This imbalance can affect the training and evaluation of machine learning models, potentially leading to inadequate prediction capabilities for minority classes. In addition, geological data often exhibit spatial and temporal correlations, where data from neighboring regions or different time periods may show interdependencies and variations. Traditional machine learning models may struggle to directly handle spatial and temporal structures in geological data, requiring additional preprocessing and feature engineering methods. Therefore, overcoming these limitations will be a crucial focus for geoscientists in future research endeavors.

5. Conclusions

This study proposes machine learning methods for zircon classification using trace element data. It examines the trace element characteristics of fertile zircons in porphyry deposits within the Tethys domain, leading to conclusions.
  • Machine learning models can effectively distinguish magmatic zircons from porphyry deposits in the Tethyan domain, and the accuracy of random forest, XGBoost, and DNN are 91%, 88.6%, and 90.8%, respectively.
  • Feature ranking results indicate that Ce, Eu, Dy, and Yb have significant impacts on distinguishing the fertility of zircons. In addition to the elements previously confirmed in prior studies, this research found that Ti, Lu, Hf, Ho, and Tb also play important roles in distinguishing zircons. The specific mechanisms of their effects require further exploration.
  • The t-SNE visualization provides a feasible method for exploring the differences in element characteristics among the porphyry deposits of the different mineralization types. The study found that in the Cu, Mo-rich, and mixed mineralization deposits, the Ce and Eu values are relatively lower compared to the Au-rich deposits, while the Lu values are higher. The Th/U ratios decrease sequentially in the Cu, Au-rich, and Mo-rich deposits.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/min14090945/s1, Table S1: Zircon trace element data. References [89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139] are cited in the supplementary materials.

Author Contributions

Conceptualization, W.-Y.H. and J.G.; methodology, W.-Y.H. and J.G.; formal analysis, J.G.; writing—original draft preparation, J.G.; writing—review and editing, W.-Y.H.; funding acquisition, W.-Y.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Key R&D Program of China (2022YFF0800902), the National Natural Science Foundation of China (42372098) and the Fundamental Research Funds for the Central Universities (2652023001).

Data Availability Statement

The data supporting this article are available in the online Supplementary Materials.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Kirkham, R.; Sinclair, W. Porphyry copper, gold, molybdenum, tungsten, tin, silver. In Geology of Canadian Mineral Deposit Types; Geological Society of America, Inc.: Boulder, CO, USA, 1995. [Google Scholar]
  2. Sinclair, W. Porphyry deposits. In Mineral Deposits of Canada: A Synthesis of Major Deposit-Types, District Metallogeny, the Evolution of Geological Provinces, and Exploration Methods: Geological Association of Canada, Mineral Deposits Division, Special Publication; Geological Association of Canada, Mineral Deposits Division: St. John’s, NL, Canada, 2007; Volume 5, pp. 223–243. [Google Scholar]
  3. Sillitoe, R.H. Porphyry copper systems. Econ. Geol. 2010, 105, 3–41. [Google Scholar] [CrossRef]
  4. Richards, J.P. Tectonic, magmatic, and metallogenic evolution of the Tethyan orogen: From subduction to collision. Ore Geol. Rev. 2015, 70, 323–345. [Google Scholar] [CrossRef]
  5. Baldwin, J.; Pearce, J.A. Discrimination of productive and nonproductive porphyritic intrusions in the Chilean Andes. Econ. Geol. 1982, 77, 664–674. [Google Scholar] [CrossRef]
  6. Lang, J.R.; Titley, S.R. Isotopic and geochemical characteristics of Laramide magmatic systems in Arizona and implications for the genesis of porphyry copper deposits. Econ. Geol. 1998, 93, 138–170. [Google Scholar] [CrossRef]
  7. Loucks, R. Distinctive composition of copper-ore-forming arcmagmas. Aust. J. Earth Sci. 2014, 61, 5–16. [Google Scholar] [CrossRef]
  8. Richards, J.P. Magmatic to hydrothermal metal fluxes in convergent and collided margins. Ore Geol. Rev. 2011, 40, 1–26. [Google Scholar] [CrossRef]
  9. Wang, R.; Tafti, R.; Hou, Z.-Q.; Shen, Z.-C.; Guo, N.; Evans, N.J.; Jeon, H.; Li, Q.-Y.; Li, W.-K. Across-arc geochemical variation in the Jurassic magmatic zone, Southern Tibet: Implication for continental arc-related porphyry CuAu mineralization. Chem. Geol. 2017, 451, 116–134. [Google Scholar] [CrossRef]
  10. Chen, X.; Richards, J.P.; Liang, H.; Zou, Y.; Zhang, J.; Huang, W.; Ren, L.; Wang, F. Contrasting arc magma fertilities in the Gangdese belt, Southern Tibet: Evidence from geochemical variations of Jurassic volcanic rocks. Lithos 2019, 324, 789–802. [Google Scholar] [CrossRef]
  11. Lu, Y.-J.; Loucks, R.R.; Fiorentini, M.; McCuaig, T.C.; Evans, N.J.; Yang, Z.-M.; Hou, Z.-Q.; Kirkland, C.L.; Parra-Avila, L.A.; Kobussen, A. Zircon compositions as a pathfinder for porphyry Cu±Mo±Au deposits. In Tectonics and Metallogeny of the Tethyan Orogenic Belt; Society of Economic Geologists: Littleton, CO, USA, 2016. [Google Scholar]
  12. Kemp, A.; Hawkesworth, C.J.; Foster, G.; Paterson, B.; Woodhead, J.; Hergt, J.; Gray, C.; Whitehouse, M. Magmatic and crustal differentiation history of granitic rocks from Hf-O isotopes in zircon. Science 2007, 315, 980–983. [Google Scholar] [CrossRef]
  13. Van Kranendonk, M.J.; Kirkland, C.L. Orogenic climax of Earth: The 1.2–1.1 Ga Grenvillian superevent. Geology 2013, 41, 735–738. [Google Scholar] [CrossRef]
  14. Lee, J.K.; Williams, I.S.; Ellis, D.J. Pb, U and Th diffusion in natural zircon. Nature 1997, 390, 159–162. [Google Scholar] [CrossRef]
  15. Rubatto, D. Zircon trace element geochemistry: Partitioning with garnet and the link between U–Pb ages and metamorphism. Chem. Geol. 2002, 184, 123–138. [Google Scholar] [CrossRef]
  16. Ballard, J.R.; Palin, J.M.; Campbell, I.H. Relative oxidation states of magmas inferred from Ce (IV)/Ce (III) in zircon: Application to porphyry copper deposits of northern Chile. Contrib. Mineral. Petrol. 2002, 144, 347–364. [Google Scholar] [CrossRef]
  17. Dilles, J.H.; Kent, A.J.; Wooden, J.L.; Tosdal, R.M.; Koleszar, A.; Lee, R.G.; Farmer, L.P. Zircon compositional evidence for sulfur-degassing from ore-forming arc magmas. Econ. Geol. 2015, 110, 241–251. [Google Scholar] [CrossRef]
  18. Lee, R.G.; Dilles, J.H.; Tosdal, R.M.; Wooden, J.L.; Mazdab, F.K. Magmatic evolution of granodiorite intrusions at the El Salvador porphyry copper deposit, Chile, based on trace element composition and U/Pb age of zircons. Econ. Geol. 2017, 112, 245–273. [Google Scholar] [CrossRef]
  19. Loader, M.A.; Wilkinson, J.J.; Armstrong, R.N. The effect of titanite crystallisation on Eu and Ce anomalies in zircon and its implications for the assessment of porphyry Cu deposit fertility. Earth Planet. Sci. Lett. 2017, 472, 107–119. [Google Scholar] [CrossRef]
  20. Pizarro, H.; Campos, E.; Bouzari, F.; Rousse, S.; Bissig, T.; Gregoire, M.; Riquelme, R. Porphyry indicator zircons (PIZs): Application to exploration of porphyry copper deposits. Ore Geol. Rev. 2020, 126, 103771. [Google Scholar] [CrossRef]
  21. Zhong, S.; Li, S.; Feng, C.; Liu, Y.; Santosh, M.; He, S.; Qu, H.; Liu, G.; Seltmann, R.; Lai, Z. Porphyry copper and skarn fertility of the northern Qinghai-Tibet Plateau collisional granitoids. Earth-Sci. Rev. 2021, 214, 103524. [Google Scholar] [CrossRef]
  22. Alpaydin, E. Introduction to Machine Learning; MIT Press: Cambridge, MA, USA, 2020. [Google Scholar]
  23. Hastie, T.; Tibshirani, R.; Friedman, J.; Hastie, T.; Tibshirani, R.; Friedman, J. Unsupervised learning. In The Elements of Statistical Learning: Data Mining, Inference, and Prediction; Springer: New York, NY, USA, 2009; pp. 485–585. [Google Scholar]
  24. Kubat, M. S; Springer: Berlin/Heidelberg, Germany, 2017. [Google Scholar]
  25. Keykhay-Hosseinpoor, M.; Kohsary, A.-H.; Hossein-Morshedy, A.; Porwal, A. A machine learning-based approach to exploration targeting of porphyry Cu-Au deposits in the Dehsalm district, eastern Iran. Ore Geol. Rev. 2020, 116, 103234. [Google Scholar] [CrossRef]
  26. Ford, A. Practical implementation of random forest-based mineral potential mapping for porphyry Cu–Au mineralization in the Eastern Lachlan Orogen, NSW, Australia. Nat. Resour. Res. 2020, 29, 267–283. [Google Scholar] [CrossRef]
  27. Abbaszadeh, M.; Hezarkhani, A.; Soltani-Mohammadi, S. An SVM-based machine learning method for the separation of alteration zones in Sungun porphyry copper deposit. Geochemistry 2013, 73, 545–554. [Google Scholar] [CrossRef]
  28. Zou, S.; Chen, X.; Brzozowski, M.J.; Leng, C.B.; Xu, D. Application of machine learning to characterizing magma fertility in porphyry Cu deposits. J. Geophys. Res. Solid Earth 2022, 127, e2022JB024584. [Google Scholar] [CrossRef]
  29. Nathwani, C.L.; Wilkinson, J.J.; Fry, G.; Armstrong, R.N.; Smith, D.J.; Ihlenfeld, C. Machine learning for geochemical exploration: Classifying metallogenic fertility in arc magmas and insights into porphyry copper deposit formation. Miner. Depos. 2022, 57, 1143–1166. [Google Scholar] [CrossRef]
  30. Hou, Z.Q.; Zheng, Y.C.; Yang, Z.M.; Yang, Z.S. Metallogesis of continental collision setting: Part I. Gangdese Cenozoic porphyry Cu-Mo systems in Tibet. Miner. Depos. 2012, 31, 647–670. [Google Scholar]
  31. Zhang, H.R.; Hou, Z.Q.; Song, Y.C.; Li, Z.; Yang, Z.M.; Wang, Z.L.; Wang, X.H.; Wang, S.X. The Temporal and spatial distribution of porphyry copper deposits in the eastern tethyan metallogenic domain: A review. Acta Geol. Sin. 2009, 83, 1818–1837. [Google Scholar]
  32. Wang, R.; Zhu, D.C.; Wang, Q.; Hou, Z.Q.; Yang, Z.M.; Zhao, Z.D.; Mo, X.X. Porphyry mineralization in the tethyan orogen. Sci. China Ser. D Earth Sci. 2020, 30, 38. [Google Scholar] [CrossRef]
  33. Chen, J.-L.; Xu, J.-F.; Ren, J.-B.; Huang, X.-X.; Wang, B.-D. Geochronology and geochemical characteristics of Late Triassic porphyritic rocks from the Zhongdian arc, eastern Tibet, and their tectonic and metallogenic implications. Gondwana Res. 2014, 26, 492–504. [Google Scholar] [CrossRef]
  34. Hou, Z. Tectono-magmatic evolution of the Yidunisland-arc and geodynamic setting of Kuroko-type sulfide deposits in Sanjiang Region, China. Resour. Geol. Spec. Issue 1993, 17, 336–350. [Google Scholar]
  35. Zeng, P.S.; Wang, H.P.; Mo, X.X.; Yu, X.H.; Li, W.C.; Li, T.G.; Li, H.; Yang, C.Z. Tectonic setting and prospects of orphyry copper deposits in Zhongdian island arc belt. Acta Geosci. Sin. 2004, 05, 535–540. [Google Scholar]
  36. Wei, S.-g.; Tang, J.-x.; Song, Y.; Liu, Z.-b.; Feng, J.; Li, Y.-b. Early Cretaceous bimodal volcanism in the Duolong Cu mining district, western Tibet: Record of slab breakoff that triggered ca. 108–113 Ma magmatism in the western Qiangtang terrane. J. Asian Earth Sci. 2017, 138, 588–607. [Google Scholar] [CrossRef]
  37. Hou, Z.; Yang, Z.; Qu, X.; Meng, X.; Li, Z.; Beaudoin, G.; Rui, Z.; Gao, Y.; Zaw, K. The Miocene Gangdese porphyry copper belt generated during post-collisional extension in the Tibetan Orogen. Ore Geol. Rev. 2009, 36, 25–51. [Google Scholar] [CrossRef]
  38. Yang, Z.; Cooke, D.R. Porphyry copper deposits in China. In Mineral Deposits of China; Society of Economic Geologists: Littleton, CO, USA, 2019. [Google Scholar]
  39. Sadigh, S.; Mirmohammadi, M.; Asghari, O.; Porwal, A. Spatial distribution of porphyry copper deposits in Kerman Belt, Iran. Ore Geol. Rev. 2023, 153, 105251. [Google Scholar] [CrossRef]
  40. He, W.-Y.; Mo, X.-X.; He, Z.-H.; White, N.C.; Chen, J.-B.; Yang, K.-H.; Wang, R.; Yu, X.-H.; Dong, G.-C.; Huang, X.-F. The geology and mineralogy of the Beiya skarn gold deposit in Yunnan, southwest China. Econ. Geol. 2015, 110, 1625–1641. [Google Scholar] [CrossRef]
  41. Hu, R.; Burnard, P.; Turner, G.; Bi, X. Helium and Argon isotope systematics in fluid inclusions of Machangqing copper deposit in west Yunnan province, China. Chem. Geol. 1998, 146, 55–63. [Google Scholar] [CrossRef]
  42. Liang, H.-Y.; Sun, W.; Su, W.-C.; Zartman, R.E. Porphyry copper-gold mineralization at Yulong, China, promoted by decreasing redox potential during magnetite alteration. Econ. Geol. 2009, 104, 587–596. [Google Scholar] [CrossRef]
  43. Zhang, H.R.; Hou, Z.Q.; Yang, Z.M. Metallogenesis and geodynamics of tethyan metallogenic domain: A review. Miner. Depos. 2010, 29, 113–133. [Google Scholar]
  44. Ren, H.; Zheng, Y.; Wu, S.; Wang, D.; Zuo, L.; Chen, L.; Gao, F.; Wei, J.; Wang, S.; Shu, D. Short-wavelength infrared characteristics and composition of white mica in the Demingding porphyry Cu-Mo deposit, Gangdese belt, Tibet: Implications for mineral exploration. Ore Geol. Rev. 2023, 164, 105833. [Google Scholar] [CrossRef]
  45. Sun, X.; Zheng, Y.; Xu, J.; Huang, L.; Guo, F.; Gao, S. Metallogenesis and ore controls of Cenozoic porphyry Mo deposits in the Gangdese belt of southern Tibet. Ore Geol. Rev. 2017, 81, 996–1014. [Google Scholar] [CrossRef]
  46. Chen, X.; Zheng, Y.; Gao, S.; Wu, S.; Jiang, X.; Jiang, J.; Cai, P.; Lin, C. Ages and petrogenesis of the late Triassic andesitic rocks at the Luerma porphyry Cu deposit, western Gangdese, and implications for regional metallogeny. Gondwana Res. 2020, 85, 103–123. [Google Scholar] [CrossRef]
  47. Xie, Z.; Xue, C.; Yang, T.; Xiang, K.; Xin, D. Petrogenesis and geodynamic implications of Early Cretaceous highly fractionated leucogranites in the northern Lanping–Simao terrane, Eastern Tibetan Plateau. J. Asian Earth Sci. 2020, 197, 104340. [Google Scholar] [CrossRef]
  48. Gao, T.; Tang, J.; Wang, L.; Li, B.; Wang, Y.; Lei, C. The first intra-oceanic island-arc porphyry Cu (Au) deposit in the western Bangong-Nujiang suture zone: Evidence from the Saidengnan Cu (Au) deposit. Gondwana Res. 2022, 104, 92–111. [Google Scholar] [CrossRef]
  49. Zengqian, H.; Hongwen, M.; Zaw, K.; Yuquan, Z.; Mingjie, W.; Zeng, W.; Guitang, P.; Renli, T. The Himalayan Yulong porphyry copper belt: Product of large-scale strike-slip faulting in eastern Tibet. Econ. Geol. 2003, 98, 125–145. [Google Scholar] [CrossRef]
  50. Zhou, H.; Sun, X.; Cook, N.J.; Lin, H.; Fu, Y.; Zhong, R.; Brugger, J. Nano-to micron-scale particulate gold hosted by magnetite: A product of gold scavenging by bismuth melts. Econ. Geol. 2017, 112, 993–1010. [Google Scholar] [CrossRef]
  51. Xu, L.; Bi, X.; Hu, R.; Zhang, X.; Su, W.; Qu, W.; Hu, Z.; Tang, Y. Relationships between porphyry Cu–Mo mineralization in the Jinshajiang–Red River metallogenic belt and tectonic activity: Constraints from zircon U–Pb and molybdenite Re–Os geochronology. Ore Geol. Rev. 2012, 48, 460–473. [Google Scholar] [CrossRef]
  52. Yang, L.-Q.; Deng, J.; Gao, X.; He, W.-Y.; Meng, J.-Y.; Santosh, M.; Yu, H.-J.; Yang, Z.; Wang, D. Timing of formation and origin of the Tongchanggou porphyry–skarn deposit: Implications for Late Cretaceous Mo–Cu metallogenesis in the southern Yidun Terrane, SE Tibetan Plateau. Ore Geol. Rev. 2017, 81, 1015–1032. [Google Scholar] [CrossRef]
  53. Moore, F.; Esmaeili, K.; Keshavarzi, B. Assessment of heavy metals contamination in stream water and sediments affected by the Sungun porphyry copper deposit, East Azerbaijan Province, Northwest Iran. Water Qual. Expo. Health 2011, 3, 37–49. [Google Scholar] [CrossRef]
  54. Amirihanza, H.; Shafieibafti, S.; Derakhshani, R.; Khojastehfar, S. Controls on Cu mineralization in central part of the Kerman porphyry copper belt, SE Iran: Constraints from structural and spatial pattern analysis. J. Struct. Geol. 2018, 116, 159–177. [Google Scholar] [CrossRef]
  55. Allahbakhshipoor, A.; Alipour-Asll, M.; Lentz, D.R. Genesis of the Kuh-Panj porphyry copper deposit, Kerman, Iran: Constraints from mineralization, geochemistry, fluid inclusion, and zircon UPb isotope systematics. J. Geochem. Explor. 2023, 253, 107276. [Google Scholar] [CrossRef]
  56. Parsapoor, A.; Khalili, M.; Tepley, F.; Maghami, M. Mineral chemistry and isotopic composition of magmatic, re-equilibrated and hydrothermal biotites from Darreh-Zar porphyry copper deposit, Kerman (Southeast of Iran). Ore Geol. Rev. 2015, 66, 200–218. [Google Scholar] [CrossRef]
  57. Aghazadeh, M.; Hou, Z.; Badrzadeh, Z.; Zhou, L. Temporal–spatial distribution and tectonic setting of porphyry copper deposits in Iran: Constraints from zircon U–Pb and molybdenite Re–Os geochronology. Ore Geol. Rev. 2015, 70, 385–406. [Google Scholar] [CrossRef]
  58. Seijo-Pardo, B.; Alonso-Betanzos, A.; Bennett, K.P.; Bolón-Canedo, V.; Josse, J.; Saeed, M.; Guyon, I. Biases in feature selection with missing data. Neurocomputing 2019, 342, 97–112. [Google Scholar] [CrossRef]
  59. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
  60. Ho, T.K. The random subspace method for constructing decision forests. IEEE Trans. Pattern Anal. Mach. Intell. 1998, 20, 832–844. [Google Scholar]
  61. Kwok, S.W.; Carter, C. Multiple decision trees. In Machine Intelligence and Pattern Recognition; Elsevier: Amsterdam, The Netherlands, 1990; Volume 9, pp. 327–335. [Google Scholar]
  62. Lee, T.-H.; Ullah, A.; Wang, R. Bootstrap aggregating and random forest. In Macroeconomic Forecasting in the Era of Big Data: Theory and Practice; Springer: Cham, Switzerland, 2020; pp. 389–429. [Google Scholar]
  63. Chen, T.; Guestrin, C. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, CA, USA, 13–17 August 2016; pp. 785–794. [Google Scholar]
  64. Li, Z.S.; Liu, Z.G. Feature Selection Algorithm Based on XGBoost. J. Commun./Tongxin Xuebao 2019, 40, 2019154-1–2019154-8. [Google Scholar]
  65. Agatonovic-Kustrin, S.; Beresford, R. Basic concepts of artificial neural network (ANN) modeling and its application in pharmaceutical research. J. Pharm. Biomed. Anal. 2000, 22, 717–727. [Google Scholar] [CrossRef]
  66. Popescu, M.-C.; Balas, V.E.; Perescu-Popescu, L.; Mastorakis, N. Multilayer perceptron and neural networks. WSEAS Trans. Circuits Syst. 2009, 8, 579–588. [Google Scholar]
  67. LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef]
  68. Sze, V.; Chen, Y.-H.; Yang, T.-J.; Emer, J.S. Efficient processing of deep neural networks: A tutorial and survey. Proc. IEEE 2017, 105, 2295–2329. [Google Scholar] [CrossRef]
  69. Mangalathu, S.; Hwang, S.-H.; Jeon, J.-S. Failure mode and effects analysis of RC members based on machine-learning-based SHapley Additive exPlanations (SHAP) approach. Eng. Struct. 2020, 219, 110927. [Google Scholar] [CrossRef]
  70. Yacouby, R.; Axman, D. Probabilistic extension of precision, recall, and f1 score for more thorough evaluation of classification models. In Proceedings of the First Workshop on Evaluation and Comparison of NLP Systems, Online, 20 November 2020; pp. 79–91. [Google Scholar]
  71. Van der Maaten, L.; Hinton, G. Visualizing data using t-SNE. J. Mach. Learn. Res. 2008, 9, 2579–2605. [Google Scholar]
  72. Van Der Maaten, L. Accelerating t-SNE using tree-based algorithms. J. Mach. Learn. Res. 2014, 15, 3221–3245. [Google Scholar]
  73. Sun, S.S.; McDonough, W.F. Chemical and isotopic systematics of oceanic basalts: Implications for mantle composition and processes. Geol. Soc. Lond. Spec. Publ. 1989, 42, 313–345. [Google Scholar] [CrossRef]
  74. Bergstra, J.; Bengio, Y. Random search for hyper-parameter optimization. J. Mach. Learn. Res. 2012, 13, 281–305. [Google Scholar]
  75. Metz, C.E. Basic principles of ROC analysis. Semin. Nucl. Med. 1978, 8, 283–298. [Google Scholar] [CrossRef] [PubMed]
  76. Candela, P.A. Controls on ore metal ratios in granite-related ore systems: An experimental and computational approach. Earth Environ. Sci. Trans. R. Soc. Edinb. 1992, 83, 317–326. [Google Scholar]
  77. Davidson, J.; Turner, S.; Handley, H.; Macpherson, C.; Dosseto, A. Amphibole “sponge” in arc crust? Geology 2007, 35, 787–790. [Google Scholar] [CrossRef]
  78. Yigit, O. Gold in Turkey—A missing link in Tethyan metallogeny. Ore Geol. Rev. 2006, 28, 147–179. [Google Scholar] [CrossRef]
  79. Yigit, O. Mineral deposits of Turkey in relation to Tethyan metallogeny: Implications for future mineral exploration. Econ. Geol. 2009, 104, 19–51. [Google Scholar] [CrossRef]
  80. Bozkurt, E.; Piper, J.D.; Winchester, J. Tectonics and Magmatism in Turkey and the Surrounding Area; Geological Society of London: London, UK, 2000. [Google Scholar]
  81. Moix, P.; Beccaletto, L.; Kozur, H.W.; Hochard, C.; Rosselet, F.; Stampfli, G.M. A new classification of the Turkish terranes and sutures and its implication for the paleotectonic history of the region. Tectonophysics 2008, 451, 7–39. [Google Scholar] [CrossRef]
  82. Bamba, T. Ophiolite and related copper deposits of the Ergani mining district, Southeastern Turkey. Bull. Miner. Res. Explor. 1976, 86, 36–50. [Google Scholar]
  83. Delibaş, O.; Moritz, R.; Ulianov, A.; Chiaradia, M.; Saraç, C.; Revan, K.M.; Göç, D. Cretaceous subduction-related magmatism and associated porphyry-type Cu–Mo prospects in the Eastern Pontides, Turkey: New constraints from geochronology and geochemistry. Lithos 2016, 248, 119–137. [Google Scholar] [CrossRef]
  84. Rabayrol, F. Late Cenozoic Post-Subduction Tectonic, Magmatic and Metallogenic Evolution of the Anatolide-Tauride Orogenic Belt, Turkey. Doctoral Dissertation, University of British Columbia, Vancouver, BC, Canada, 2018. [Google Scholar]
  85. Zhao, J.X.; Su, B.X.; Xiao, Y.; Hui, K.X.; Qin, K.Z. Evolution of magma oxidation states and volatile components in the Cenozoic porphyry ore systems in the western Turkey, Tethyan domain: Constrains from the compositions of zircon and apatite. Acta Petrol. Sin. 2021, 37, 2339–2363. [Google Scholar]
  86. Shen, Z.M.; Feng, Z.J.; Ren, Q.J. Classification Quantitative Discrimination Model of Porphyry Copper (Molybdenum) Deposits of Different Mineralization Types. Contrib. Geol. Miner. Resour. Res. 1993, 8, 95–103. [Google Scholar]
  87. Kirkland, C.; Smithies, R.; Taylor, R.; Evans, N.; McDonald, B. Zircon Th/U ratios in magmatic environs. Lithos 2014, 212–215, 397–414. [Google Scholar] [CrossRef]
  88. Kinny P D, Maas R. Lu–Hf and Sm–Nd isotope systems in zircon. Reviews in mineralogy and geochemistry. Rev. Mineral. Geochem. 2003; 53, 327–341. [CrossRef]
  89. Xu, L.-L.; Zhu, J.-J.; Huang, M.-L.; Pan, L.-C.; Hu, R.; Bi, X.-W. Genesis of hydrous-oxidized parental magmas for porphyry Cu (Mo, Au) deposits in a post-collisional setting: Examples from the Sanjiang region, SW China. Miner. Depos. 2023, 58, 161–196. [Google Scholar] [CrossRef]
  90. Yang, M.; Zhao, F.; Liu, X.; Qing, H.; Chi, G.; Li, X.; Lai, C. Contribution of magma mixing to the formation of porphyry-skarn mineralization in a post-collisional setting: The Machangqing Cu-Mo-(Au) deposit, Sanjiang tectonic belt, SW China. Ore Geol. Rev. 2020, 122, 103518. [Google Scholar] [CrossRef]
  91. Bao, X.; He, W.; Mao, J.; Liang, T.; Wang, H.; Zhou, Y.; Wang, J. Redox states and genesis of Cu-and Au-mineralized granite porphyries in the Jinshajiang Cu–Au metallogenic belt, SW China: Studies on the zircon chemistry. Miner. Depos. 2023, 58, 1123–1142. [Google Scholar] [CrossRef]
  92. Fu, Y.; Sun, X.M.; Lin, H.; Zhou, H.Y.; Li, X.; Ouyang, X.Q.; Jiang, L.Y.; Shi, G.Y.; Liang, Y.H. Geochronology of the giant Beiya gold-polymetallic deposit in Yunnan Province, Southwest China and its relationship with the petrogenesis of alkaline porphyry. Ore Geol. Rev. 2015, 71, 138–149. [Google Scholar] [CrossRef]
  93. Gao, X.-Q.; He, W.-Y.; Gao, X.; Bao, X.-S.; Yang, Z. Constraints of magmatic oxidation state on mineralization in the Beiya alkali-rich porphyry gold deposit, Western Yunnan, China. Solid Earth Sci. 2017, 2, 65–78. [Google Scholar] [CrossRef]
  94. Meng, X.; Mao, J.; Zhang, C.; Zhang, D.; Liu, H. Melt recharge, fO2-T conditions, and metal fertility of felsic magmas: Zircon trace element chemistry of Cu-Au porphyries in the Sanjiang orogenic belt, Southwest China. Miner. Depos. 2018, 53, 649–663. [Google Scholar] [CrossRef]
  95. Huang, M.-L.; Bi, X.-W.; Richards, J.P.; Hu, R.-Z.; Xu, L.-L.; Gao, J.-F.; Zhang, X.-C. High water contents of magmas and extensive fluid exsolution during the formation of the Yulong porphyry Cu-Mo deposit, Eastern Tibet. J. Asian Earth Sci. 2019, 176, 168–183. [Google Scholar] [CrossRef]
  96. Gao, X.; Yang, L.-Q.; He, W.-Y.; Groves, D. Redox conditions, compositional parameters, and indirect subduction-related source of Cretaceous Sn and Cu–Mo fertile post-subduction granites in the Yidun Terrane of Eastern Tibet. Ore Geol. Rev. 2021, 139, 104506. [Google Scholar] [CrossRef]
  97. Li, J.; Qin, K.; Li, G.; Cao, M.; Xiao, B.; Chen, L.; McInnes, B.I. Petrogenesis and thermal history of the Yulong porphyry copper deposit, Eastern Tibet: Insights from U-Pb and U-Th/He dating, and zircon Hf isotope and trace element analysis. Mineral. Petrol. 2012, 105, 201–221. [Google Scholar] [CrossRef]
  98. Yang, Z.-M.; Goldfarb, R.; Chang, Z.-S. Generation of post-collisional porphyry copper deposits in Southern Tibet triggered by subduction of the Indian continental plate. In Tectonics and Metallogeny of the Tethyan Orogenic Belt; Society of Economic Geologists: Littleton, CO, USA, 2016. [Google Scholar]
  99. Wang, R.; Richards, J.P.; Hou, Z.-Q.; Yang, Z.-M.; Gou, Z.-B.; DuFrane, S.A. Increasing magmatic oxidation state from Paleocene to Miocene in the Eastern Gangdese Belt, Tibet: Implication for collision-related porphyry Cu-Mo±Au mineralization. Econ. Geol. 2014, 109, 1943–1965. [Google Scholar] [CrossRef]
  100. Hezarkhani, A. Petrology of the intrusive rocks within the Sungun porphyry copper deposit, Azerbaijan, Iran. J. Asian Earth Sci. 2006, 27, 326–340. [Google Scholar] [CrossRef]
  101. Asadi, S.; Moore, F.; Zarasvandi, A. Discriminating productive and barren porphyry copper deposits in the southeastern part of the Central Iranian volcano-plutonic belt, Kerman region, Iran: A review. Earth-Sci. Rev. 2014, 138, 25–46. [Google Scholar] [CrossRef]
  102. Zhang, W.; Zhao, Z.; Liu, D.; Qiu, K.; Wang, Q.; Zhu, D.-C.; Hou, Z. Mineralogy and geochemistry of the Zedong Late Cretaceous (~94 Ma) biotite granodiorite in the Southern Lhasa Terrane: Implications for the tectonic setting and Cu-Au mineralization. Lithos 2023, 446, 107158. [Google Scholar] [CrossRef]
  103. Wu, S.; Zheng, Y.; Sun, X. Subduction metasomatism and collision-related metamorphic dehydration controls on the fertility of porphyry copper ore-forming high Sr/Y magma in Tibet. Ore Geol. Rev. 2016, 73, 83–103. [Google Scholar] [CrossRef]
  104. Hu, Y.B. Petrogenesis and metallogenetic implications of adakites in the Gangdese porphyry copper belt. Doctoral Dissertation, University of Chinese Academy of Sciences, Beijing, China, 2015. [Google Scholar]
  105. Zhang, B.; Liu, J.; Chen, W.; Zhu, Z.; Sun, C. Late Eocene magmatism of the Eastern Qiangtang Block (Eastern Tibetan Plateau) and its geodynamic implications. J. Asian Earth Sci. 2020, 195, 104329. [Google Scholar] [CrossRef]
  106. Tang, P.; Tang, J.; Wang, Y.; Lin, B.; Leng, Q.; Zhang, Q.; Wu, C. Genesis of the Lakang’e porphyry Mo (Cu) deposit, Tibet: Constraints from geochemistry, geochronology, Sr-Nd-Pb-Hf isotopes, zircon and apatite. Lithos 2021, 380, 105834. [Google Scholar] [CrossRef]
  107. Sun, X.; Lu, Y.-J.; McCuaig, T.C.; Zheng, Y.-Y.; Chang, H.-F.; Guo, F.; Xu, L.-J. Miocene ultrapotassic, high-Mg dioritic, and adakite-like rocks from Zhunuo in Southern Tibet: Implications for mantle metasomatism and porphyry copper mineralization in collisional orogens. J. Petrol. 2018, 59, 341–386. [Google Scholar] [CrossRef]
  108. Li, Q.; Yang, Z.; Wang, R.; Sun, M.; Qu, H. Zircon trace elemental and Hf-O isotopic compositions of the Miocene magmatic suite in the giant Qulong porphyry copper deposit Southern Tibet. Acta Petrol. Mineral. 2021, 40, 1–26. [Google Scholar]
  109. Zhang, J.; Chen, X.; Zhang, L.; Li, J. Insights into magma evolution and fluid evolution from zircon geochronology and geochemistry of the Xiongcun porphyry Cu-Mo deposit, Tibet, China. Econ. Geol. 2022, 117, 879–899. [Google Scholar]
  110. Zhu, X.; Li, G.; Liu, H.; Chen, H.; Ma, D.; Liu, C.; Wei, L. High oxidation magmatic evolution in the Naruo porphyry Cu deposit, Tibet, China. Gondwana Res. 2019, 76, 26–43. [Google Scholar] [CrossRef]
  111. Li, G.-M.; Qin, K.-Z.; Li, J.-X.; Evans, N.J.; Zhao, J.-X.; Cao, M.-J.; Zhang, X.-N. Cretaceous Magmatism and Metallogeny in the Bangong–Nujiang Metallogenic Belt, Central Tibet: Evidence from Petrogeochemistry, Zircon U–Pb Ages, and Hf–O Isotopic Compositions. Gondwana Res. 2017, 41, 110–127. [Google Scholar] [CrossRef]
  112. Zhang, Z.; Wang, L.; Tang, P.; Lin, B.; Sun, M.; Qi, J.; Yang, Z. Geochemistry and zircon trace elements composition of the Miocene ore-bearing biotite monzogranite porphyry in the Demingding porphyry Cu–Mo deposit, Tibet: Petrogenesis and implication for magma fertility. Geol. J. 2020, 55, 4525–4542. [Google Scholar] [CrossRef]
  113. Li, Q.; Sun, X.; Lu, Y.; Wang, F.; Hao, J. Apatite and zircon compositions for Miocene mineralizing and barren intrusions in the Gangdese porphyry copper belt of southern Tibet: Implication for ore control. Ore Geol. Rev. 2021, 139, 104474. [Google Scholar] [CrossRef]
  114. Lu, Y.-J.; Loucks, R.R.; Fiorentini, M.; McCuaig, T.C.; Evans, N.J.; Yang, Z.-M.; Kobussen, A. Zircon compositions as a pathfinder for porphyry Cu±Mo±Au deposits. Miner. Depos. 2016, 51, 249–267. [Google Scholar]
  115. Sun, X.; Hollings, P.; Lu, Y.-J. Geology and origin of the Zhunuo porphyry copper deposit, Gangdese Belt, Southern Tibet. Miner. Depos. 2021, 56, 457–480. [Google Scholar] [CrossRef]
  116. Hwang, J.; Park, J.-W.; Wan, B.; Honarmand, M. Contrasting platinum-group element geochemistry of post-collisional porphyry Cu±Au ore-bearing and barren suites in the central and southeastern Urumieh-Dokhtar magmatic arc, Iran. Miner. Depos. 2023, 58, 1583–1603. [Google Scholar] [CrossRef]
  117. Xie, F.; Tang, J.; Chen, Y.; Lang, X. Apatite and zircon geochemistry of Jurassic porphyries in the Xiongcun district, Southern Gangdese porphyry copper belt: Implications for petrogenesis and mineralization. Ore Geol. Rev. 2018, 96, 98–114. [Google Scholar] [CrossRef]
  118. Yinqiao, Z.; Wenting, H.; Huaying, L.; Jing, W.; Shuping, L.; Xiuzhang, W. Identification of porphyry genetically associated with mineralization and its zircon U-Pb and biotite Ar-Ar age of the Xiongcun Cu-Au deposit, Southern Gangdese, Tibet. Acta Petrolog. Sin. 2015, 31, 2053–2062. [Google Scholar]
  119. Wang, L.; Zheng, Y.; Hou, Z.; Xue, C.; Yang, Z.; Shen, Y.; Li, X.; Ghaffar, A. The subduction-related Saindak porphyry Cu-Au deposit formed by remelting of a thickened juvenile lower crust underneath the Chagai Belt, Pakistan. Ore Geol. Rev. 2022, 149, 105062. [Google Scholar] [CrossRef]
  120. Paolillo, L.; Chiaradia, M.; Ulianov, A. Zircon petrochronology of the Kışladaǧ porphyry Au deposit (Turkey). Econ. Geol. 2022, 117, 401–422. [Google Scholar] [CrossRef]
  121. Zhou, Y.; Xu, B.; Hou, Z.Q.; Wang, R.; Zheng, Y.C.; He, W.Y. Petrogenesis of Cenozoic high–Sr/Y shoshonites and associated mafic microgranular enclaves in an intracontinental setting: Implications for porphyry Cu-Au mineralization in Western Yunnan, China. Lithos 2019, 324, 39–54. [Google Scholar] [CrossRef]
  122. Zhang, J.; Peng, T.; Fan, W.; Zhao, G.; Dong, X.; Gao, J.; Chen, L. Petrogenesis of the Early Cretaceous granitoids and its mafic enclaves in the Northern Tengchong Terrane, Southern Margin of the Tibetan Plateau and its tectonic implications. Lithos 2018, 318, 283–298. [Google Scholar] [CrossRef]
  123. Zeng, Y.-C.; Xu, J.-F.; Huang, F.; Li, M.-J.; Chen, Q. Generation of the 105–100 Ma Dagze volcanic rocks in the North Lhasa Terrane by lower crustal melting at different temperature and depth: Implications for tectonic transition. GSA Bull. 2020, 132, 1257–1272. [Google Scholar] [CrossRef]
  124. Li, H.; Wang, M.; Zeng, X.-W.; Luo, A.-B.; Yu, Y.-P.; Zeng, X.-J. Generation of Jurassic high-Mg diorite and plagiogranite intrusions of the Asa area, Tibet: Products of intra-oceanic subduction of the Meso-Tethys Ocean. Lithos 2020, 362, 105481. [Google Scholar] [CrossRef]
  125. Babazadeh, S.; D’Antonio, M.; Cottle, J.M.; Ghalamghash, J.; Raeisi, D.; An, Y. Constraints from geochemistry, zircon U-Pb geochronology and Hf-Nd isotopic compositions on the origin of Cenozoic volcanic rocks from Central Urumieh-Dokhtar magmatic arc, Iran. Gondwana Res. 2021, 90, 27–46. [Google Scholar] [CrossRef]
  126. Liu, H.-N.; Li, X.-W.; Mo, X.-X.; Xu, J.-F.; Liu, J.-J.; Dong, G.-C.; Yu, H.-X. Early Mesozoic crustal evolution in the NW segment of West Qinling, China: Evidence from diverse intermediate–felsic igneous rocks. Lithos 2021, 396, 106187. [Google Scholar] [CrossRef]
  127. Ji, C.; Yan, L.-L.; Lu, L.; Jin, X.; Huang, Q.; Zhang, K.-J. Anduo Late Cretaceous high-K calc-alkaline and shoshonitic volcanic rocks in Central Tibet, Western China: Relamination of the subducted Meso-Tethyan oceanic plateau. Lithos 2021, 400, 106345. [Google Scholar] [CrossRef]
  128. Zeng, Y.-C.; Xu, J.-F.; Li, M.-J.; Chen, J.-L.; Wang, B.-D.; Huang, F.; Ren, S.-H. Late Eocene two-pyroxene trachydacites from the Southern Qiangtang Terrane, Central Tibetan Plateau: High-temperature melting of overthickened and dehydrated lower crust. J. Petrol. 2021, 62, egab080. [Google Scholar] [CrossRef]
  129. Chen, H.-B.; Ji, W.-Q.; Zhang, S.-H.; Jiang, H.; Xu, Q.; Wu, F.-Y. Disclosing the most intense magmatic flare-up in Southern Tibet: Insights from study of the Pangduo volcanic-plutonic complex. Lithos 2023, 454, 107267. [Google Scholar] [CrossRef]
  130. Fan, H.; Zhang, M.; Huang, F.; Xu, J.; Liu, X.; Zeng, Y.; Yu, H. Subducted Oceanic Plateau Fed Crustal Growth: Insights from Amdo Dacites in Central Tibetan Plateau. Lithos 2022, 434, 106944. [Google Scholar] [CrossRef]
  131. Zhang, Y.-F.; Xu, B.; Hou, Z.-Q.; Zhao, Y.; Wang, Z.-X.; Shen, J.-Q.; Li, E.-Q. Zircon Xenocryst Geochronology and Implications for the Lhasa Terrane Evolution: Insights from Cenozoic Volcanic Rocks (Coqen, Tibet). J. Asian Earth Sci. 2023, 255, 105763. [Google Scholar] [CrossRef]
  132. Li, Q.-H.; Lu, L.; Zhang, K.-J.; Yan, L.-L.; Huangfu, P.; Hui, J.; Ji, C. Late Cretaceous Post-Orogenic Delamination in the Western Gangdese Arc: Evidence from Geochronology, Petrology, Geochemistry, and Sr–Nd–Hf Isotopes of Intermediate–Acidic Igneous Rocks. Lithos 2022, 424, 106763. [Google Scholar] [CrossRef]
  133. Chen, X.-D.; Li, B.; Yu, M.; Zhang, W.-D.; Zhu, L. Generation of Crystal-Rich Rhyodacites by Fluid-Induced Crystal-Mush Rejuvenation: Perspective from the Late Triassic Nageng (Sub-)Volcanic Complex of the East Kunlun Orogen, NW China. Chem. Geol. 2022, 599, 120833. [Google Scholar] [CrossRef]
  134. Liu, X.; Zhan, Q.-Y.; Zhu, D.-C.; Weinberg, R.F.; Wang, Q.; Xie, J.-C.; Zhao, Z.-D. Large Zircon Age Spans Record Multi-Stage History of Batholith Assembly: Insights from the Late Triassic Dongcuo Batholith in the Eastern Tibetan Plateau. J. Asian Earth Sci. 2022, 231, 105220. [Google Scholar] [CrossRef]
  135. Sepidbar, F.; Mirnejad, H.; Ma, C.; Moghadam, H.S. Identification of Eocene-Oligocene Magmatic Pulses Associated with Flare-Up in East Iran: Timing and Sources. Gondwana Res. 2018, 57, 141–156. [Google Scholar] [CrossRef]
  136. Honarmand, M.; Rashidnejad Omran, N.; Neubauer, F.; Hashem Emami, M.; Nabatian, G.; Liu, X.; Chen, B. Laser-ICP-MS U–Pb Zircon Ages and Geochemical and Sr–Nd–Pb Isotopic Compositions of the Niyasar Plutonic Complex, Iran: Constraints on Petrogenesis and Tectonic Evolution. Int. Geol. Rev. 2014, 56, 104–132. [Google Scholar] [CrossRef]
  137. Ma, X.; Xu, Z.; Chen, X.; Meert, J.G.; He, Z.; Liang, F.; Ma, S. The Origin and Tectonic Significance of the Volcanic Rocks of the Yeba Formation in the Gangdese Magmatic Belt, South Tibet. J. Earth Sci. 2017, 28, 265–282. [Google Scholar] [CrossRef]
  138. Wei, Y. The Geochronology, Geochemistry and Petrogenesis of the Volcanic Rocks of Yeba Formation, Southern Tibet. Master’s Thesis, China University of Geosciences, Beijing, China, 2014; pp. 1–59. [Google Scholar]
  139. Razique, A. Magmatic Evolution and Genesis of the Giant Reko Diq H14-H15 Porphyry Copper-Gold Deposit, District Chagai, Balochistan-Pakistan. Master’s Thesis, University of British Columbia, Vancouver, BC, Canada, 2013. [Google Scholar]
Minerals 14 00945 g001
Figure 1. Tectonic setting and locations of deposits in the dataset of this study in the Tethys (adapted from [13,15,26]).
Figure 1. Tectonic setting and locations of deposits in the dataset of this study in the Tethys (adapted from [13,15,26]).
Minerals 14 00945 g001
Minerals 14 00945 g002
Figure 2. (a) Original histograms of the zircon trace elements; (b) the histograms of the zircon trace elements after the log transformation; (c) the histograms of the zircon trace elements after the Z-score standardization. Due to the abundance of indicators, only five trace elements were selected as representatives for plotting.
Figure 2. (a) Original histograms of the zircon trace elements; (b) the histograms of the zircon trace elements after the log transformation; (c) the histograms of the zircon trace elements after the Z-score standardization. Due to the abundance of indicators, only five trace elements were selected as representatives for plotting.
Minerals 14 00945 g002
Minerals 14 00945 g003
Figure 3. (a) Schematic diagram of random forest; (b) schematic diagram of DNN.
Figure 3. (a) Schematic diagram of random forest; (b) schematic diagram of DNN.
Minerals 14 00945 g003
Minerals 14 00945 g004
Figure 4. Chondrite-normalized zircon trace elements patterns of zircons from the fertile samples (a–e) and barren samples (f) in the Tethyan domain. Normalization after Sun and McDonough [73].
Figure 4. Chondrite-normalized zircon trace elements patterns of zircons from the fertile samples (a–e) and barren samples (f) in the Tethyan domain. Normalization after Sun and McDonough [73].
Minerals 14 00945 g004
Minerals 14 00945 g005
Figure 5. Scatter plots of zircons in porphyry Cu-related deposits based on traditional partitioning methods: (a) Eu/Eu* vs. (Ce/Nd)/Y; (b) Eu/Eu* vs. Ce/Ce*; (c) Dy/Yb vs. Eu/Eu*; (d) (Ce/Nd)/Y vs. (Eu/Eu*)/Y*10,000; (e) Dy/Yb vs. (Eu/Eu*)/Y*10,000. Partition conditions are from [11,20].
Figure 5. Scatter plots of zircons in porphyry Cu-related deposits based on traditional partitioning methods: (a) Eu/Eu* vs. (Ce/Nd)/Y; (b) Eu/Eu* vs. Ce/Ce*; (c) Dy/Yb vs. Eu/Eu*; (d) (Ce/Nd)/Y vs. (Eu/Eu*)/Y*10,000; (e) Dy/Yb vs. (Eu/Eu*)/Y*10,000. Partition conditions are from [11,20].
Minerals 14 00945 g005
Minerals 14 00945 g006
Figure 6. Plot of zircon Eu/Eu* vs. Ce/Nd from different ages of porphyry copper deposits.
Figure 6. Plot of zircon Eu/Eu* vs. Ce/Nd from different ages of porphyry copper deposits.
Minerals 14 00945 g006
Minerals 14 00945 g007
Figure 7. Confusion matrix diagrams of test set using three machine learning models.
Figure 7. Confusion matrix diagrams of test set using three machine learning models.
Minerals 14 00945 g007
Minerals 14 00945 g008
Figure 8. Feature importance. (a) Random forest feature ranking; (b) SHAP scatter plot.
Figure 8. Feature importance. (a) Random forest feature ranking; (b) SHAP scatter plot.
Minerals 14 00945 g008
Minerals 14 00945 g009
Figure 9. Scatter plots of zircon trace element content: (a) Eu vs. Ti; (b) Ti vs. Ce; (c) Hf vs. Dy; (d) Lu vs. Dy; (e) Yb vs. Lu.
Figure 9. Scatter plots of zircon trace element content: (a) Eu vs. Ti; (b) Ti vs. Ce; (c) Hf vs. Dy; (d) Lu vs. Dy; (e) Yb vs. Lu.
Minerals 14 00945 g009
Minerals 14 00945 g010
Figure 10. Scatter plots of zircons in Turkey: (a) Eu vs. Ti; (b) Dy vs. Ce. The classification results stem from the random forest.
Figure 10. Scatter plots of zircons in Turkey: (a) Eu vs. Ti; (b) Dy vs. Ce. The classification results stem from the random forest.
Minerals 14 00945 g010
Minerals 14 00945 g011
Figure 11. Feature importance ranking based on random forest.
Figure 11. Feature importance ranking based on random forest.
Minerals 14 00945 g011
Minerals 14 00945 g012
Figure 12. t-SNE scatter plots of different types of deposits.
Figure 12. t-SNE scatter plots of different types of deposits.
Minerals 14 00945 g012
Minerals 14 00945 g013
Figure 13. The t-SNE scatter plot of zircon trace elements from different types of porphyry deposits.
Figure 13. The t-SNE scatter plot of zircon trace elements from different types of porphyry deposits.
Minerals 14 00945 g013
Table 1. Information on the deposit where the compiled zircon data are located [43,44,45,46,47,48,49,50,51,52,53,54,55,56,57].
Table 1. Information on the deposit where the compiled zircon data are located [43,44,45,46,47,48,49,50,51,52,53,54,55,56,57].
DepositTypeTonnage(Mt) and GradeZircon CountFertility
GangdeseBelt
XiongcunCu-Au874, Cu: 2.96 @0.34%, Au: 218.4t @0.25g/t283Fertile
QulongCu-Mo2200, Cu: 8 @0.3–0.6%, Mo: 0.03–0.06%97Fertile
Jiama
Jiru		Cu-Mo1055, Cu: >5 @1.16%, Mo: 0.024%77Fertile
Cu-Mo42, Cu: 0.18 @0.43%32Fertile
Lakang’e
Zhunuo
Demingding
Chongjiang		Mo-CuCu: >0.1, Cu: 0.13–0.49%37Fertile
Cu-Mo403, Cu: 2.3 @0.57%54Fertile
Mo-CuCu: >0.5 @0.26 %, Mo: >0.5 @0.14 %15Fertile
Cu-Mo45, Cu: 0.18 @0.4%, Mo: 0.01 @0.02%28Fertile
Dabu
Luerma		Cu-Mo250, Cu: 0.5 @0.2%, Mo: 0.05 @0.02%15Fertile
CuCu: 0.26% ~ 0.73%, Au: 1.2 ~ 37.7 g/t69Fertile
\\\485Barren
Bangong–Nujiang Belt
NaruoCu578, Cu: 2.37 @0.41%, Mo: 104 @0.18%100Fertile
GuaguaziCu\35Fertile
SaidengnanCu-Au\11Fertile
\\\214Barren
Ailaoshan–Red River Belt
YulongCu-Mo-Au1006, Cu: 6.24 @0.62%, Mo: 0.42 @0.04%, Au: 50.3t @0.05g/t118Fertile
MachangqingCu-Mo-AuCu: 0.25 @0.44%, Mo: 0.14%, Au: 0.06g/t65Fertile
BeiyaAuAu: 323t @2.47g/t, Cu: 56.68 @0.52%170Fertile
TongchangCu-Mo-AuCu: 37,958t @1.48%, Mo: 30,370t @0.218%, Au: 0.13–0.25g/t47Fertile
\\\302Barren
Yidun–Zhongdian Belt
PulangCu-Au804, Cu: 446.8 @0.52%, Au: 0.18 g/t64Fertile
RelinCu-Mo6.6, Mo: 0.049%22Fertile
XiuwacuMo-W6.6, Mo: 13,627t @0.34%, WO3: 8431t @0.28%74Fertile
HongshanCu-Mo65.1, Cu: 0.64 @1.23 %, Mo: 5769t @0.03 %122Fertile
TongchanggouMo-Cu142.5, Mo: 0.3 @0.3 %, Cu: 34 kt @0.8%45Fertile
\\\146Barren
Sahand–Bazman Belt
SungunCu-Mo850, Cu: 0.76 %, Mo: 0.01%45Fertile
Sar CheshmehCu-Mo1700, Cu: 0.65%, Au: 0.06 g/t, Mo: 0.03%103Fertile
Kuh-PanjCu100, Cu: 100 @0.21%18Fertile
Darreh-ZarCu-Mo283, Cu: 1.1 @0.38%1Fertile
MeidukCu-Mo500, Cu: 0.86%, Mo: 0.007%2Fertile
\\\104Barren
Chagai Belt
SaindakCu-AuCu: 27Mt 64Fertile
Reko diqCu-AuCu: 1.23Mt@0.41 %, Au: 59.3 t@0.22 g/t88Fertile
\\\30Barren
Other Districts
\\\83Fertile
\\\486Barren
Table 2. The best parameter combination selected by random search and the classifier’s accuracy on the cross-validation set and independent test set.
Table 2. The best parameter combination selected by random search and the classifier’s accuracy on the cross-validation set and independent test set.
Base ClassifierSelected ParametersCross-Validation AccuracyTest
Accuracy		AUC
XGBoostlearning_rate = 0.3
min_child_weight = 3
n_estimators = 150		87.0%88.6%0.985
Random Forestmin_samples_leaf = 1
min_samples_split = 2
n_estimators= 190		89.5%91.0%0.990
DNNActivation = tanh
hidden_layer_sizes = (70,10,10)
Solver = adam		90.6%90.8%0.982
Table 3. Sample sources and prediction results of zircons.
Table 3. Sample sources and prediction results of zircons.
DepositSampleLithologyPrediction
(0-Barren; 1-Fertile)
RFXGBoostDNN
Saricayiryayla
Porphyry Cu-Mo		18UD08-2Granite001
18MU03Granodiorite001
Halilaǧa
Porphyry Cu-Mo		18HA10-2Andesite111
18HA02-1Rhyodacite000
18HA03Streatted Lava000
18HA06-1Dacite111
18HA04-1Feldsparphyric Rock111
18HA05-1Andesite000
Kisladaǧ
Porphyry Au		18US12-3Trachyandensite001
18US10-1Andesite111
18US08-1Andesite111
18US09-1Dacite000
18US05-1Dacite001
18US07-2Dacite001
18US06-2Dacite111
Pınarbaşı
Porphyry Mo-Cu		18GD02-1CGranite110
18GD07-5Granite111
18GD02-1DMafic Enclave000
18GD04-3Monzonite100
18GD06-2Granite010
Table 4. Classification report of the mineral deposit types based on the random forest model.
Table 4. Classification report of the mineral deposit types based on the random forest model.
TypePrecisionRecallF1-ScoreSupport
Au-Rich0.800.840.8219
Cu0.950.950.9522
Cu-Au1.000.800.8910
Cu-Mo0.800.670.7318
Cu-Mo-Au0.860.930.8945
Mo-Rich0.880.880.8816
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Guo, J.; He, W.-Y. Application of Machine Learning to Characterize Metallogenic Potential Based on Trace Elements of Zircon: A Case Study of the Tethyan Domain. Minerals 2024, 14, 945. https://doi.org/10.3390/min14090945

AMA Style

Guo J, He W-Y. Application of Machine Learning to Characterize Metallogenic Potential Based on Trace Elements of Zircon: A Case Study of the Tethyan Domain. Minerals. 2024; 14(9):945. https://doi.org/10.3390/min14090945

Chicago/Turabian Style

Guo, Jin, and Wen-Yan He. 2024. "Application of Machine Learning to Characterize Metallogenic Potential Based on Trace Elements of Zircon: A Case Study of the Tethyan Domain" Minerals 14, no. 9: 945. https://doi.org/10.3390/min14090945

APA Style

Guo, J., & He, W.-Y. (2024). Application of Machine Learning to Characterize Metallogenic Potential Based on Trace Elements of Zircon: A Case Study of the Tethyan Domain. Minerals, 14(9), 945. https://doi.org/10.3390/min14090945

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop