2.1. Regional geological characteristics

The Songshugang mining district occupies Geyuan Town, Hengfeng County, Jiangxi Province. Positioned approximately 3 km northwest of the Lingshan Pluton contact zone, it spans coordinates 117°37’28”E-117°38’58”E and 28°37’15”N-28°38’30”N. Tectonically, the mining district occupies a critical position within the suture zone between the Yangtze Block and Cathaysia Block; metallogenically, it lies within the Huaiyu Mountains Metallogenic Subzone of the Qinzhou-Hangzhou Metallogenic Belt’s eastern segment [29].

The region primarily exposes strata of the Jixian, Qingbaikou, Nanhua, Sinian, Cambrian, and Jurassic Systems (Fig 1). The Jixian System is composed predominantly of phyllites and schists; the Qingbaikouan System comprises phyllites, rhyolites, and metaclastic lithic conglomerates; the Nanhua System comprises phyllites, slates, pebbly greywackes, tuffites, and carbonaceous shales; the Sinian System comprises siltstones and fine-grained sandstones; the Cambrian System consists of calcareous mudstones and argillaceous limestones; the Jurassic System includes mudstones, conglomeratic sandstones, and siltstones. Magmatic rocks are dominated by the Lingshan composite batholith, exhibiting distinct lithofacies zoning. Three concentric zones are identified from core to periphery: Fine-grained glomeroporphyritic rapakivi amphibole-biotite monzogranite, Medium-grained rapakivi amphibole-biotite monzogranite, Medium-coarse-grained biotite alkali-feldspar granite [30].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Regional geological map (modified from [32]).

1. Jurassic System; 2. Cambrian System; 3. Sinian System; 4. Nanhua System; 5. Qingbaikouan System; 6. Jixian System; 7. Yangqiao Unit; 8. Xiazhaiwu Unit; 9. Lingshan Superunit; 10. Normal fault, dip angle; 11. Reverse fault, dip angle; 12. Strike-slip fault; 13. Fault (nature undefined); 14. Circular structure (RS-interpreted); 15. Unconformity contact; 16. Lithofacies boundary; 17. Geological contact; 18. Tantalum-niobium deposit.

https://doi.org/10.1371/journal.pone.0331937.g001

The area exhibits well-developed NE-SW trending linear structures, with subordinate NW-SE orientations, with subordinate NW-SE trends. Sub-EW trending basin-controlling faults occur along the southern margin of the Jurassic basin. Under the influence of the Lingshan batholith and deep-seated concealed intrusions, three distinct circular structures-Mopanshan, Songshugang, and Lingshan-are manifested on remote sensing imagery. Proven deposits include the Songshugang Ta-Nb and Huangshan Nb-Ta deposits. Ore bodies are predominantly occur within altered granitic intrusions, comprising potassic-altered granite, sodic-altered granite, and greisenized granite types. Surface country rocks exhibit intense alteration assemblages featuring potassic alteration, albitization, greisenization, hematitization, sericitization, and chloritization [31].

Reprinted from [32] under a CC BY license, with permission from [Geocarto International], original copyright [2025].

2.2. Geological setting of the mining district

The Songshugang mining district constitutes a concealed, altered granite-hosted rare-metals deposit. Tantalum (Ta) and niobium (Nb) represent the primary economic metals, with associated mineralization including tungsten (W), tin (Sn), rubidium (Rb), and lithium (Li). This deposit exhibits distinct metallogenic characteristics, with ore bodies occurring within alteration shells atop concealed granite cupolas. This mineralization pattern features spatially coexistent and genetically associated polymetallic assemblages. Surface exposures comprise the Nanhua System, dominated by low-grade metamorphic phyllites and slates. The deep-seated concealed rock bodies associated with mineralization are divided into six alteration zones: pegmatite-type niobium-tantalum mineralization zone, greisenized granite-type niobium-tantalum mineralization zone, potassic feldsparized granite-type niobium-tantalum mineralization zone, weakly albitized granite-type subzone, moderately albitized granite-type subzone, and strongly albitized granite-type subzone (Fig 2) [33]. The mineralogical characteristics of each alteration zone are presented in Table 1. The wall-rock alteration shows a vertical zonation. From the bottom up, there are a sodic alteration zone, a greisen zone, and a potassic alteration zone, which have a close spatial and genetic relationship with mineralization. In terms of structure, the mining district is on the northern limb of the Geyuan-Zhengfang composite syncline, with many secondary folds. The fault structures in the district are mainly small-scale NE and NW-trending faults [31].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Mineral Characteristics of Rock Alteration Zones.

https://doi.org/10.1371/journal.pone.0331937.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Section along exploration line 0 in the songshugang mining area, Hengfeng County, Jiangxi Province.

1. Nanhua System; 2. Pegmatite-type Nb-Ta mineralized zone; 3. Greisenized granite-type Nb-Ta mineralized zone; 4. K-feldspathized granite-type Nb-Ta mineralized zone; 5. Weak albitized granite subzone; 6. Moderate albitized granite subzone; 7. Strong albitized granite subzone; 8. Boreholes.

https://doi.org/10.1371/journal.pone.0331937.g002

3.1. FastRBF introduction

Radial Basis Function (RBF) has become one of the most widely used functions in implicit 3D geological modeling due to its mathematical flexibility, accurate characterization of complex spatial relationships, and strong adaptability to unstructured geological data. In contrast to traditional interpolation methods like Kriging and Discrete Smooth Interpolation (DSI), RBF transforms spatial scatter points into a continuous implicit field by employing kernel functions. This capability not only facilitates the effective representation of nonlinear geometric features inherent in complex geological interfaces but also accommodates sparse and heterogeneously distributed geological data with remarkable efficiency.

RBF modeling generates a continuous, implicit surface from scattered data points. The model uses inputs like point coordinates, normal vectors, and other constraints to calculate the weighting coefficients for the basis functions. By incorporating global polynomial terms to correct low-frequency errors, this method produces a smooth, accurate implicit surface equation [20,34,35,36,37]. Obtain the zero isosurface F(r)=0, constitutes the surface of the geological body model, with the implicit surface equation constructed from scattered data points [38]:

(1)

In the formula: r is any data point on the generated surface; q_i are the control points (known sample points, i.e., surface points); w_i are the weights corresponding to each control point; is the radial basis function; and P(r) is a first-order polynomial in r.

(2)

The kernel function type of the FastRBF function is the improved spheroidal basis function, and the specific formula is:

(3)

In the formula: r is the distance between the calculation point and the center of the basis function; m is the order of the ellipsoidal function (an odd number between 3 and 9), which controls the rate of decay of the function, with smaller m values resulting in slower decay and smoother functions, and larger m values resulting in faster decay and more localized functions; c is the parameter that controls the “support radius” of the kernel function; this formula is a decreasing function that asymptotically approaches zero.

In Leapfrog software, the basis functions are represented in the form of spherical variograms:

(4)

In the formula: is the semivariance value; S is the sill value, which in this study is set to the variance of the data samples; N is the nugget effect value, representing small-scale noise or measurement error, and in this study is set to 5% of the variance; R is the range parameter of the semivariance; is a proportionality constant dependent on the order m of the spherical function.

In the Leapfrog software, the support radius c in this study is set to 1.5 times the basic survey spacing; the regularization parameter λ is set to 1 × 10⁻⁶ to balance the fitting accuracy and the stability of the solution; the hierarchical partitioning depth is automatically optimized based on the data volume; the truncation threshold is set to 1 × 10⁻⁴ to ensure the computational efficiency of the far-field fast multipole approximation. Under this configuration, the complexity of a single matrix-vector multiplication is O(N), and if the number of iterations is k, the total complexity of the construction phase is:

(5)

k is typically a constant, so the complexity of constructing the algorithm simplifies to O(NlogN).

To determine the weight coefficients w_i and the first-order polynomial coefficients a_i, the orthogonal condition (6) and the interpolation constraint condition (7) must be satisfied.

(6)

(7)

where: j = 1, 2, …, n.

Let , and from equations (6) and (7), we can obtain:

(8)

Equation (8) can be simplified as follows:

(9)

In the formula, , , , , (10)

Under typical conditions, the left-hand side of equation (9) is an invertible non-positive semi-definite matrix, and thus there is only one set of solutions (w₁, w₂, …, w_n, a₀, a₁, a₂, a₃,). The polynomial coefficients obtained by solving equation (9) can be used to determine P(r), and this should not vary with different geological units.

FastRBF represents an improved version of RBF that focuses on algorithmic acceleration and computational efficiency optimization, delivering enhanced performance and accuracy when processing large-scale datasets [39–40]. FastRBF employs hierarchical spatial partitioning by decomposing the computational domain into multi-level subspaces and implementing approximation schemes for distant regions. This reduces computational complexity from O(N²) in conventional RBF to O(NlogN), enabling efficient modeling with millions of scattered points [41]. FastRBF addresses the dense matrix generated by global RBF interpolation through localized support kernel functions or truncation strategies, combined with Krylov subspace iterative solvers (e.g., GMRES), significantly reducing memory footprint and computational load. Compared to RBF, FastRBF achieves a 10-100x speedup in interpolation [42]; On the other hand, FastRBF employs multi-scale kernel functions (e.g., hybrid compactly supported Gaussian and global linear kernels) with data-driven bandwidth optimization to balance local detail resolution and global trend fitting. Adaptive adjustment of regularization parameters effectively mitigates interference from data noise and outliers [41].

3.2. Modeling workflow

The complex 3D geology of the study area makes it difficult to create accurate implicit models directly from borehole and exploration data. To address this challenge, our study introduces an optimized modeling workflow that integrates geological expertise as an interpretive constraint. Using the Songshugang Mining Area as a case study, we based this expertise on a comprehensive analysis of the region’s geological patterns and structural evolution. This knowledge-driven approach significantly enhances modeling accuracy and reliability, establishing a refined 3D geological modeling workflow (Fig 3). The implemented workflow comprises four critical phases: 1) Data collection, 2) Data processing, 3) Model construction, and 4) Model analysis with prospective mineralization targets prediction, detailed as follows:

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. 3D Geological modeling workflow.

https://doi.org/10.1371/journal.pone.0331937.g003

1. (1). Data Collection. Prior to modeling, systematically collect the required input datasets, including surface boundaries, topographic contours, geological maps, borehole data, and exploration sections.
2. (2). Data Processing. This phase comprises format conversion, projection transformation, database integration, project boundary delineation, and georeferenced digitization of geological contacts.
3. (3). 3D Model Construction. Key procedures include expert knowledge constraint implementation, geological interface reconstruction, 3D structural framework modeling, and grade distribution modeling.
4. (4). Model Analysis and Mineral Prospectivity Mapping. The 3D structural and grade models enable comprehensive analysis of deep-seated geological characteristics and orebody distribution patterns, facilitating prediction of prospective mineralization targets at depth.

4.1. Modeling software introduction

The 3D geological modeling for this study was performed using Leapfrog Geo 2023.1, an industry-standard platform for implicit modeling. Its core strength is the FastRBF algorithm, which enables the rapid interpolation of diverse datasets to build high-precision models. These models effectively visualize complex geological features, including structural frameworks and orebody distribution. In addition to its primary algorithm, Leapfrog Geo enhances modeling efficiency through robust data processing, an intuitive interface, and powerful real-time updating and interactive editing functions. Its capabilities have led to its widespread adoption in fields such as mineral exploration, engineering geology, and hydrogeology.

4.2. Modeling data introduction

(1) Surface Boundaries. The modeling area extends 1.90 km (E-W) and 0.93 km (N–S), covering approximately 1.77 km². These boundaries define the lateral extents of the 3D model.

1. (2). Topographic Contours. Contour data from topographic maps serve as primary inputs for constructing the Digital Elevation Model (DEM). The generated DEM surface accurately represents terrain morphology, forming the upper boundary surface of the geological model.
2. (3). Geological Map. A 1:2,000-scale geological map incorporating 14 exploration sections and 57 borehole locations provides spatial constraints on geological unit boundaries in the 3D geological model, defining the volumetric extent of geological bodies.
3. (4). Borehole Data. A total of 57 boreholes were integrated, with depths ranging from 326.74 m to 1,000.25 m (mean 574.89 m). Boreholea are concentrated in the central and western sectors, with sparse coverage in the east. Datasets include collar coordinates, deviation surveys, stratigraphic logging, and assay analyses. Stratigraphic data supports 3D structural modeling, while assay data enables grade distribution modeling. Integration borehole data with exploration sections significantly enhances modeling accuracy [18,43–45].
4. (5). Exploration Sections. Fourteen exploration sections with NW-SE orientation were incorporated. Partial sections do not fully traverse the mining area. The central and western parts of the mining area have complex geological structures, with densely arranged exploration lines and profiles, while the eastern part has relatively simple geological structures and a sparser arrangement of exploration lines, resulting in a data distribution pattern characterized by “dense in the west and sparse in the east” (Fig 4). Accordingly, a differentiated modeling strategy was applied: virtual profiles were added in the structurally complex western area to better constrain 3D geometry, while no additional virtual profiles were required in the simpler eastern area.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Boreholes and exploration section location map. 1.

Exploration lines and labels; 2. Boreholes; 3. Nanhua System.

https://doi.org/10.1371/journal.pone.0331937.g004

4.3. Modeling data processing

(1) Projection Transformation. Surface boundaries, geological maps, exploration sections, and borehole collar coordinates underwent projection transformation to a unified 1:1,000 scale (for display purposes, not representing data precision) using CGCS 2000 ellipsoid parameters with Gauss-Krüger projection (3° zone 39).

1. (2). Data Format Conversion. Surface boundaries and topographic contours were converted to  .dxf format; raster geological maps and exploration sections to  .JPG format; borehole data were structured into relational tables: Collar, Survey, Interval, and Assay, linked by borehole IDs.
2. (3). Data Import. Import  .dxf format data directly into modeling software; spatially eoreferenced geological maps and exploration section maps in  .JPG format using the 3D coordinates (x, y, z) of four corner points; sequentially import drillhole collar tables, survey tables, interval tables, and assay tables to generate 3D drillhole trajectories. Perform error verification on imported data through multi-angle visualization inspection and software-based logical validation. Identified errors may be corrected either via direct edits or through source data amendments followed by reimportation.
3. (4). Model Domain Establishment. The model roof is defined by a DEM surface (peaking at 692.98 m), while the base constitutes a horizontal plane at −700 m elevation. Lateral boundaries are formed by vertical planes extending from the surface modeling delineation line.
4. (5). Geological Boundary Digitization. Perform interactive digitization of exploration sections based on geological unit distribution and contact relationships. ① Extract intersection points of geological interfaces from drillholes to obtain scattered point datasets; ② Digitize geological boundaries along exploration sections to capture interface traces. Strictly control node spacing during digitization, optimize boundary precision through precise node manipulation, ensure smooth and accurate representation of geological boundaries, and convert digitized boundaries into scattered point datasets.

5.1. Knowledge-driven constraint implementation

(1) The deep rock mass in the study area exhibits a complex three-dimensional morphology with significant local variations, making it difficult to construct an accurate geological model using data-driven methods alone. Therefore, it is necessary to integrate expert knowledge to constrain the modeling process, thereby enhancing the accuracy and reliability of the model. Based on modeling data, regional geological background, and expert experience, the age relationships and spatial distribution characteristics of various geological bodies, as well as their three-dimensional morphology, can be clearly defined, providing a priori constraints for three-dimensional geological modeling.

1. (2). Expert knowledge constraints refer to the process of constraining and optimizing the construction of 3D geological models based on a deep understanding of regional geological characteristics by geological experts, integrating multiple sources of data such as geological maps, borehole data, and exploration line profiles. The key steps for implementing expert knowledge constraints in this study are as follows: first, constructing virtual geological cross-sections using expert knowledge; second, converting these virtual cross-sections into discrete 3D spatial points; and third, using these points as data sources to constrain the model, thereby optimizing the morphology of the 3D geological model. In situations where modeling data is insufficient, expert knowledge can leverage long-term accumulated geological research experience and regional geological patterns to make scientific inferences and supplement data in areas where information is missing or unclear. This study employs the following two methods to enhance model quality:

① In sections where geological bodies exhibit significant undulations, virtual cross-sections are obtained through equidistant interpolation based on existing exploration line profiles. Simultaneously, geological experts adjust the geological boundaries within these virtual cross-sections based on their understanding of the three-dimensional morphology of the geological bodies (Fig 5). The number of additional virtual cross-sections required should be determined by the complexity of the local geological bodies. For areas with complex morphology, a single virtual cross-section may not be sufficient to control the three-dimensional shape of the geological bodies, necessitating an appropriate increase in the number of cross-sections to supplement critical geological interface details, thereby significantly enhancing the accuracy and reliability of the modeling.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Distribution of extended exploration sections and interpolated virtual sections.

1. Original exploration sections; 2. Extended exploration sections; 3. Interpolated virtual sections.

https://doi.org/10.1371/journal.pone.0331937.g005

② When constructing geological interfaces using implicit function interpolation, the surfaces tend to extend automatically to the model boundaries. Because these boundaries are often poorly constrained by data, the automatically extended interfaces may deviate substantially from real geological conditions. To address this issue, the boundaries of existing exploration-line profiles can be expanded based on the variations in the direction and dip of the geological boundaries within the original exploration line profiles. This method ensures that the boundaries of the constructed geological interfaces are more consistent with actual geological conditions, effectively enhancing the continuity and geological rationality of the model boundary regions.

5.2. Geological surface construction

Modeling Units and Surface Typology. The model domain comprises eight primary units: Nanhua System (Nh), pegmatite-type Nb-Ta mineralization zone (ρ), greisenized granite-type Nb-Ta mineralization zone (γg), K-feldspathized granite-type Nb-Ta mineralization zone (γd), weak albitized granite subzone (γc), moderate albitized granite subzone (γb), strong albitized granite subzone (γa), and biotite granite zone (γ). Defined surface categories include DEM surfaces and geological interfaces.

1. (1). DEM Surface Construction. Convert contour data with elevation attributes into scattered point datasets. Perform data validation to eliminate duplicate points. Generate 3D surfaces by triangulating the processed point set (Fig 6a). Where drillhole collar coordinates deviate from the DEM surface, spatially adjust triangulation nodes to ensure correspondence with actual collar positions.
2. (2). Geological Interface Construction. Digitize boundaries from exploration sections, densify nodes along digitized traces, and convert boundary data into scattered point datasets. Spatially interpolate these point datasets combined with borehole interval data through triangulation to construct geological interfaces (Fig 6b).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Display of DEM reference surface and nanhua system (Nh) geological interface.

(a) DEM reference surface; (b) Geological interface of Nanhua System (Nh).

https://doi.org/10.1371/journal.pone.0331937.g006

Refining Complex Geological Interfaces. Geological interfaces built solely from exploration sections and borehole data are often inaccurate for two reasons: (1) protracted exploration cycles resulting in section drafting based predominantly on drillhole data along individual lines, lacking comprehensive 3D geological continuity analysis that causes topological inconsistencies between adjacent sections; (2) uneven and sparse data distribution yielding inadequate constraint density for interface geometry. To correct these inaccuracies, we employ an integrated analysis that incorporates multiple sources of information: existing sections, borehole data, local structural knowledge, regional geological principles, and other site-specific features. Enhance constraint density through strategic insertion of virtual sections between existing exploration lines and extension of section coverage. Following expert-guided interpretive digitization of new boundaries on virtual/extended sections, fuse these with original data to reconstruct updated geological interfaces, completing the initial update cycle.

To ensure that the constructed geological interfaces align with actual geological conditions, it is necessary to validate the updated geological interfaces. If discrepancies are found between the geological interfaces and the actual geological conditions, the exploration line profiles, virtual geological profiles, and extended exploration line profiles should be revised based on expert knowledge, and the geological interfaces should be updated again. The iteration can be terminated when the geological structures, shapes of geological bodies, and other key geological elements reflected in the model in three-dimensional space are consistent with objective geological data and expert judgment. Through repeated “validation-correction” iterative optimization processes, a geological interface that closely matches the actual conditions can ultimately be obtained (Fig 7).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Schematic workflow of knowledge-constrained geological interface construction.

1. Nanhua System (Nh); 2. Pegmatite-type Nb-Ta mineralization zone (ρ); 3. Greisenized granite-type Nb-Ta mineralization zone (γg); 4. K-feldspathized granite-type Nb-Ta mineralization zone (γd); 5. Weak albitized granite subzone (γc); 6. Moderate albitized granite subzone (γb); 7. Strong albitized granite subzone (γa); 8. Biotite granite zone (γ); 9. Extended exploration sections; 10. Interpolated virtual sections.

https://doi.org/10.1371/journal.pone.0331937.g007

5.3. 3D Geological structure model construction

First, an initial geological model was constructed by sequentially truncating a base volume. This process was guided by a chronological sequence of all geological interfaces (from youngest to oldest), determined from their spatial and cross-cutting relationships. During this stage, the appearance of any voids within the model served as a key validation check, signaling an incorrect chronological sequence that required immediate readjustment.

After the model is constructed, its consistency with the actual geological conditions must be verified. If discrepancies are identified, expert knowledge should be integrated to revise the existing exploration line cross-sections (including virtual geological cross-sections and extended exploration line cross-sections). If necessary, additional virtual geological cross-sections can be added, and the geological interfaces should be updated accordingly. Through this iterative optimization process based on expert knowledge, the model should be refined until it closely matches the existing boreholes, measured cross-sections, and other modeling data. The model should also ensure smooth transitions between rock bodies and geological interfaces, and its overall morphology should conform to regional geological patterns and align with the understanding of geological experts (Fig 8). The accuracy of the model can be validated by comparing model slices with existing actual cross-sections and boreholes (Fig 9). The final model should be spatially consistent with all existing boreholes and cross-section data.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. Schematic diagram of the 3D geological structure model.

1. Nanhua System (Nh); 2. Weakly Albitized Granite Subzone (γc); 3. Moderately Albitized Granite Subzone (γb); 4. Strongly Albitized Granite Subzone (γa); 5. Biotite Granite Zone (γ).

https://doi.org/10.1371/journal.pone.0331937.g008

Download:

• PNG
  larger image
• TIFF
  original image

Fig 9. Comparison of model cross-section and actual profile.

(a) Exploration Line 1 Model Slice; (b) Profile of Exploration Line 1. 1. Nanhua System (Nh); 2. Pegmatitic niobium-tantalum mineralization zone (ρ); 3. Greisenized granite-hosted niobium-tantalum mineralization zone (γg); 4. Potassic-altered granite-hosted niobium-tantalum mineralization zone (γd); 5. Weakly albitized granite subzone (γc); 6. Moderately albitized granite subzone (γb); 7. Intensely albitized granite subzone (γa); 8.Biotite granite zone (γ).

https://doi.org/10.1371/journal.pone.0331937.g009

5.4 3D Ore grade modeling construction

The borehole assay data in the modeling domain primarily include tungsten trioxide (WO₃), tin (Sn), molybdenum (Mo), tantalum pentoxide (Ta₂O₅), niobium pentoxide (Nb₂O₅), and rubidium oxide (Rb₂O). Utilizing the 3D geological structure model as a geometric constraint framework, create an attribute-empty 3D orebody solid model. Employ the FastRBF function to interpolate ore block cells within this model based on grade data of individual oxides, ultimately constructing the 3D ore grade model. Conduct mathematical statistical analysis of oxide grade data within the model, establish grade classification ranges, and visualize the results through color-coded rendering (Fig 10).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 10. Schematic diagrams of the 3D ore grade models.

(a) Mo grade model; (b) WO₃ grade model; (c) Sn grade model; (d) Ta₂O₅ grade model; (e) Nb₂O₅ grade model; (f) Rb₂O grade model.

https://doi.org/10.1371/journal.pone.0331937.g010

6.1. Deep geological characteristics analysis

Within the study area, the following geological units develop sequentially from surface to subsurface: Nanhua System (Nh), Pegmatitic Nb-Ta Mineralization Zone (ρ), Greisenized Granite Nb-Ta Mineralization Zone (γg), Potassic-Altered Granite Nb-Ta Mineralization Zone (γd), Weakly Albitized Granite Subzone (γc), Moderately Albitized Granite Subzone (γb), Strongly Albitized Granite Subzone (γa), and Biotite Granite Zone (γ) (Fig 11). The subsurface rock mass exhibits distinct ore-type zoning, vertically from top to bottom, four natural ore types are developed: the Pegmatitic Nb-Ta Mineralization Zone (ρ), the Potassic-Altered Granite Nb-Ta Mineralization Zone (γd), the Greisenized Granite Nb-Ta Mineralization Zone (γg), and the Albitized Granite Nb-Ta Mineralization Zone. The Albitized Granite Nb-Ta Mineralization Zone is further subdivided based on alteration intensity and spatial position: the Weakly Albitized Granite Subzone (γc), the Moderately Albitized Granite Subzone (γb), the Strongly Albitized Granite Subzone (γa), and Unaltered Biotite Granite. The deposit is dominated by albitized granite-type ores, which form thick, stratoid bodies concentrated in the central and lower parts of the rock mass. Overlying these are greisenized granite-type ores, characterized by saddle-like distributions and zones of localized, intense greisenization. In contrast, the two remaining ore types are discontinuous and have poor lateral continuity, occurring in varied forms such as banded, thin-shelled, saddle-like, or massive bodies.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 11. 3D visualization of geological bodies.

(a) 3D geological structure model; (b) Nanhua System (Nh) geobody; (c) Pegmatitic niobium-tantalum mineralization zone (ρ) geobody; (d) Greisenized granite-hosted niobium-tantalum mineralization zone (γg) geobody; (e) Potassic-altered granite-hosted niobium-tantalum mineralization zone (γd) geobody; (f) Weakly albitized granite subzone (γc) geobody; (g) Moderately albitized granite subzone (γb) geobody; (h) Intensely albitized granite subzone (γa) geobody; (i) Biotite granite zone (γ) geobody. 1. Nanhua System (Nh); 2. Pegmatitic niobium-tantalum mineralization zone (ρ); 3. Greisenized granite-hosted niobium-tantalum mineralization zone (γg); 4. Potassic-altered granite-hosted niobium-tantalum mineralization zone (γd); 5. Weakly albitized granite subzone (γc); 6. Moderately albitized granite subzone (γb); 7. Intensely albitized granite subzone (γa); 8. Biotite granite zone (γ).

https://doi.org/10.1371/journal.pone.0331937.g011

1. (1). The Nanhua System (Nh) occurs in the shallow subsurface, and exposed at the surface, spanning elevations from −601.33 m to 692.98 m.
2. (2). The Pegmatite-type Niobium-Tantalum Mineralization Zone (ρ) occupies the intrusion’s upper contact zone at 38.34–412.25 m elevations, exhibiting relatively consistent thickness (0–18 m).
3. (3). The Potassic-altered Granite-hosted Niobium-Tantalum Mineralization Zone (γd) is located at the southeastern apex of the pluton. Minor occurrences underlie pegmatitic mineralization zones, while the majority directly contacts the Nanhua System strata. This zone exhibits a thin-lens, extending 150–280 m downdip to the southeast and 30–160 m along strike northeast. It displays discontinuous distribution with irregular thickness variations ranging from 0 to 35 m, within an elevation interval of 397.12 to 261.13 m.
4. (4). The Greisenized Granite-hosted Niobium-Tantalum Mineralization Zone (γg) is stratigraphically constrained between the underlying albitized granite-hosted ores and the overlying pegmatitic and potassic-altered zones. Exhibiting a saddle-shaped morphology, this unit has a variable thickness of 15–170 m and occurs within an elevation range of 19.70 m to 395.64 m.
5. (5). The Weak Albitized Granite Subzone (γc) is located below the greisenized granite mineralization zone and above the moderately albitized mineralized zone. The ore body is thicker between exploration lines 2 and 3 and becomes thinner to the east. It exhibits a saddle-shaped or pseudo-stratiform distribution, with a wide distribution range. The elevation range of the ore body is from 299.32 to −259.00 m, and the maximum thickness controlled by a single engineering project is 247.28 m. The thickness is mainly concentrated between 40 and 80 m.
6. (6). The Moderately Albitized Granite Subzone (γb) underlies the weakly albitized mineralization zone and overlies the intensely albitized mineralization zone. The subzone contacts metamorphic rocks or overlying porphyritic granite along its northwestern and southeastern margins. Occurring at −312.00 to 189.40 m elevations, it demonstrates a maximum borehole intersection of 135.66 m, with predominant thicknesses concentrated between 40–180 m.
7. (7). The Strongly Sodic Plagioclase-altered Granite Subzone (γa) is located beneath the moderately sodic plagioclase-altered mineralized zone and represents the deepest mineralized rock exposed by drilling. The elevation range of the ore body is −25.33 to −330.76 m, with a maximum thickness of 207.36 m controlled by a single engineering project. The thickness is primarily concentrated between 65.61 and 153.16 m.
8. (8). The Biotite Granite (γ) forms the basal unit of the model, underlying the intensely albitized granite zone. The orebody occurs below approximately −230 m elevation.

6.2. Analysis of orebody distribution characteristics

(1) Molybdenum, tungsten, and tin orebodies primarily occur within the Nanhua System strata, concentrated in the exocontact zone of the granite pluton (Figs 10 and 12). The granite underwent multistage evolution from magmatic crystallization to hydrothermal alteration, with greisenization and potassic alteration representing key metallogenic episodes for molybdenum mineralization. These alteration zones exhibit enriched Mo-W-Sn mineralization. Spatially, the deposit exhibits pronounced zonation, with grades peaking in the alteration centers and diminishing outwards. This pattern varies by metal: molybdenum orebodies, though limited in size, are concentrated in the northwestern sector, whereas tungsten and tin are enriched along the northern and southern margins, defining a depleted central core. In near-surface quartz veins, tungsten and tin orebodies demonstrate close paragenetic relationships. At depth, this association expands to include mineralization hosted in pegmatitic, potassic-altered, and greisenized granite. Mineralogically, tungsten principally occurs as wolframite and tin as cassiterite within quartz veins, typically co-crystallizing in these structures.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 12. E-W and N-S section views of 3D grade model.

(a) E-W section location diagram of 3D geological structure model; (b) Multi-element grade distribution section along E-W orientation; (c) N-S section location diagram of 3D geological structure model; (d) Multi-element grade distribution section along N-S orientation.

https://doi.org/10.1371/journal.pone.0331937.g012

1. (2). Tantalum, niobium, and rubidium orebodies predominantly occur within the granite pluton, demonstrating peak mineralization enrichment in albitized and greisenized granites, with minor occurrences are observed in the exocontact zone (Figs 10 and 12). Orebodies typically exhibit disseminated distribution patterns, with morphology intrinsically controlled by pluton contacts and alteration zones. Spatially, grade distribution displays marked zonation, with maximum values in central alteration zones decreasing peripherally. Specifically, rubidium orebodies concentrate at the pluton apex, the granite-Nanhua System contact, and transitional domains between greisenized and albitized granites. Tantalum and niobium orebodies primarily distribute within the pluton, particularly along greisenized-albitized granite interfaces. Although greisenized granites yield higher Ta-Nb grades, limited development of such rocks renders albitized granites the principal host. Within these units, tantalum and niobium minerals occur as complex oxides-showing close spatial association but distinct mineralization characteristics-with rubidium orebodies demonstrating significant spatial correlation with Ta-Nb mineralization.

6.3. Prospective mineralization targets

The tantalum-niobium orebodies in the Songshugang mining area are primarily controlled by hydrothermal alteration, with their spatial distribution closely linked to the 3D morphology of the granitic body and alteration zones. The 3D geological model (Fig 13) reveals significant enrichment of tantalum-niobium mineralization within granitic alteration zones. The orebodies are found primarily along the upper and lower contacts of the albitized granite, and notably, the eastward extension of this enrichment zone remains undrilled, indicating substantial deep-seated prospecting potential.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 13. Predictive model of the songshugang Ta-Nb deposit.

(a) 3D geological structure model; (b) 3D Ta-Nb grade model. 1. Nanhua System (Nh); 2. Weakly albitized granitic subzone (γc); 3. Moderately albitized granitic subzone (γb); 4. Strongly albitized granitic subzone (γa); 5. Biotite granitic zone (γ); 6. Ta > 0.01% and Nb > 0.015%; 7. Ta > 0.015% and Nb > 0.02%; 8. Ta > 0.018% and Nb > 0.024%; 9. Predicted favorable target zones; 10. Exploration line locations and numbers.

https://doi.org/10.1371/journal.pone.0331937.g013

Based on the three-dimensional geological model, two new favorable mineral exploration targets have been delineated (Fig 13). Target Zone A, located east of Exploration Line 13 (elevations −300 m to −400 m), this is an unexplored area adjacent to the Lingshan pluton. It has been prioritized for systematic new exploration. Target Zone B, located between Exploration Lines 4 and 7 (elevations −600 m to −700 m), this zone represents a potential deep extension of known orebodies. This target is supported by our model, which indicates that Ta-Nb mineralization continues downward beyond the limits of current deep-level exploration. Both predicted zones demonstrate promising mineralization prospects and are designated as priority targets for the next exploration phase.

7.1. Conclusions

This study utilized the Leapfrog Geo platform to integrate multisource data, including borehole logs and exploration line profiles. Employing the FastRBF algorithm, we constructed expert knowledge-constrained 3D geological structure and grade models of the Songshugang Ta-Nb mining area in Hengfeng, Jiangxi Province. Systematic analysis of deep geological structures and Ta-Nb mineralization prediction yielded the following key findings:

1. (1). For the complex geological setting of the Songshugang mining area in Hengfeng, Jiangxi, an innovative expert knowledge-constrained implicit 3D modeling approach was developed. An iterative “modeling-verification-modification” workflow was employed to produce structural and grade models that are highly consistent with real-world geology, effectively overcoming the challenges of representing complex geological structures.
2. (2). Analysis of the 3D geological structure and grade models revealed two primary zones of significant Ta-Nb enrichment: the contact between albitized granite and the overlying K-feldspathized/greisenized granite, and its interface with the underlying biotite granite. Based on these findings, we have delineated two high-priority exploration targets, providing a robust geological basis for future prospecting efforts.

7.2. Discussion

While this study has yielded significant findings, certain aspects require further refinement:

1. (1). In the southwestern and northeastern sectors of the mining area, limited borehole density and sparse exploration line coverage constrained expert knowledge application, introducing modeling uncertainties. Subsequent model refinement requires supplemental data acquisition to enhance model accuracy.
2. (2). Although the expert knowledge-constrained implicit 3D geological modeling approach significantly enhanced modeling accuracy for complex structures, divergent geological interpretations among experts-regarding regional tectonic evolution, geological contact relationships, and deep 3D geometry-may lead to inconsistent constraint criteria. Future work could integrate machine learning algorithms to optimize expert knowledge constraints.
3. (3). Although the currently delineated favorable prospecting targets based on the 3D geological model are geologically constrained, they require engineering verification. Particularly given the complex deep-seated tectonic evolution and multiphase mineralization in the mining area, unidentified ore-controlling factors may impact prediction reliability.