New Insights into the Xiongbaxi–Yalongri Cu-W(-Mo) Deposit (Tibet): Scheelite Geochemistry and Machine Learning Constraints on Ore-Forming Fluid Evolution and Genetic Type

Abstract

The Zhunuo ore district, at the western end of the Gangdese porphyry Cu belt, hosts significant Cu mineralization and newly recognized W mineralization dominated by scheelite. However, the genetic relationship between scheelite and porphyry mineralization, and the evolution of ore-forming fluids remain poorly constrained. To address this, scheelite samples from multiple locations were analyzed for major elements (EMPA), in situ trace elements (LA-ICP-MS), and internal textures (cathodoluminescence, CL). These data, combined with machine learning methods, were used to determine scheelite genetic types and reconstruct fluid evolution. REE patterns and CL textures reveal three scheelite generations in Yalongri (early Sch I c, middle Sch I b, late Sch I a), two in Zhigunong (early Sch II a, late Sch II b), and one in Xiongbaxi (Sch III). Low Na (0–329 ppm) and Nb (3.9–39 ppm) relative to high ΣREE + Y-Eu (16–3857 ppm), indicate that the dominant substitution mechanism is 3Ca²⁺ = 2REE³⁺ + □Ca (□Ca = Ca vacancy). δEu values > 1 in Sch I a, Sch I b, Sch II a, and Sch II b indicate reducing fluids, whereas δEu < in Sch I c and Sch III reflects oxidizing conditions. Variations in REE, Mo, and Sr contents suggest that ore-forming fluids in Yalongri evolved from oxidizing to reducing conditions, with late-stage scheelite undergoing dissolution–reprecipitation. Zhigunong records two reducing stages: an early REE-rich-Mo-poor stage and a later REE-poor-Mo-rich stage. Xiongbaxi records a single oxidizing, REE-rich, Mo-rich stage. Scheelite exhibits low-to-moderate Sr/Mo ratios (0.02–6.10), consistent with a magmatic–hydrothermal origin, and relatively uniform Y/Ho ratios (12–59) indicating stable crystallization conditions. A Random Forest model classifies scheelite into orogenic, porphyry, skarn, and greisen types. Overall, the results indicate that ore-forming fluids evolved from oxidizing to reducing conditions, favoring metal transport and enrichment. Integrated geochemical and machine learning evidence suggest, strong potential for porphyry-type Cu-W(-Mo) mineralization in Yalongri and Zhigunong, and skarn-type W-Mo mineralization in Xiongbaxi, providing important guidance for future exploration in the western Gangdese metallogenic belt.

1. Introduction

Scheelite is the principal host mineral for tungsten and is widely distributed in magmatic-hydrothermal deposits, including orogenic, skarn, porphyry, greisen, and quartz-vein systems [1,2,3,4,5,6,7]. Its crystal structure, composeds of (WO₄) tetrahedra and (CaO₈) dodecahedra, allows for a wide range of trace elements, such as REE, Sr, Pb, Y, Mn, and Mo substitutitute for Ca²⁺ or W⁶⁺ through coupled substitutions [5,7,8,9,10,11,12,13]. The most common substitution mechanisms are [9,14,15]: (1) 2Ca²⁺ = REE³⁺ + Na⁺; (2) Ca²⁺ + W⁶⁺ = REE³⁺ + Nb⁵⁺; and (3) 3Ca²⁺ = 2REE³⁺ + □Ca, where □Ca denotes a Ca vacancy. Because of its distinctive crystal chemistry and trace element composition, scheelite is an effective tracer of ore-forming fluid sources, physicochemical conditions, and fluid evolution, providing critical constraints on deposit genesis [1,4,5,9,15,16,17,18,19,20,21,22,23].

With advances in LA-ICP-MS trace element analysis, scheelite geochemistry has been increasingly applied to deposit genetic classification. Song et al. (2014) proposed an LREE-MREE-HREE ternary diagram to distinguish skarn-type W-Mo, vein-type Au-W, and porphyry-type W-Mo deposits [20]. Sciuba et al. (2020) analyzed scheelite from 25 orogenic gold deposits using Partial Least Squares-Discriminant Analysis (PLS-DA), and showed enrichment in Sr, Mo, Eu, As, and Sr/Mo ratios, with less distinctive REE patterns than scheelite from other deposit types [3]. Nie et al. (2023) applied Discriminant Projection Analysis (DPA) to scheelite from 86 global deposits and demonstrated that Mo, Sr, REE, Y, Ta, Pb, and δEu effectively discriminate among deposit types [6]. Despite these advances, classification accuracy and precision remain limited, highlighting the need for more robust discrimination approaches.

The Gangdese metallogenic belt in Tibet is a major Cu-Mo province, hosting numerous porphyry deposits, including Sharang, Zhunuo, Jiru, Chongjiang, Tinggong, Dabu, Qulong, and Jiama [24,25,26,27,28,29,30]. The giant Qulong porphyry deposit contains abundant scheelite at depth [19], providing new constraints on ore-forming fluid evolution, while scheelite occurences in the Jiama and Nuri Cu-polymetallic deposits further inform deposit classification [31,32,33]. Recent exploration in the western Gangdese belt has identified several scheelite-bearing deposits, including Xiongbaxi, Zhigunong, and Yalongri [34], offering valuable opportunities to investigate fluid evolution and genetic types of W-bearing mineralization. In this study, EMPA and LA-ICP-MS analyses of scheelite from the Xiongbaxi-Yalongri area are used to constrain trace element substitution mechanisms, reconstruct ore-forming fluid evolution, and determine deposit genetic types. In addition, a global scheelite database was compiled and analyzed using a Random Forest model to classify deposit types and predict the genetic affinity of scheelite in the study area. These integrated petrographic, geochemical, and machine learning approaches provide new insights into scheelite genesis and W mineralization in the western Gangdese belt.

2. Regional Geological Background

The Tibetan Plateau, a globally significant collisional orogenic belt, is subdivided from north to south into the Songpan-Ganzi, Qiangtang, Lhasa, and Himalayan terranes, separated by the Jinshajiang, Bangong-Nujiang, and Yarlung Zangbo suture zones [35] (Figure 1a). Among these, the Lhasa Terrane represents the most extensive, and structurally complex Mesozoic–Cenozoic tectono-magmatic belt on the plateau [36]. It is further divided into the North, Central, and South Lhasa terranes by the Shiquanhe–Namco Ophiolitic Mélange Zone (SNMZ) and the Luobadui–Milashan Fault (LMF) [37]. The region is characterized by widespread Gangdese Batholith intrusions and Linzizong volcanic rocks [38,39,40,41,42], with minor Oligocene–Miocene high-Sr/Y granitoid stocks [43,44,45,46]. The Gangdese porphyry Cu metallogenic belt extends east-west and hosts major porphyry deposits, including Zhunuo, Jiru, Chongjiang, Tinggong, Dabu, Qulong, and Jiama (Figure 1b). Magmatism and mineralization in the belt are predominantly dated between 20 and 13 Ma [47,48].

Yalongri, Zhigunong, and Xiongbaxi are located within the Zhunuo ore district in the western segment of the Gangdese Cu-polymetallic metallogenic belt, Tibet, and occur within volcanic–magmatic rocks of the South Lhasa Terrane (Figure 2). Magmatic rocks are widely developed in the district and consist mainly of Paleocene, Eocene, and Miocene granites. Paleocene granites dominate the western part of the district, and occur mainly as stocks. Eocene granites are concentrated in the southern part, extending east–west as batholiths and stocks, whereas Miocene granites are sporadically distributed as stocks or dikes. The Eocene and Miocene granitoids are closely associated with mineralization [49]. The principal stratigraphic units exposed in the district include the Eocene Nianbo (E₂n), Pana (E₂p), Dianzhong (E₁d), as well as the Angren (K_1-2a), Bima (K₁b), and Mamuxia (J₃-K₁m) formations. The Nianbo, Pana and Dianzhong formations of the Linzizong Group, are widely distributed in the central, eastern and northern parts of the district and consist mainly of intermediate to acidic volcanic lavas and pyroclastic rocks, including tuff, andesite, and rhyolite [50]. The Angren Formation, exposed in the southwestern corner, is composed primarily of coarse and fine-grained sandstone and siltstone. The Bima Formation, occurring in the northeastern area, consists mainly of tuff, mudstone, and siltstone, whereas the Mamuxia Formation exposed in the central-eastern area, is dominated by siltstone and limestone. Structural features are well developed and include NNE-, NNW-, and near E-W-trending faults. Among these, the NNE- and NNW-trending structures show the strongest association with mineralization [26].

3. Geological Characteristics of the Deposit

The Yalongri mining area lies in the southwestern part of the Zhunuo ore district, adjacent to the Luobuzhen deposit. The principal exposed unit is the Paleogene Dianzhong Formation of the Linzizong Group, distributed in the northeastern and northwestern parts of the area. This formation consists mainly of dacite affected by silicification, sericitization, and propylitic alteration, with minor pyrite occurring along fractures. Intrusive rocks dominate the area and consist mainly of biotite monzogranite emplaced as a batholith. This granite exhibits pervasive sericitic and propylitic alteration and forms the main ore-hosting body, with widespread malachite, azuritee, chalcopyrite, and molybdenite mineralization. Dikes are rare and represented by a single, ~75 m long lamprophyre dike in the western part of the area; this dike is fresh and unaltered. Faulting is weakly developed, with two principal sets trending near E-W and N-E. Based on regional structural patterns, NE-trending faults are interpreted as the primary ore-controlling structures in the Yalongri area.

Two major Cu-Mo mineralized bodies have been delineated: CuI and CuII (Figure 3). The CuI body, located in the northwestern part of the area, is ~1600 m long and ~800 m wide, covering an area of ~0.97 km² and exhibiting an irregular shape. The CuII body, situated in the central–southern area, is ~750 m long and ~300 m wide, covers ~0.18 km², and displays a comb-like geometry. Ore minerals include malachite, azurite, pyrite, chalcopyrite, molybdenite, and scheelite, occurring mainly in quartz veins and along fractures.

The Zhigunong mining area lies in the northwestern part of the Zhunuo ore district, adjacent to the Zhunuo and Beimulang porphyry Cu deposits. The principal exposed strata are the Paleocene–Eocene Pana and Nianbo formations of the Linzizong Group, which generally dip to the northwest. The Nianbo Formation, exposed mainly in the central–southern and northwestern parts of the area, consists of rhyolitic crystal tuff, rhyolitic lithic–crystal tuff, and purplish-red volcanic breccia, and is conformably overlain by the Pana Formation. The Pana Formation, exposed chiefly in the central–eastern and southern parts of the area, is composed primarily of dacitic crystal tuff and dacite. Magmatic rocks are limited and occur mainly as late-stage dikes, including granite porphyry, diorite porphyry, and lamprophyre. Structural features are relatively simple and dominated by NW- and NE-trending faults, which are closely associated with mineralization. In addition, fracture, systems (including joints), and tourmaline–quartz crypto-explosive breccia veins are well developed and strongly related to mineralization [34].

One Cu and one W mineralized body have been delineated in the Zhigunong mining area. The Cu body, located in the eastern part of the area, is ~570 m long and 200–220 m wide, occurs at depths of 73–132 m, strikes northwest, and shows a crescent-shaped plan geometry (Figure 4). The W body is spatially associated with the Cu body, extends for ~200 m, and is buried at a depth of ~210 m. The host rocks are mainly dacitic crystal tuff, with tungsten occurring predominantly as scheelite. Ore minerals consist mainly of chalcopyrite, pyrite, and scheelite, with minor chalcocite, occurring as disseminations, veinlets, and locally as clots or medium-to-fine-grained veins with a discontinuous distribution. Alteration is dominated by silicification and propylitization, with subordinate kaolinization and tourmalinization. Both mineralized bodies are concealed and form a typical “copper above, tungsten below” vertical zonation, indicating strong exploration potential [34].

The Xiongbaxi area, located in the western part of the Zhunuo ore district, is underlain mainly by the Nianbo Formation of the Linzizong Group, composed primarily of tuff. Magmatic rocks are well developed and include Cretaceous medium-to-coarse-grained porphyritic hornblende monzogranite, medium-to-coarse-grained porphyritic biotite monzogranite, and fine-grained porphyritic biotite monzogranite. The first two form extensively exposed batholiths, whereas the latter occurs as stocks or dikes restricted to the eastern and southeastern parts of the area. Ore minerals include molybdenite, malachite, pyrite, and scheelite. Alteration is dominated by limonitization and propylitization, with limonitization closely associated with mineralization.

One W-Mo mineralized body and two Cu mineralized bodies have been identified. The W-Mo body, located in the northern part of the study area, is ~700 m long and ~270 m wide, covering ~1.2 km² (Figure 5). It is hosted by biotite monzogranite and exhibits intense surface limonitization, with molybdenization commonly associated with limonitize, indicating a close genetic relationship. The two Cu bodies, located in the northern and central parts of the area, cover ~0.46 km² and ~0.21 km², respectively. They are also hosted by biotite monzogranite and show strong limonitization.

4. Sample Description and Analytical Methods

4.1. Sample Location and Selection

Samples were collected from the Yalongri, Zhigunong, and Xiongbaxi areas. Seven samples were selected under ultraviolet illumination and prepared as polished thin sections (~50 μm thick), including four from Yalongri, two from Zhigunong, and one from Xiongbaxi. In the Yalongri area, scheelite occurs mainly in biotite monzogranite, typically as disseminated grains along fractures. Grain sizes are mostly < 2 mm, with anhedral to subhedral granular morphologies. Scheelite is associated with pyrite, chalcopyrite, malachite, azurite, molybdenite, and quartz (Figure 6d). Under UV light, it displays pale blue to bluish-white fluorescence (Figure 6a–c). Under reflected light, scheelite displays surface cracks and locally contains inclusions of pyrite and biotite (Figure 7a). In the Zhigunong area, scheelite is hosted by crystal tuff and diorite porphyry, forming quartz + scheelite ± pyrite assemblages. In crystal stuff, scheelite is well crystallized, occurring as brownish tetragonal dipyramids with grain size of 0.5–5 mm (Figure 6e). In contrast, scheelite in diorite porphyry is mostly anhedral and distributed along vein margins (Figure 6f). Both types show pale blue fluorescence under UV light (Figure 6e,f). Under plane-polarized light, scheelite exhibits pleochroism from pale pink to pale green (Figure 7b), and under crossed polars, it shows second-order blue to blue–green interference colors (Figure 7c). Pyrite commonly occurs along scheelite margins in reflected light (Figure 7e,f). In the Xiongbaxi area, scheelite is hosted by medium-grained porphyritic hornblende monzogranite, with grain sizes around 1 mm. It displays bluish-white fluorescence under UV light (Figure 6g) and appears triangular with smooth surfaces under reflected light (Figure 7f). Scheelite is associated mainly with pyrite and molybdenite (Figure 6h,i).

Scheelite displays distinct CL characteristics among the three areas. In Yalongri, individual grains show three CL zones that, based on cross-cutting relationships, defined three generations from late to early: Sch I a (grayish-white), Sch I b (gray), and Sch I c (dark gray) (Figure 8a–c). In Zhigunong, scheelite grains contain dark gray and gray domains;the dark gray domains exhibit clear growth zoning and are locally cross-cut by gray domains (Figure 8d,e), allowing recognition of two generations: early Sch II a and late- Sch II b. In Xiongbaxi, scheelite shows uniform and homogeneous CL textures, indicating a single generation, designated Sch III (Figure 8f).

4.2. Sample Limitations

Sample collection in the Yalongri, Zhigunong, and Xiongbaxi areas was guided by spatial distribution, paragenetic relationships, and lithological variability. Samples from Yalongri and Zhigunong are considered representative of regional characteristics, whereas only one sample was collected from Xiongbaxi due to limited exposure, which may reduce its regional representativeness. Samples from the study area were excluded from model training and used solely as a test set to evaluate predictive performance, thereby preserving model objectivity. The prediction results serve only as reference; genetic interpretations rely primarily on regional geological features, with both factors jointly defining the genetic classification of mineralization in the area.

4.3. Analytical Methods

4.3.1. Major Element Analyses of Scheelite

Cathodoluminescence (CL) imaging of scheelite was conducted at the Geological Exploration Institute of the Shougang Group (Beijing) using a Thermo Scientific Apreo 2C scanning electron microscope operated at 10 kV acceleration voltage and 3.2 nA beam current.

Major element analyses were performed at the Electron Probe Microanalysis Laboratory, China University of Geosciences (Beijing) using an EMPA-1600 electron microprobe. Analytical conditions included an accelerating voltage of 15 kV, a beam current of 20 nA, and a beam diameter of 1 μm. The standards used were powellite (Mo), magnetite (Fe), scheelite (Ca, W), rhodochrosite (Mn), rutile (Ti), chalcopyrite (Cu), galena (Pb), and sphalerite (Zn). All analytical data were corrected using the ZAF method.

4.3.2. Trace Element Analyses of Scheelite

In situ trace element analyses of scheelite were conducted at the Chinese Academy of Geological Sciences using LA-ICP-MS. The system comprised a GeoLasPro 193 nm excimer laser ablation system (Coherent, USA) coupled to an Agilent 7900 ICP-MS. Analytical parameters included a 50 μm spot size, 5 Hz repetition rate, and ~5 J/cm² energy density. The ablated aerosol was transported by He as the carrier gas and mixed with Ar via a T-connector before entering the mass spectrometer. Each analysis consisted of a 20 s background acquisition followed by a 50 s signal acquisition. NIST SRM 610 and 612 were used as external standards. Data reduction followed the methods of Pearce et al. (1997) and Liu et al. (2008) [51,52], with offline processing performed using the ICPMSDataCal 12.2 software [52].

4.3.3. Machine Learning

Machine learning uses algorithms that learn from data to construct predictive models. Models are trained using labeled datasets and validated with independent data of known outcomes. Model performance and generalization ability are evaluated using standard metrics, after which well-performing models are applied to predict outcomes for previously unseen data [53].

Random Forest (RF) is an ensemble supervised learning algorithm, proposed by Breiman [54]. It uses bootstrap resampling to generate multiple training subsets, builds a decision tree for each subset, and combines their predictions through majority voting. This approach improves classification accuracy and generalization while providing robustness to noise and outliers [54].

Model performance was evaluated using precision, recall, accuracy, F1-score, area under the curve (AUC), and the confusion matrix (Figure 9). Precision measures the proportion of correctly predicted positive samples, recall measures the proportion of actual positives correctly identified, and accuracy represents the overall proportion of correct predictions. The F1-score, the harmonic mean of precision and recall, summarizes classification performance. AUC assesses binary classifier performance, with higher values indicating better discrimination. The confusion matrix compares predicted and true classes in an n × n matrix, where diagonal elements represent correct predictions.

All data in this experimental dataset were obtained through LA-ICP-MS analysis and measured using Ca as an internal standard. For the exclusion of inconsistent data, we followed two criteria: first, excluding data points lacking any of the 17 key elements (La-Lu, Y, Sr, Mo) required in the model; second, excluding data points where key discriminators (e.g., Mo, Sr) in the model fell below the detection limit.

5. Analytical Results

5.1. Major Elements of Scheelite

The major element compositions of scheelite determined by EMPA are summarized in Supplementary Materials Table S1. Sch I a, Sch I b, and Sch III display relatively high CaO contents, averaging 18.5 wt%, 18.9 wt%, and 19.0 wt%, respectively, whereas Sch I c, Sch II a, and Sch II b show lower CaO contents (Figure 10a), generally below the theoretical values of 19.5 wt%. The WO₃ contents of Sch I a, Sch I b, Sch I c, Sch II a and Sch II b are relatively high (averaging 80.9 wt%, 81.2 wt%, 80.9 wt%, and 81.8 wt%, respectively), exceeding the theoretical value of 80.53 wt%, whereas Sch I b, and Sch III have lower WO₃ contents, averaging 79.8 wt% and 78.6 wt%, respectively (Figure 10b). The MoO₃ contents of Sch I a, Sch I b, Sch I c, and Sch IIa are similar and relatively low, whereas Sch II a, Sch II b, and Sch III show higher values (Figure 10c), with Sch III exhibiting the highest average MoO₃ content (1.25 wt%).

5.2. Trace Elements of Scheelite

The LA-ICP-MS trace-element and REE data for scheelite are presented in Supplementary Materials Table S2. Scheelite from the Zhunuo ore district shows moderate compositional variations in Sr, Y, Nb, Th, U, and REEs among different generations.

As shown in Figure 10, Sch I a and Sch I b display similar contents of V, As, Y, and Nb (Figure 10d–h), whereas Sch I b is relatively enriched in Hf, Ta, Th, and U (averages: 0.2 ppm, 1.2 ppm, 11.9 ppm, and 1.7 ppm, respectively; Figure 10i–l). In comparison, Sch I c shows higher V, As, and Nb contents (averages: 21.4 ppm, 49.3 ppm, and 34 ppm, respectively), while its Sr, Y, Hf, Ta, Th, and U contents are similar to those of Sch I a (Figure 10d–l). Sch II a and Sch II b have comparable V, Nb, Hf, Th, and U contents; however, Sch II b exhibits lower As and Y (averages: 4.9 ppm and 3.5 ppm) and higher Sr (average: 942 ppm) (Figure 10d–i). Sch III shows similarly low Hf, Th, and U contents to Sch II a- Sch II, but higher V, As, Nb, and Ta contents (Figure 10h–l). Sch I a, Sch I b, Sch I c, and Sch II a display comparable LREE, MREE, and HREE contents (Figure 10m–o). Sch I a and Sch I b exhibit wide variation in LREE and ΣREE (LREE = 216–1909 ppm and 390–2192 ppm; ΣREE = 400–2550 ppm and 539–2640 ppm, respectively). In contrast, Sch II b and Sch III have much lower MREE and HREE contents (MREE = 1.4 ppm and 56.2 ppm; HREE = 1.7 ppm and 73.4 ppm, respectively), whereas Sch III shows relatively high LREE (average: 1261 ppm) and ΣREE contents (average: 1335 ppm).

REE patterns further discriminate scheelite generations (Figure 11). Sch I a-Sch I c show relatively flat patterns, differing mainly in Eu anomalies: Sch I a exhibits weak or no Eu anomaly, Sch I b a strong positive Eu anomaly, and Sch I c a pronounced Eu negative anomaly (Figure 11a). Sch II a displays a hump-shaped pattern with MREE enrichment and a weak positive Eu anomaly, whereas Sch II b shows a right-sloping pattern with LREE enrichment, HREE depletion, a strong positive Eu anomaly, and the lowest ΣREE (average: 19.6 ppm) (Figure 11b). Sch III exhibits a relatively uniform right-sloping pattern with LREE enrichment, HREE depletion, a strong negative Eu anomaly, and weak LREE-HREE fractionation with (La/Lu)_N = 35.9–37.8 (Figure 11c).

5.3. Random Forests of Trace Elements

This study compiled 3347 LA-ICP-MS analyses of scheelite from 90 tungsten deposits worldwide (Supplementary Materials Table S3), covering 17 elements (La, Ce, Pr, Nd, Sm, Eu, Gd, Tb, Dy, Ho, Er, Tm, Yb, Lu, Y, Sr, and Mo) across six deposit types. The dataset was analyzed using two-dimensional geochemical plots and machine learning methods to construct a predictive classification model. The dataset was sourced from [57,58,59,60,61,62,63,64,65,66].

REE patterns from the dataset show that, except for vein-type gold deposits, which display a left- sloping pattern characterized by HREE enrichment and LREE depletion, most deposit types exhibit overlapping REE distributions and cannot be reliably distinguished using REE patterns alone (Figure S1). In contrast, binary geochemical plots reveal that Sr content, Mo content, and the Sr/Mo ratios provide effective discrimination among deposit types. In Sr-Mo, Mo-(Gd/Lu)_N, and Sr/Mo-δEu diagrams (Figure S2a–c), orogenic deposits are characterized by high Sr, low Mo, high Sr/Mo ratios, and positive Eu anomalies, whereas vein-type gold deposits show high Mo contents and low Sr/Mo ratios. Porphyry and greisen deposits display variable Mo contents and Sr/Mo ratios, although greisen deposits generally have higher Sr contents. Skarn deposits are enriched in both Sr and Mo but exhibit low Sr/Mo ratios, while hydrothermal vein-type deposits are characterized by distinctly low Sr and Mo contents. Therefore, Sr and Mo contents and the Sr/Mo ratio effectively discriminate among different deposit types. In the δCe-δEu diagram (Figure S2d), orogenic, vein-type gold, porphyry, and quartz vein-type deposits commonly exhibit positive Eu anomalies. In contrast, greisen and skarn deposits display a wider range of Eu anomaly values but are predominantly positive, resulting in weaker discrimination in this diagram.

The ROC curves for the test dataset yield AUC values close to 1, indicating excellent classification performance (Figure S3a). Optimization of the Random Forest (RF) model produced a feature-importance ranking (Figure S3b), with Sr, Mo, and δEu identified as the most influential variables (importance > 2.50). Eu, La, δCe, and Ce show intermediate importance (1.71–2.22), whereas the remaining elements contribute less, particularly Ho, Tm, and Er, which rank lowest (1.01–1.09). The confusion matrix, based on 3347 data points, demonstrates high prediction accuracy and low error rates for four deposit types; reduced performance for vein-type gold and hydrothermal vein-type deposits reflects limited sample sizes (Figure S3c). Overall, these results confirm the robustness and reliability of the RF classifier. The learning curve stabilizes at ~174 decision trees (Figure S3d), indicating model convergence and demonstrating that the RF model is an efficient, interpretable, and practical tool for cost-effective geochemical exploration.

The Random Forest model was applied to scheelite trace-element data from the Zhunuo ore district to predict deposit types in the Yalongri, Zhigunong, and Xiongbaxi areas (Supplementary Materials Table S4). Of the 16 samples from Yalongri, 14 were classified as porphyry, 1 as orogenic, and 1as greisen. The 8 samples from Zhigunong were assigned to four deposit types: 3 porphyry, 2 skarn, 1 orogenic, and 2 greisen. Both samples from Xiongbaxi were classified as skarn-type deposits (Figure S3e). Figures S1–S3 are provided in the Supplementary Materials.

6. Discussion

6.1. Substitution Mechanisms in Scheelite

Previous studies indicate that REE³⁺ enter the scheelite lattice mainly through three substitution mechanisms [9,14,15]: (1) 2Ca²⁺ = REE³⁺ + Na⁺; (2) Ca²⁺ + W⁶⁺ = REE³⁺ + Nb⁵⁺; and (3) 3Ca²⁺ = 2REE³⁺ + □Ca, where □Ca denotes a Ca vacancy. If mechanism (1) dominates, Na contents should be comparable to ΣREE + Y-Eu, and preferential incorporation of MREE³⁺ (due to ionic-radius similarity with Ca²⁺), should produce hump-shaped REE pattern. Dominance of mechanism (2) would result in strong Nb enrichment, with Nb = ΣREE + Y-Eu, as reported in some gold deposits (e.g., Nova Scotia, Canada) [11]. Both mechanisms (1) and (2) are controlled by ionic radius and thus predict an approximately 1:1 relationship between ΣREE + Y-Eu and Na or Nb [9]. In contrast, mechanism (3) is independent of ionic radius, shows no REE selectivity, and is primarily governed by the REE composition of the ore-forming fluid, typically producing flat REE patterns [9]. Scheelite from the Zhunuo ore district, except for Sch II b, exhibits high ΣREE + Y-Eu contents (698–3857 ppm), far exceeds Na (22–330 ppm) or Nb (4.5–39.2 ppm), effectively ruling out mechanisms (1) and (2). For Sch II b, Na is below detection limits and Nb (3.9–4.3 ppm) remains much lower than ΣREE + Y-Eu (16.1–28.4 ppm), further excluding these mechanisms. In addition, ΣREE + Y-Eu vs. Na and Nb plots show no 1:1 linear relationship (Figure 12a,b), confirming that substitution is not controlled by ionic radius. The REE distribution patterns of scheelite further support this interpretation [9]. Scheelite from the district displays relatively flat REE patterns with weak LREE-HREE fractionation (La_N/Lu_N = 0.22–75, average = 7.64), indicating that the dominant substitution mechanism is 3Ca²⁺ = 2REE³⁺ + □Ca.

6.2. Physicochemical Conditions of the Hydrothermal Fluids

Numerous studies have shown that the redox conditions of scheelite-forming fluids can be constrained using indicators such as Eu and Ce valence states and Mo contents [9,20,65,67].

Europium occurs as Eu²⁺ and Eu³⁺. Under reducing conditions, Eu is preferentially enriched as Eu²⁺, which readily substitutes for Ca²⁺ in the scheelite lattice due to their similar charge and coordination number, producing a positive Eu anomaly. Under oxidizing conditions, Eu is dominated by Eu³⁺, typically resulting in a negative Eu anomaly. The average δEu values of Sch I a (1.21), Sch I b (5.86), and Sch I c (0.45) indicate weakly reducing, strongly reducing, and moderately oxidizing conditions, respectively. Similarly, Sch II a (1.49) and Sch II b (3.54) reflect reducing fluids, whereas Sch III (0.33) shows a pronounced negative Eu anomaly, indicative of oxidizing conditions. Because Eu³⁺, Sm³⁺, and Gd³⁺ have similar partition coefficients, a near 1:1 correlation between Eu_N and Eu_N* (Eu_N* = $\sqrt{{Sm}_N\ast {Gd}_N}$) suggests Eu³⁺ dominance [9]. Such correlations are observed in Sch I a, Sch I c, and Sch II a (Figure 13a). While this is consistent with oxidizing conditions for Sch I c, it conflicts with the reducing environments inferred for Sch I a and Sch II a, indicating that their Eu anomalies were influenced by factors other than in oxygen fugacity. In addition to redox conditions, the Eu anomalies can be affected by plagioclase fractionation, fluid-rock interaction, and temperature [9,15,68,69,70]. From Sch I c to Sch I b, the Eu anomaly changes markedly, yet shows no clear correlation between with Sr content (Figure 13d), excluding a dominant role for fluid-rock interaction. Temperature effects on Eu anomalies in scheelite are generally limited and insufficient to produce large variations [71]. Consequently, the observed Eu anomaly variations are most likely related to plagioclase fractionation, consistent with the lithology of the study area. Scheelite crystallized during the early stages did not undergo significant fluid-rock interaction and thus preserves the characteristics of the parental magma (Figure 13e), which is inferred to have been REE-enriched and oxidized. In contrast, Sch I b, Sch I c exhibits higher Sr contents and lower Eu positive anomalies, indicating that fluid-rock interaction became important during the evolution of the hydrothermal system (Figure 13f). CL textures and trace-element characteristics of scheelite further indicate that late-stage hydrothermal fluids partially dissolved earlier-formed scheelite, allowing later- generations to inherit geochemical features of the early- and middle-stage scheelite. Previous studies suggest that if Eu anomalies are primarily controlled by redox conditions, δEu should correlate positively with Mo content and δCe [65]. However, scheelite from the Zhunuo ore district shows scattered distributions in Mo-δEu and δCe-δEu plots (Figure 13b,c), indicating that oxygen fugacity was not the sole control on Eu anomalies. Cerium provides additional redox constraints because it occurs as Ce⁴⁺ under oxidizing conditions and Ce³⁺ under reducing conditions [23,72]. The observed δCe values (0.97–1.55) are generally close to or greater than unity, indicating an overall reducing fluid environment, consistent with the Eu-based interpretations. Together, these observations imply that Eu anomalies were controlled by multiple factors, including fluid redox conditions, fluid-rock interactions, and/or external fluid input. Molybdenum behavior further constrains redox conditions.

In hydrothermal systems, Mo valence is redox-sensitive: under oxidizing conditions, Mo⁶⁺ substitutes for W⁶⁺ in scheelite, forming a solid solution toward powellite (CaMoO₄) [73], whereas under reducing conditions, Mo⁴⁺ combines with sulfur to form molybdenite (MoS₂) [74,75]. Consequently, Mo content in scheelite is a sensitive redox indicator [20,76,77]. In the Zhunuo district, average Mo contents of Sch I a (407 ppm), Sch I b (442 ppm), and Sch I c (124 ppm) suggest an increase in oxygen fugacity (fO₂) from early to middle stages, followed by a decreased during the late stage. In contrast, Sch II a (529 ppm) and Sch II b (1762 ppm) record a progressive increase in oxygen fugacity. Sch III, characterized by high Mo contents (8898–8964 ppm), indicates crystallization from a highly oxidizing fluid, consistent with its strong negative Eu anomaly.

6.3. Constraints on Fluid Evolution from Scheelite

As discussed above, the variable δEu values of Sch I a, Sch I b, and Sch I c indicate that scheelite-forming fluids in the Yalongri area evolved progressively from oxidizing to reducing conditions from early to late stages. REE patterns and trace element characteristics suggest that the early fluid was REE-rich, had a relatively high Eu³⁺/Eu²⁺ ratio, and contained low Mo. During early crystallization, Eu entered scheelite lattice mainly as Eu³⁺, producing a gradual decrease in the Eu³⁺/Eu²⁺ ratio and resulting in the flat REE pattern with a negative Eu anomaly observed in Sch I c. The relatively high Mo contents of Sch I b indicates the influx of Mo-rich fluids after Sch I c crystallization, with Mo substituting for W⁶⁺ in the scheelite lattice. As Eu³⁺ was largely consumed during Sch I c formation, Eu²⁺ became dominant, shifting the fluid toward reducing conditions. Consequently, Eu entered scheelite mainly as Eu²⁺ substituting for Ca²⁺, consistent with the high δEu values (2.99–13.65) observed in Sch I b. In the δEu-Sr diagram (Figure 13d), Sr remains nearly constant with increasing δEu, indicating negligible dissolution of Eu-rich minerals (e.g., plagioclase). Late-stage Sch I a shows ΣREE, Th, U, and Hf contents similar to Sch I c but a wide range of Mo contents (52.6–1657 ppm), suggesting formation through dissolution-precipitation of Sch I b and Sch I c by late-stage fluids. CL images (Figure 8a) show Sch I a as fine veins crosscutting or rimming earlier scheelite (Sch I b and Sch I c), supporting this interpretation. The consumption of Eu²⁺ during Sch I b crystallization likely resulted in the weakly positive Eu anomaly observed in Sch I a. In the Zhigunong area, contrasting REE patterns (Figure 11b), ΣREE contents (Sch II a: 693–1987 ppm; Sch II b: 11.5–20.7 ppm), Mo contents (Sch II a: 71–1135 ppm; Sch II b: 1690–1833 ppm), and CL textures indicate two distinct fluid stages. The early-stage was REE-rich, Mo-poor, and weakly reducing, whereas the late-stage was REE-poor, Mo-rich, and more oxidizing, characterized by a pronounced positive Eu anomaly. From early to late stages, increasing oxygen fugacity was accompanied by rising δEu and Sr contents (Figure 14b,d), likely due to plagioclase dissolution during fluid migration. In the Xiongbaxi area, uniform REE patterns and very high Mo contents of Sch III (8898–8964 ppm) indicate crystallization from a single-stage oxidizing fluid, enriched in LREEs, depleted in HREEs, and characterized by high Mo availability.

The Sr/Mo ratio has been proposed as an effective indicator of scheelite formation environments [67]. In magmatic–hydrothermal systems, scheelite derived from highly fractionated felsic magmatic typically exhibits low Sr/Mo ratios, whereas metamorphic systems produce higher ratios due to Sr released during metamorphism [67]. Scheelite from the Zhunuo ore district shows consistently low Sr/Mo ratios (0.02–6.10). In the Sr/Mo-δEu diagram (Figure 14a), all data plot within the magmatic-hydrothermal field, confirming this genetic affinity. Because Y and Ho have similar ionic radii and charges, their ratios remain stable under closed-system crystallization but can be modified by external fluid input [11,78]. Except for Sch II b (Y/Ho = 36–59), scheelite from the Zhunuo district displays relatively constant Y/Ho ratios (12–38), indicating crystallization under stable conditions (Figure 14b). The distinct Y/Ho values in Zhigunong scheelite further support the presence of two discrete fluid stages in this area.

In summary, scheelite from the Zhunuo ore district is of magmatic-hydrothermal origin and crystallized under a relatively stable conditions. In Yalongri, ore-forming fluids evolved from oxidizing to reducing conditions from early to late stages, with oxygen fugacity initially increasing and then decreasing, without significant dissolution-precipitation of Eu-rich minerals. In Zhigunong, scheelite formed from two distinct fluid stages: an early REE-rich, Mo-poor stage and a late REE-depleted, Mo-rich stage, accompanied by increasing oxygen fugacity and plagioclase dissolution. In Xiongbaxi, scheelite crystallized from a single oxidizing fluid enriched in LREEs, depleted in HREEs, and characterized by high Mo contents.

6.4. Fingerprinting Deposit Types

Previous studies have demonstrated that scheelite trace element compositions are effective for distinguishing deposit types and guiding mineral exploration [3,6,79]. Results from the Random Forest (RF) model indicate that deposit types in the Yalongri and Xiongbaxi areas are relatively uniform, whereas those in Zhigunong are more complex, likely reflecting multiple stages of ore-forming fluids.

In the Yalongri area, scheelite is predominantly classified as porphyry type, characterized by variable Mo contents, Sr/Mo ratios, and relatively high Sr contents, comparable to scheelite from the Nuri Cu-W-Mo and Julong Cu-Mo deposits [19,33]. Minor scheelite classified as orogenic type shows low Mo (50 ppm) and Sr (254 ppm), relatively high Sr/Mo (5.40), and weak MREE enrichment, resembling scheelite from the Huangjindong orogenic gold deposit [7]. Greisen-type scheelite displays elevated Y and Sr contents (890 ppm and 320 ppm) and high Sr/Mo ratios (6.10), similar to the Chuankou W deposit in the Nanling belt [80]. The occurrence of greisen- and orogenic-type signatures in Yalongri likely due to strong late-stage fluid-rock interactions affecting Sr behavior. However, based on regional geological and mineralization style, Yalongri is best classified as a porphyry type system. Scheelite from the Zhigunong area show greater compositional diversity, reflecting multiple fluid sources. Porphyry-type scheelite exhibits moderate Sr (550–650 ppm), high Mo (830–1135 ppm), and low Sr/Mo (0.54–0.78), comparable to the Nuri Cu-W-Mo deposit [33]. Orogenic-type scheelite contains high Sr and Mo (1031 ppm and 1690 ppm) with low Sr/Mo (0.61), similar to the Doranasai deposit [81]. Greisen-type scheelite shows wide ranges in Sr (244–854 ppm), Y (4.5–1045 ppm), and Mo (72.7–1833 ppm), with Sr/Mo ratios of 0.47–3.36, resembling the Chuankou W deposit [80]. Skarn-type scheelite is characterized by very low ΣREE (11–21 ppm), high Mo (1691–1833 ppm), and high Sr (854–1031 ppm), comparable to scheelite from the Brejui deposit [82]. The complexity of scheelite geochemistry in Zhigunong likely results from strong in Sr and Mo caused by the involvement of multiple fluid pulses. Its proximity to Zhunuo porphyry deposit further suggests a strong potential for porphyry type mineralization. In the Xiongbaxi area, scheelite exhibits relatively high ΣREE (average: 1335 ppm), a pronounced negative Eu anomaly, and right-sloping REE patterns, consistent with scheelite from the super-large Sandaozhuang W-Mo skarn deposit on the southern margin of the North China Craton [13]. This interpretation is consistent with the widespread occurrence of molybdenite and intense limonitization observed in the area (Figure 6h,i). In summary, the study area records strong fluid–rock interaction, consistent with the view that intense hydrothermal alteration is a key control on mineralization, as documented in the Yangla deposit [83]. However, due to the relatively limited exploration coverage, the scale and economic potential of mineralization in the study area require further investigation.

7. Conclusions

(1)

Based on REE distribution patterns and cathodoluminescence (CL) imaging, scheelite in the Zhunuo ore district can be divided into multiple generations. Three generations are recognized in the Yalongri area, from late to early: Sch I a, Sch I b, and Sch I c. Two generations occur in the Zhigunong area, from early to late: Sch II a and Sch II b. Scheelite from the Xiongbaxi area comprises a single generation: Sch III.

(2)

Scheelite from the Zhunuo district exhibits Na and Nb contents significantly lower than ΣREE + Y-Eu and relatively flat REE patterns, indicating that the dominant substitution mechanism is 3Ca²⁺ = 2REE³⁺ + □Ca. The δEu values of Sch I a, Sch I b, Sch II a, and Sch II b are mostly ≥1, reflecting reducing ore-forming fluids, whereas Sch I c and Sch III show δEu < 1, indicating oxidizing conditions.

(3)

REE patterns and trace element characteristics indicate that ore-forming fluids in the Yalongri area evolved from oxidizing to reducing conditions, with late-stage scheelite forming through dissolution–reprecipitation processes. The Zhigunong area records two fluid stages: an early REE-rich, Mo-poor stage and a later REE-poor, Mo-rich stage. The Xiongbaxi area records a single oxidizing stage characterized by REE- and Mo-rich fluids. Scheelite from the district shows low to moderate Sr/Mo ratios (0.02–6.10), consistent with a magmatic-hydrothermal origin, and relatively uniform Y/Ho ratios (12–59), indicating a stable crystallization environment.

(4)

Random Forest model predictions classify scheelite from the Zhunuo ore district into four genetic types. Scheelite from the Yalongri area is predominantly porphyry type, with minor greisen and orogenic affinities. In the Zhigunong area, scheelite spans porphyry, skarn, greisen, and orogenic types, whereas scheelite from the Xiongbaxi area is classified as skarn type.