minerals

Article
Metallogenic Prediction of the Zaozigou Gold Deposit Using
3D Geological and Geochemical Modeling
Cheng Li 1,2 , Bingli Liu 2,3, *, Keyan Xiao 1 , Yunhui Kong 2 , Lu Wang 2 , Rui Tang 2 , Miao Xie 2 and Yixiao Wu 2

1 SinoProbe Laboratory, Institute of Mineral Resources, Chinese Academy of Geological Sciences,

Beijing 100037, China; leecheng88@163.com (C.L.); kyanxiao@sohu.com (K.X.)
2 Geomathematics Key Laboratory of Sichuan Province, Chengdu University of Technology,
Chengdu 610059, China; kongyunhui@stu.cdut.edu.cn (Y.K.); wanglu0834@aliyun.com (L.W.);
tangrui0223@outlook.com (R.T.); xiemiao@stu.cdut.edu.cn (M.X.); wuyx@stu.cdut.edu.cn (Y.W.)
3 UNESCO International Center on Global-Scale Geochemistry, Langfang 065000, China
* Correspondence: liubingli@cdut.edu.cn; Tel./Fax: +86-028-8407-3610

Abstract: Deep-seated mineralization prediction is an important scientific problem in the area of min-
eral resources exploration. The 3D metallogenic information extraction of geology and geochemistry
can be of great help. This study uses 3D modeling technology to intuitively depict the spatial distribu-
tion of orebodies, fractures, and intrusive rocks. In particular, the geochemical models of 12 elements
are established for geochemical metallogenic information extraction. Subsequently, the front halo
element association of As-Sb-Hg, the near-ore halo element association of Au-Ag-Cu-Pb-Zn, and the
tail halo element association of W-Mo-Bi are identified. Upon this foundation, the 3D convolutional
neural network model is built and used for deep-seated mineralization prediction, which expresses a
high performance (AUC = 0.99). Associated with the metallogenic regularity, two mineral exploration
targets are delineated, which might be able to serve as beneficial achievements for deep exploration
in the Zaozigou gold deposit.

Keywords: metallogenic prediction; 3D geological modeling; 3D geochemical modeling; Zaozigou

gold deposit; 3D convolutional neural network

Citation: Li, C.; Liu, B.; Xiao, K.;

Kong, Y.; Wang, L.; Tang, R.; Xie, M.;
Wu, Y. Metallogenic Prediction of the
Zaozigou Gold Deposit Using 3D
1. Introduction
Geological and Geochemical Deep exploration is an important research direction of mineral resource prediction,
Modeling. Minerals 2023, 13, 1205. which requires continuous innovation in theoretical and methodological approaches, elabo-
https://doi.org/10.3390/ rate 3D modeling of the complicated geological and geochemical mass, and the technologi-
min13091205 cal support of modern statistical methods, such as artificial intelligence, to extract deeper
Academic Editor: Stanisław Mazur
information and inference.
The current the mainstream method for deep-earth resource prediction is 3D-modeling-
Received: 17 August 2023 based quantitative mineral prediction at depth, and its key technologies are: (i) the 3D visual
Revised: 8 September 2023 representation of metallogenic geological mass, metallogenic structural zone of contact, and
Accepted: 11 September 2023
geochemistry [1–4]; (ii) 3D mineralization information extraction based on spatial analysis
Published: 13 September 2023
technology [5–10]; and (iii) studies on the distribution and prediction of mineralization
based on data analysis methods [11–21]. Three-dimensional quantitative mineral resource
prediction is gaining growing significance in the exploration of deep mineral resources. As
Copyright: © 2023 by the authors.
3D models become more refined and methods for extracting mineralization information
Licensee MDPI, Basel, Switzerland.
continue to evolve and deepen, the scientific rigor and practical utility of mineral resource
This article is an open access article prediction are poised for significant enhancement.
distributed under the terms and The basis of 3D Mineral Prospectivity Mapping is driven by a 3D geological model,
conditions of the Creative Commons and, therefore, 3D geological modeling has become an important part of 3D metallogenic
Attribution (CC BY) license (https:// prediction. The relevant approach is to integrate geological information, such as topography,
creativecommons.org/licenses/by/ geological mapping, drill core logging, geological section, and geophysical inversion
4.0/). results, to improve the characterization of a complex geological mass in 3D space. Simon W.

Minerals 2023, 13, 1205. https://doi.org/10.3390/min13091205 https://www.mdpi.com/journal/minerals

Minerals 2023, 13, 1205 2 of 22

Houlding first proposed 3D geological modeling [1] and raised the theory of 3D geological
modeling in conjunction with geological modeling systems, including the establishment
of spatial databases, spatial data models and data structures, and the 3D representation
of geological mass. The 3D geological modeling approach is mostly built based on the
modeling of drill data, the exploration profiles and tunnels, etc. It is mainly applied to
the high-precision geometric modeling of rock masses, faults, and orebodies at the ore
deposit scale, and the key is controlling the modeling accuracy for meeting the elaborate
requirement of 3D visualization of complicated geological masses and structures [6,22–28].
Primary geochemical halo identification is a common geochemical approach for detect-
ing orebodies near or below the ground surface [29]. A primary geochemical halo is an area
of rocks surrounding orebodies enriched in ore-forming elements [30]. Primary geochemi-
cal characteristics of mineral deposits provide important information for predicting deep
mineral resources, as they reflect the geochemical processes of metal precipitation and min-
eral formation. Many methods have been used to identify the primary halo characteristics
of mineral deposits, including vertical element zonation [31–33], element ratios vectoring
toward ore zones [34–36], Pearce element ratios [29,37–40], and alteration indices [41–49].
A major aspect of these methods is the determination of the spatial distributions of single
elements and/or element associations.
The practice of geochemical surveys has shown that primary halo or geochemical rock
surveys are one of the most effective methods for inferring the depth of erosion of ore
deposits, determining the occurrence of orebodies, searching for hidden orebodies, and
making predictions for mineralization [50–54]. Geochemical anomalies in the primary halo
are an important indicator of the presence of potential orebodies at depth, and many case
studies have shown the presence of primary haloes dominated by ore-forming elements
around orebodies of gold, silver, copper, lead, and zinc, especially hydrothermal deposits,
where a wide range of primary geochemical anomalies are developed [53,55].
Convolutional Neural Network (CNN), as the most successful neural network model
applied in deep learning, has been successfully applied in the field of geology, such as in
lithology recognition [56], geological mapping [57], and three-dimensional geological structure
inversion [58]. In recent years, many scholars have also achieved good results in applying
convolutional neural networks to mineral exploration and prediction research [59–62].
Based on the above analysis, this study addresses the scientific problem of “3D elabo-
rate modeling of mineralization space”, takes the Zaozigou gold deposit in the West Qinling
Mountains as the study area, builds geological and geochemical models using the deep
learning algorithms of 3D convolutional neural network, and predicts two exploration
targets at depth.

2. Geological Setting
2.1. Regional Geological Background
The West Qinling region is situated in the western part of the Qinling orogenic belt. It
is bordered by the Qilian orogenic belt to the north, the Qaidam massif to the west, and
the Songpan-Garze terrane to the south (Figure 1a). Its geological evolution is closely
intertwined with the history of the Tethys tectonic domain and the convergence of the
Yangtze Massif and the North China Plate [63–67].
The geotectonic position of the Xiahe-Hezuo area is in the northwestern part of the
West Qinling orogenic belt and represents the western extension of the Qinling-Qilian-
Kunlun central orogenic belt (Figure 1a). The intricate geological structure in this region
creates a favorable environment for mineralization [68].
The outcropping strata in the Xiahe-Hezuo area are primarily of Mesozoic and Late
Paleozoic formations, and they are generally distributed along fracture lines trending north-
westward. North and south of the Xiahe-Hezuo fracture, Mesozoic and Late Paleo-zoic
strata are, respectively, developed, with the northern region dominated by Carboniferous
and Permian formations and the southern region primarily comprising Triassic formations,
sporadically exposing Neogene and Quaternary units (Figure 1b). The geological structure
Minerals 2023, 13, 1205 3 of 22

within the area is primarily composed of folds and fractures. Folds, in conjunction with
fractures, jointly control the distribution of gold deposits. The fractures predominantly
trend in the north-west direction, and these fractures serve as the primary conduits for
magma-hydrothermal fluid migration. It is evident from the figure that gold deposits are
primarily located near fractures, and their distribution trend aligns with the orientation
of the fractures. Additionally, the presence of fractures also influences the distribution
pattern of Yanshanian intermediate-acidic rocks. With the Xiahe-Hezuo fracture as the
demarcation, larger-scale batholiths are developed in the northern region. In contrast, the
southern region primarily features vein-like formations, resulting in an overall “northwest
belt, southeast cluster” distribution pattern. Specifically, the Zaozigou gold deposit is a
typical shallow-seated low-temperature hydrothermal gold deposit in the Xiahe-Hezuo
area. The exact location is indicated by a black box in Figure 1b.

Figure 1. (a) Geological sketch of the demonstration area. NQL—North Qinling Tectonic Belt; SDS—
Figure 1. (a) Geological sketch of the demonstration area. NQL—North Qinling Tectonic Belt;
Shangdan ancient suture zone; CBS—North Qaidam ancient suture zone; AMS—A’nyemaqen
SDS—Shangdan ancient suture zone; CBS—North Qaidam ancient suture zone; AMS—A’nyemaqen
ancient suture zone; MLS—Mianlue suture zone. (b) The distribution map of main minerals in
ancient suturearea
Xiahe-Hezuo zone; MLS—Mianlue
(modified suture[69]).
from reference zone. (b) The distribution map of main minerals in
Xiahe-Hezuo area (modified from reference [69]).

2.2. Mineral Deposit Features

The Zaozigou Gold deposit is a typical representative of gold deposit associated with
intermediate-acid dike rocks in the south of the Xiahe-Hezuo fracture. It is located at
approximately 9km southwest of the Hezuo city, Gansu Province, with convenient access
to the mine site. The main ore-bearing position is between Gully 1 and Gully 4, with a total
area of approximately 2.6 km2 (Figure 2).
 2.2. Mineral Deposit Features
The Zaozigou Gold deposit is a typical representative of gold deposit associated with
intermediate-acid dike rocks in the south of the Xiahe-Hezuo fracture. It is located at ap-
proximately 9km southwest of the Hezuo city, Gansu Province, with convenient access to
Minerals 2023, 13, 1205 4 of 22
the mine site. The main ore-bearing position is between Gully 1 and Gully 4, with a total
area of approximately 2.6 km2 (Figure 2).

Figure 2. Geological map of Zaozigou gold deposit (modified from reference [70]).
Figure 2. Geological map of Zaozigou gold deposit (modified from reference [70]).

The geological formations

The geological exposed in the exposed
formations Zaozigou in are
the relatively
Zaozigou are homogeneous, pri-
relatively homogeneous, pri-
marily consisting of Triassic
marily and Quaternary
consisting of Triassic and strata. The Quaternary
Quaternary strata. Thedeposits
Quaternary are predom-
deposits are predomi-
inantly located in the southern
nantly located inand the eastern
southern parts
and of the study
eastern partsarea.
of theAt the same
study area. time,
At thethesame time, the
Triassic
Triassic formations belongformations
mainly belong mainlysection
to the lower to the lower
of thesection
MiddleofTriassic
the Middle Triassic Gulangdi
Gulangdi
Group (T2g1). The Group (T2 g1 ).comprises
formation The formation
a largecomprises
sedimentary a large
cyclesedimentary
from bottom cycle from
to top con-bottom to top
consisting of siltstone → argillaceous slate → calcareous slate
sisting of siltstone → argillaceous slate → calcareous slate [19,71]. The region exhibits well- [19,71]. The region exhibits
well-developed fractures, broadly categorized into two
developed fractures, broadly categorized into two sets: those trending in an east–west sets: those trending in an east–west
(EW) direction and(EW) direction
those trending andinthose trending
a northeast in adirection.
(NE) northeastThere
(NE) direction.
are also a few There
frac-are also a few
fractures oriented in a north–south (NS) and northwest (NW) direction, although these
tures oriented in a north–south (NS) and northwest (NW) direction, although these are
are localized. As depicted in Figure 1, the NW-oriented fractures will likely represent
localized. As depicted in Figure 1, the NW-oriented fractures will likely represent major
major regional fractures, primarily serving as conduits for ore-forming hydrothermal fluid
regional fractures, primarily
migration. In serving
contrast,as conduits
fractures for ore-forming
trending hydrothermal
in other directions, such asfluid mi- and EW, are
NE, NS,
gration. In contrast, fractures trending in other directions, such as NE, NS,
considered secondary fractures and constitute the primary sites for ore deposition. and EW, are The
considered secondary fractures
study area and constitute
is influenced the primary
by the NE-oriented sites for
fractures ore deposition.
resulting The movements
from tectonic
study area is influenced
during the byYanshan
the NE-oriented
period, whichfractures
controlresulting from tectonic
the orientation movements
of rock bodies. It is evident from
during the Yanshan period, which control the orientation of rock bodies. It
the figure that Quartz diorite porphyrite, Granodiorite-porphyry, and Plagioclase is evident from granite
the figure that Quartz
porphyry diorite
are allporphyrite,
predominantly Granodiorite-porphyry,
distributed along theand NE Plagioclase
trend, including granite
the distribution
porphyry are allpattern
predominantly distributed along the NE trend, including the distribution
of gold orebodies.
The fractures exhibit a robust development and intricate morphology, displaying an
pattern of gold orebodies.
The fracturesintersecting patterndevelopment
exhibit a robust that suggestsand multiple episodes
intricate of geological
morphology, activityanin the region.
displaying
intersecting pattern that suggests multiple episodes of geological activity in the and
These fractures exert precise control over the distribution of orebodies vein formations
region.
within the deposit. They can be categorized into four principal orientations: NW (blue),
NE (red), E–W (green), and NNE (orange) [72–74].
1 The E–W (green) orientation fracture group is likely the earliest to appear, as nearly
all other fractures intersect. Among this group, the longest fracture is F20, located in the
Minerals 2023, 13, 1205 5 of 22

study area’s central part, traversing the entire mining site and intersecting with most of the
gold deposits within the area.
2 The NW (blue) orientation fracture group emerged after the E–W fractures and is
relatively rare within the study area, intersecting with NNE (orange) fractures and being
intersected by E–W (green) fractures, suggesting a chronological placement between these
two orientations. Referring to Figure 1b, it is apparent that this fracture belongs to the NW
regional trend and likely served as a conduit for hydrothermal fluid migration rather than
as a site for mineral enrichment. Consequently, Figure 2 indicates that gold deposits are
generally located away from this fracture.
3 The NNE (orange) orientation fractures are predominantly distributed in the north-
western part of the study area. The illustration shows that F37 intersects the NW (blue)
orientation fractures, and F40 intersects the NE (red) fracture F19, indicating its later forma-
tion. Despite three fractures within this group, only F40 is relatively close to a gold deposit.
However, the orientation of this gold deposit aligns more closely with NE (red) fracture
F19, indicating a less direct relationship with F40.
4 The NE (red) orientation fractures are the most extensively distributed within the
study area, with several gold deposits exhibiting orientations consistent with this group of
fractures. These fractures, characterized by their later formation, intersect with fractures of
other orientations.
Despite the broad classification of fractures into the four groups mentioned above,
there are also localized NS (north–south) orientation fractures. While some fractures
may exhibit a closer spatial relationship with the distribution of mineral deposits, such
as the NE (red) orientation fractures, the formation of mineral deposits is a complex
and protracted process. Single fractures, on their own, are insufficient to account for
mineralization. Instead, the complex interplay of overlapping fractures enhances the
potential for mineralization.
The geological context encompasses an intermediate-acid dike featuring fine crys-
talline diorite, diorite porphyrite, biotite dioritic porphyrite, and quartz diorite porphyrite,
notable for its densely developed porphyritic structure. This dike exhibits a predominant
trend in the NNE direction, which gradually transitions to an approximately N–S orienta-
tion within the western section of the deposit, with additional occurrences trending in a
few NW directions. The complex mineralization within the deposit has been significantly
influenced by regional multi-period deep fracture activity, resulting in the overlapping of
successive phases of magmatism across multiple mineralization stages [75].
There are 147 gold orebodies that have been found in the Zaozigou gold deposit, of
which 17 orebodies are main orebodies with gold reserves greater than 1 ton, with the
total gold reserve being more than 100 tons [76]. According to the spatial distribution and
combination of the mineralization, the Zaozigou deposit can be divided into eastern and
western ore blocks (Table 1).

Table 1. Orebodies and ore-bearing fracture.

Ore Block Orientation Fracture Orebody

Au1, Au9, Au10, Au15, Au45,
NE F19~F27 Au46, Au16, Au17, Au21, Au37,
Eastern ore block
Au52, Au55
NWW F3 M6, M4
NW F42, F41, F34 Au14
Western ore block
N–S F13~F11, F9~F8 Au29, Au31

The eastern ore block is primarily situated between Gully 1 and Gully 3, oriented in
a northeast direction. It encompasses Au1, which is controlled by F24, Au9 controlled by
F21, and Au15 controlled by F25. These orebodies extend over 1000 m long and 300 m
wide, with a NW direction tendency and a steep dip near the ground surface, locally nearly
upright. In the depths, these orebodies have been staggered by a gently dipping fracture,
Minerals 2023, 13, 1205 6 of 22

causing the tendency to change to a SE orientation (Figure 2). In addition, orebodies M4

and M6 lie underground, with a strike of NWW orientation, a tendency of SW orientation,
and a dip of 8◦ ~26◦ . These orebodies cross the NE-striking orebodies obliquely, staggering
them, and their own mineralization behavior occurs simultaneously [67].
The western ore block is primarily distributed between Gully 3 and Gully 4, extending
in a nearly north–south direction. The main orebodies in this area are Au29 to Au31.
The orebodies extend over 1000 m long and wider than 500 m, with a strike of 350◦ ~10◦ ,
varying tendency, and dips greater than 75◦ , which are locally subvertical. These orebodies
extend stably, and the mineralization is weaker in the steeper parts and stronger in the
slower parts.
The alteration of the surrounding rocks is mainly controlled by different tectonic and
vein intrusion activities. From the fractures, veins, and the center of the orebody to both
sides of the surrounding rocks, the alteration of the surrounding rocks weakens gradually.
This can be generally divided into silicification (stibnite mineralization, pyritization, and
arsenopyritization) → sericitization (chloritization and epidotization) → hematitization
and carbonation.
According to previous studies in the Zaozigou gold deposit, the metallogenic process
can be divided into two parts: the hydrothermal mineralization and the supergene min-
eralization period. The hydrothermal period is further divided into four phases: (i) early
silicification–hematitization phase; (ii) quartz–pyrite–arsenopyrite phase, where the min-
eral assemblage and production sequence are quartz → pyrite → arsenopyrite → native
gold. It is the main gold-forming phase of the deposit; (iii) quartz–stibnite–orpiment–
calcite (ankerite) phase, where the mineral assemblage and production sequence are
quartz → pyrite → arsenopyrite → stibnite→ native gold; and (iv) quartz–calcite phase,
where the mineral assemblage and production sequence are quartz → calcite → kaolin.
During the supergene mineralization period, which belongs to the Yanshanian–Himalayan
stage, the primary ores near the ground surface were oxidized under the supergene geologi-
cal process, forming limonite and reddening associated with regional collisional orogenesis
and plateau uplift, and the rocks are of allotriomorphic texture, metasomatic relict texture,
and alveolate texture [67,76,77].

3. Methods
3.1. Convolutional Neural Network
Convolutional Neural Networks (CNN) [78], as widely used algorithms in deep
learning, effectively learn feature patterns from sample data. Leveraging multi-layer
convolution operations, they automatically extract feature information from input data,
revealing intricate patterns within the data. In a convolutional neural network, apart from
the input and output layers, the hidden layers include convolutional layers, activation
function layers, pooling layers, and fully connected layers.
The convolutional layer is the core, reflecting the local connectivity and weight-sharing
characteristics of convolutional neural networks. It uses filters to extract local features from
the overall data, and then, by shifting with a certain stride, extracts data features from
different positions, resulting in a feature map [79].

Oi,j = ∑u,v Ii+u,j+v × Ku,v (1)

where Oi,j is an element of the output feature map, I is the input image, K is the convolution
kernel, and u and v are the spatial indices of the convolution kernel.
After the convolution operation, a non-linear activation function is introduced. This
article employs the ReLU function [80], which not only ensures training effectiveness but
also accelerates the training speed. The calculation formula is as shown in Equation (2).

x, if x≥0
f (x) = (2)
0, if x<0
  x, if x≥0
f ( x) =  (2)
0, if x<0
The pooling layer serves to gradually reduce spatial dimensions, thereby reducing
Minerals 2023, 13, 1205 7 of 22
the number of parameters and computational demands while simultaneously increasing
the receptive field [81]. The most common polling operation is maximum pooling [82].
Within the fully connected layer, each neuron maintains connections to all preceding
The pooling layer serves to gradually reduce spatial dimensions, thereby reducing the
layer neurons. These layers
number are responsible
of parameters for tasks such
and computational as classification
demands or regression,
while simultaneously increasing the
making use of universally extracted features [83].
receptive field [81]. The most common polling operation is maximum pooling [82].
O =σ W × I + b ( )
Within the fully connected layer, each neuron maintains connections to all preceding
(3)
layer neurons. These layers are responsible for tasks such as classification or regression,
where 𝑊 is the weight
makingmatrix, 𝐼 is the input,
use of universally 𝑏 is features
extracted vector, and 𝜎 is an activation
the bias[83].
function such as Softmax.
O = σ (W × I + b ) (3)
In this study, the Lenet-5 CNN consisting of seven network structure layers was uti-
lized [84]. The input
where layer
W is followed
is the weightby two combinations
matrix, I is the input, bofisconvolutional
the bias vector,and
andpooling
σ is an activation
layers, which thenfunction
pass through
such as the transposed convolutional layers. Finally, the Softmax
Softmax.
In this study,
function is used to normalize the Lenet-5
the output layer.CNN consisting
Within of seven
this model, thenetwork structureislayers was
ReLU function
utilized [84]. The input layer is followed by two combinations
employed as the excitation function of the convolution layer in this model. of convolutional and pooling
layers, which then pass through the transposed convolutional layers. Finally, the Softmax
function is used to normalize the output layer. Within this model, the ReLU function is
3.2. The 3D-CNN Model Architecture
employed as the excitation function of the convolution layer in this model.
Three-dimensional convolutional neural networks have the unique advantage of
3.2. The 3D-CNN
learning spatial positional Model Architecture
relationships among data [85–87]. In the context of three-di-
mensional ore depositThree-dimensional
prediction tasks,convolutional
the three-dimensional spatial
neural networks haveinformation inher- of learn-
the unique advantage
ing spatial positional relationships among data [85–87]. In the
ently contains mineralization location relationships, making three-dimensional convolu-context of three-dimensional
ore deposit
tional neural networks prediction
particularly tasks, the three-dimensional
advantageous for such tasks. spatial information inherently con-
The networktains mineralization
architecture location
designed inrelationships, making three-dimensional
this study comprises convolutional neural
five main components
networks particularly advantageous for such tasks.
(Figure 3), the input layer, three-dimensional convolutional layers (Conv), pooling layers,
The network architecture designed in this study comprises five main components
three-dimensional(Figure
transposed
3), theconvolutional layers (TConv), convolutional
input layer, three-dimensional and the output layer.
layers Among
(Conv), pooling lay-
these, Conv, pooling
ers, three-dimensional transposed convolutional layers (TConv), and multi-
layers, and TConv collectively constitute the hidden layers of a the output layer.
layer perceptron. Among
To address
these,the goalpooling
Conv, of ore layers,
depositand
prediction, the network
TConv collectively outputthe
constitute layer
hidden layers
is modified to output
of a the posterior
multi-layer probabilities
perceptron. of mineralization
To address the goal of in
orea deposit
binary classification
prediction, the network
form. output layer is modified to output the posterior probabilities of mineralization in a binary
classification form.

Figure 3. DesigningFigure
the 3D3.convolutional neural
Designing the 3D network neural
convolutional (modified from
network reference
(modified [88]).
from reference [88]).

4. Dataset and 3D4.Modeling

Dataset and 3D Modeling
4.1. Dataset
4.1. Dataset
Before the establishment of the 3D geological model of the Zaozigou gold deposit,
Before the establishment
the collection of
of the 3D geological
historical geologicalmodel of the Zaozigou
and geochemical gold
data was deposit, including
completed,
the collection of historical
geological reports, geological exploration maps, drills geochemical data, etc.ge-
geological and geochemical data was completed, including (Table 2).
ological reports, geological exploration maps, drills geochemical data, etc. (Table 2).
Table 2. Dataset of Zaozigou gold deposit.

Category Description
Geological map Topographic-geological map with the scale of 1:2000
Horizontal section maps of elevations of 2450 m, 2530 m, 3610 m, 2850 m, 3010 m, 3050 m, 3090 m,
Horizontal section map
and 3170 m
 Table 2. Dataset of Zaozigou gold deposit.

Category Description
Geological
Minerals map
2023, 13, 1205 Topographic-geological map with the scale of 1:2000 8 of 22

Horizontal section Horizontal section maps of elevations of 2450 m, 2530 m, 3610 m, 2850 m, 3010 m, 3050 m, 3090
map m, and 3170 m
Table 2. Cont.geological profiles of exploration lines of 54#, 58#, 62#, 66#, 70#, 74#, 78#, 81#,
NE-orientation
Category 89#, 93#, 96#, 98#, 100#, 102#, 106#, 110#, 114#, 118#, 120#, 128#
Description
Geological profile NW-orientation geological
NE-orientation geological profiles
profiles of exploration
of exploration lines
lines of 54#, 58#,of62#,
192#,
66#,196#, 199#,
70#, 74#, 78#, 200#, 202#,
81#, 89#, 93#, 204#,
96#, 98#, 100#, 102#,208#,
106#, 212#, 216#118#, 120#, 128#
110#, 114#,
Geological profile
NW-orientation
N–S-orientation geological
geological profiles of
profiles of exploration
exploration lines of 192#,
lines 196#, 287#,
of 283#, 199#, 200#,
291#,202#,
292#,204#, 208#,
295#, 298#, 302#
212#, 216#
Drilling log 72 drillings
N–S-orientation geological profiles of exploration lines of 283#, 287#, 291#, 292#, 295#, 298#, 302#
Geochemical data
Drilling log 5028 samples; 12 elements of Ag, 72 As,drillings
Au, Cu, Hg, Pb, Zn, Sb, W, Bi, Co, and Mo
Geochemical data 5028 samples; 12 elements of Ag, As, Au, Cu, Hg, Pb, Zn, Sb, W, Bi, Co, and Mo
4.2. The 3D Geological Modeling
4.2. The 3D Geological Modeling
The commencement of three-dimensional geological modeling is initiated by the crit-
ical The
stepcommencement
of standardizing of the
three-dimensional
coordinates of the geological modeling
collected data. Theis initiated
subsequentby the
phase in-
critical step of standardizing the coordinates of the collected data.
volves the importation of the standardized data into the Micromine software The subsequent phase
platform,
involves the importation of the standardized data into the Micromine software platform,
marking the beginning of the modeling process.
marking the beginning of the modeling process.
Drillhole data represent the most direct and accurate source of deep-seated geologi-
Drillhole data represent the most direct and accurate source of deep-seated geological
cal information.
information. In this In study,
this study,
a totala of
total
157ofdrillhole
157 drillhole datasets
datasets were collected.
were collected. By importing
By importing
drillholelocation
drillhole location data,
data, drillhole
drillhole deviational
deviational surveysurvey data, sampling
data, sampling records,records, and lithologi-
and lithological
cal data, a comprehensive drillhole model is constructed, as illustrated
data, a comprehensive drillhole model is constructed, as illustrated in Figure 4. in Figure 4.

Figure 4. The 3D model of drillings in Zaozigou gold deposit. (a) Plan view of drillings distribution.
Figure 4. The 3D model of drillings in Zaozigou gold deposit. (a) Plan view of drillings distribution.
(b) Side view of drillings distribution. (c) Sample grade distribution of drillings.
(b) Side view of drillings distribution. (c) Sample grade distribution of drillings.

Subsequently,
Subsequently, thethe prospecting
prospecting lineline section
section underwent
underwent coordinate
coordinate transformation
transformation to to
ensurethethe
ensure alignment
alignment of drillholes
of drillholes on the on the exploration
exploration cross-sections
cross-sections with the
with the drillhole drillhole
model.
model.
This Thisinvolved
process processconnecting
involved connecting theboundary
the geological geological boundary
lines lines corresponding
corresponding to adjacent to
cross-sections, enabling theenabling
adjacent cross-sections, construction
theofconstruction
orebody models, magmaticmodels,
of orebody rock models, and rock
magmatic
fracture
models,models.
and fracture models.
The orebody model accurately depicts the shape and attitude of the orebody. At the
Zaozigou gold deposit, the orebody trends northeast, with a steep inclination ranging
from 50◦ to 70◦ , and in some localized areas, it is even vertical. Certain segments of the
steeply dipping orebody are intersected and displaced by ore-controlling faults associated
Minerals 2023, 13, x FOR PEER REVIEW 9 of 22

The orebody model accurately depicts the shape and attitude of the orebody. At the
Minerals 2023, 13, 1205 Zaozigou gold deposit, the orebody trends northeast, with a steep inclination ranging
9 of 22
from 50° to 70°, and in some localized areas, it is even vertical. Certain segments of the
steeply dipping orebody are intersected and displaced by ore-controlling faults associated
withgently
with gentlydipping
dipping orebodies
orebodies (Figure
(Figure 5). This
5). This information
information provides
provides a valuable
a valuable range for
range for
selecting positive samples in subsequent ore prediction.
selecting positive samples in subsequent ore prediction.

Figure5.5.Zaozigou
Figure Zaozigou orebody
orebody model.
model. (a) Plan
(a) Plan view.
view. (b,c)(b,c)
Side Side
view.view.

During
During thethe
formation process
formation of mineral
process deposits,
of mineral fractures
deposits, typicallytypically
fractures play a pivotal
playrole
a pivotal
as they
role asprovide pathways
they provide for magmatic
pathways hydrothermal
for magmatic fluids, contributing
hydrothermal fluids, to the enrichment
contributing to the en-
ofrichment
valuable ofminerals.
valuable Atminerals.
the Zaozigou gold
At the deposit, these
Zaozigou goldfracture
deposit,structures can be structures
these fracture broadly can
categorized into five main groups: northwest-trending, north–south-trending,
be broadly categorized into five main groups: northwest-trending, north–south-trending, northeast-
trending, nearly east–west-trending, and north–northeast-trending. Through an analysis of
northeast-trending, nearly east–west-trending, and north–northeast-trending. Through
the relationship between the orebody and these structural features, we have determined
an analysis of the relationship between the orebody and these structural features, we have
that fractures have their most significant impact on mineralization within an approximate
determined
radius of 30 mthat
[77].fractures
Therefore,have theirdelineated
we have most significant impact
a 30 m buffer zone onaround
mineralization within an
these fracture
approximate radius of 30 m [77]. Therefore, we have delineated a
structures as one of the elements for subsequent mineralization prediction (Figure 6). 30 m buffer zone around
theseThefracture
Zaozigou structures as one
gold deposit of features
area the elements for subsequent
the development mineralization
of various rock veins,prediction
pri-
(Figurecomposed
marily 6). of diorite porphyrite, granite diorite porphyrite, and quartz diorite
The Zaozigou
porphyrite. gold deposit
Diorite porphyrite veins area featuresless
are relatively the abundant
development of predominantly
and are various rock veins,
ex- pri-
posed in the northwest, extending deeper towards the southeast with significant
marily composed of diorite porphyrite, granite diorite porphyrite, and quartz diorite por- variations
inphyrite.
dip angles. In the
Diorite central partveins
porphyrite of theare
granite dioriteless
relatively porphyrite
abundant veinand
zone,
arethe orientation
predominantly ex-
isposed
northeastward, dipping towards the southwest at angles ranging from
in the northwest, extending deeper towards the southeast with significant varia-approximately
30 ◦ to 60◦ . Quartz diorite porphyrite veins are concentrated and extend along a north-
tions in dip angles. In the central part of◦ the granite diorite porphyrite vein zone, the ori-
east direction with orientations between 40 and 60◦ . Mineralization is closely associated
entation is northeastward, dipping towards the southwest at angles ranging from approx-
with intermediate rock veins, particularly quartz diorite porphyrite, establishing it as a
imately 30° to 60°. Quartz diorite porphyrite veins are concentrated and extend along a
significant geological feature related to mineralization in the mining area (Figure 7).
northeast direction with orientations between 40° and 60°. Mineralization is closely
Minerals 2023, 13, x FOR PEER REVIEW 10 of 22

associated with intermediate rock veins, particularly quartz diorite porphyrite, establish-
Minerals 2023, 13, 1205 10 of 22
ing it as a significant geological feature related to mineralization in the mining area (Figure
7).

Minerals 2023, 13, x FOR PEER REVIEW 2 of 14

Figure6.6.The
Figure The3D3D
fracture surface
fracture and 3D
surface andfracture model.model.
3D fracture (a) Plan(a)
view. (b,c)
Plan Side(b,c)
view. view.Side view.

Figure 7.
Figure 7. Intrusive
Intrusive model
model of
of Zaozigou
Zaozigou gold
gold deposit.
deposit. (a)
(a) Plan
Plan view.
view. (b,c)
(b,c) Side
Side view.
view.
Figure 7. Intrusive model of Zaozigou gold deposit. (a) Plan view. (b,c) Side view.

4.3.The
4.3. The3D
3DGeochemical
GeochemicalModeling
Modeling
Theprimary
The primarygeochemical
geochemicalhalo
halodata
datafrom
fromthe
thedrilling
drillingcore
coreare
arecollected
collectedfrom
fromthe
the
“Zaozigougold
“Zaozigou goldsuccessive
successiveresources
resourcesexploration
explorationproject
projectininHezuo
Hezuocity,
city,Gansu
GansuProvince”,
Province”,
withaatotal
with totalof
of72
72 drillings
drillings and 5028 samples.
samples. The
The1212elements
elementsofofAg, As,
Ag, Au,
As, Cu,Cu,
Au, Hg,Hg,
Pb,
Zn, Sb, W, Bi, Co, and Mo were analyzed. The analytical methods were plasma mass
spectrometry, atomic absorption, emission spectrometry, atomic fluorescence, and X-ray
fluorescence spectrometry [70].
 trometry, atomic absorption, emission spectrometry, atomic fluorescence, and X-ray fluo-
rescence spectrometry [70].

4.3.1. Descriptive Statistical Analysis

Descriptive
Minerals 2023, 13, 1205 statistical analysis was carried out on the primary geochemical halo data 11 of 22
from drilling the core of the Zaozigou gold deposit (Table 3). In terms of elemental enrich-
ment (K) (Figure 8a), the enrichment coefficients for the Au, As, and Sb elements were
greater than 100 and standard
Pb, Zn, deviations
Sb, W, Bi, Co, and Mowere greater
were thanThe
analyzed. 1000, reflecting
analytical the high
methods werede-
plasma mass
gree of enrichment spectrometry, atomic absorption,
of these elements in the area.emission
In terms spectrometry, atomic fluorescence,
of the coefficient of variation and X-ray
fluorescence
(CV) (Figure 8b), elements spectrometry
with coefficients[70].of variation greater than 5 are Au, Hg, Sb, and
Bi, indicating that 4.3.1.
the distribution in the mine area is extremely heterogeneous with large
Descriptive Statistical Analysis
fluctuations and is more prone to aggregation and mineralization potential. Elements with
Descriptive statistical analysis was carried out on the primary geochemical halo
strong coefficientsdata of variation (greater
from drilling thanof1the
the core andZaozigou
less thangold
5) are Ag, As,
deposit Cu,3).Mo,
(Table and W,
In terms of elemental
indicating that theyenrichment (K) (Figure 8a), the enrichment coefficients for the Au, As,enrich-
are unevenly distributed in space and have some potential for and Sb elements
ment. were greater than 100 and standard deviations were greater than 1000, reflecting the high
In summary, degree
the standard deviations,
of enrichment enrichment
of these elements in coefficients,
the area. In and
termscoefficients of var-of variation
of the coefficient
(CV) (Figure 8b), elements with coefficients of variation
iation for Au and Sb elements are characterized by large dispersions and strong enrich- greater than 5 are Au, Hg, Sb,
and Bi, indicating
ment coefficients and degrees of variation. that the distribution in the mine area is extremely heterogeneous with
large fluctuations and is more prone to aggregation and mineralization potential. Elements
Table 3. Descriptivewith stronganalysis
statistical coefficients of variation
of drilling (greater
core data than 1 Gold
of Zaozigou and less than 5) are Ag, As, Cu, Mo,
Mine.
and W, indicating that they are unevenly distributed in space and have some potential
Minimum Mean Median Maximum for enrichment.
Standard Deviation Anomaly Threshold K CV
0.15 202.53 11 35,300 1064.67 2331.87 238.27 5.26
Table 3. Descriptive statistical analysis of drilling core data of Zaozigou Gold Mine.
0.5 639.1 95.6 25,140 1677.3 3993.6 145.25 2.62
0.05 Element
391.12 Minimum
21.4 Mean 130,300 Median 4329.22 Standard Deviation
Maximum 9049.57 1150.35 11.07
Anomaly Threshold K CV
0 Au 58 0.1523 202.5394,931 11 1397
35,300 1064.67 2852 2331.87 4.86 23.96
238.27 5.26
0.5 As34.1 0.5
28.6 639.13418 95.6 25,140
67.4 1677.3 168.8 3993.6 2 145.25
1.98 2.62
Sb 0.05 391.12 21.4 130,300 4329.22 9049.57 1150.35 11.07
0.8 Hg22.8 020.3 58 381.8 23 16.6
94,931 1397 56.1 2852 1.2 0.73
4.86 23.96
1.9 Cu71.3 74.1
0.5 34.1 317 28.6 34.8
3418 67.4 140.8 168.8 1.05 0.49
2 1.98
Pb 0.8 22.8 20.3 381.8 16.6 56.1 1.2 0.73
0.1 24.5 17.5 142.6 22.5 69.5 2.45 0.92
Zn 1.9 71.3 74.1 317 34.8 140.8 1.05 0.49
0.007 Co0.136 0.088
0.1 24.59.795 17.5 0.272
142.6 22.5 0.68 69.5 2.27 2
2.45 0.92
0.01 Ag1.75 0.007
0.44 0.136304.05 0.088 9.795
7.3 0.272 16.34 0.68 2.82 2.27
4.18 2
Mo 0.01 1.75 0.44 304.05 7.3 16.34 2.82 4.18
0.05 W
17.75 0.057.5 1639.75 7.5
17.75
51.48
1639.75 51.48
120.71 120.71
18.3 2.9
18.3 2.9
0 Bi1.29 00.46 1.29337.5 0.46 7.49
337.5 7.49 16.27 16.27 7.14 5.83
7.14 5.83
Note: The units of Ag, Au,
Note: and
The Hg
units of is
Ag,10
Au,; and
−9 the rest of −the elements
Hg is 10 9 ; the rest of theare in units
elements ofunits
are in 10−6of
. 10−6 .

Figure 8. Enrichment coefficient

Figure and variation
8. Enrichment coefficient
coefficient diagram
and variation of Zaozigou
coefficient diagramgold deposit.gold deposit.
of Zaozigou

In summary, the standard deviations, enrichment coefficients, and coefficients of vari-

ation for Au and Sb elements are characterized by large dispersions and strong enrichment
coefficients and degrees of variation.

4.3.2. Defining the Modeling Range of 3D Data Volume

The geochemistry-based prediction method is usually constricted by the controlled
width and depth of prospecting engineering, such as drillings. Because the deep orebodies
Minerals 2023, 13, x FOR PEER REVIEW 12 of 22

4.3.2. Defining the Modeling Range of 3D Data Voludme

Minerals 2023, 13, 1205 The geochemistry-based prediction method is usually constricted by the controlled 12 of 22
width and depth of prospecting engineering, such as drillings. Because the deep orebodies
of the Zaozigou gold deposit are obviously controlled by fractures, this study constructs
the 3D primary geochemical halo data volume model within a fracture buffer of 30 m so
of the Zaozigou gold deposit are obviously controlled by fractures, this study constructs
that
the 3Disolated drilling
primary at large depths
geochemical halo datahasvolume
reliablemodel
prediction
withinspace.
a fracture buffer of 30 m so
that isolated drilling at large depths has reliable prediction space. engineering, and sam-
Therefore, based on the geological background, prospecting
plingTherefore,
intervals, based
the cube sizegeological
on the was determined as 10 prospecting
background, m × 10 m × engineering,
10 m for lengthand×sampling
width ×
height. A total of 314,788 cubes were filled in the modeling range (Figure 9).
intervals, the cube size was determined as 10 m × 10 m × 10 m for length × width × height. The spatially
interpolated cubes
A total of 314,788 willwere
cubes carryfilled
geological and geochemical
in the modeling data,9).asThe
range (Figure well as inferred
spatially metal-
interpolated
logenic
cubes willinformation, providing
carry geological a 3D visualization
and geochemical carrier
data, as well for further
as inferred 3D mineral
metallogenic prospec-
information,
tivity mapping.
providing a 3D visualization carrier for further 3D mineral prospectivity mapping.

Figure
Figure 9.
9. Buffer
Bufferzone
zoneofof3030mmdistance from
distance fracture
from of of
fracture Zaozigou gold
Zaozigou deposit.
gold (a) Buffer
deposit. zone.
(a) Buffer (b)
zone.
Buffer zone superimposed by orebodies.
(b) Buffer zone superimposed by orebodies.

4.3.3. Spatial
4.3.3. Spatial Interpolation
Interpolation MethodMethod of of Primary
Primary Geochemical Halo Data
This study adopts
This adoptsthe theordinary
ordinarykriging
kriging interpolation
interpolation method
method to construct
to constructa 3Da primary
3D pri-
geochemical halo data volume
mary geochemical halo data volume model. model.
The steps
The steps ofof the
the spatial
spatial interpolation
interpolation ofof the
the elemental
elemental content
content mainly
mainly include:
include: (i)
(i) data
data
pre-processing, including statistical analysis and the normalization of
pre-processing, including statistical analysis and the normalization of data; (ii) variogramdata; (ii) variogram
calculation and
calculation and fitting,
fitting, including
including calculating
calculating thethe parameters
parameters of of the
the variogram
variogram and and using
using aa
spherical model
spherical model to to fit
fitthe
thevariogram;
variogram;and and(iii)
(iii)3D3Dkriging
kriginginterpolation,
interpolation, including
including establish-
estab-
ing a 3D spatial search ellipsoid, calculating the coefficient matrix and
lishing a 3D spatial search ellipsoid, calculating the coefficient matrix and distance vector,distance vector, and
employing the ordinary kriging method to gain the spatial interpolation
and employing the ordinary kriging method to gain the spatial interpolation results [89]. results [89]. Finally,
a 3D primary
Finally, geochemical
a 3D primary halo data
geochemical volume
halo model was
data volume modelobtained to provide
was obtained geochemical
to provide geo-
evidence for deep mineral prediction.
chemical evidence for deep mineral prediction.
Practically, the
Practically, spatial distribution
the spatial distribution of the elements
of the elements in in the
the depths
depths of of the
the Zaozigou
Zaozigou gold gold
deposit is
deposit is anisotropic,
anisotropic,with withaaclear
clearenrichment
enrichment pattern
patternalong with
along thethe
with fracture. TheThe
fracture. orebody
ore-
occurrences are nearly consistent with the fracture occurrences. Therefore,
body occurrences are nearly consistent with the fracture occurrences. Therefore, the inter- the interpolated
ellipsoidellipsoid
polated specification was determined
specification by combining
was determined the distribution
by combining the distributionof fractures and
of fractures
orebodies. Firstly, data pre-processing is necessary for primary geochemical halo data
and orebodies. Firstly, data pre-processing is necessary for primary geochemical halo data
analysis, because extreme values can greatly influence the element distribution regularity
analysis, because extreme values can greatly influence the element distribution regularity
extraction. Taking three times the mean value as a threshold value, the extreme values are
extraction. Taking three times the mean value as a threshold value, the extreme values are
replaced by the threshold. Secondly, the determination of the main and secondary range
replaced by the threshold. Secondly, the determination of the main and secondary range
orientations is critical for the accuracy of the kriging interpolation results. The direction
orientations is critical for the accuracy of the kriging interpolation results. The direction
of the Main orientation is aligned with the orientation of the orebody, while the direction
of the Secondary orientation corresponds to the dip angle of the orebody. The Vertical
orientation is perpendicular to the plane formed by the Main orientation and the Secondary
orientation (Table 4; Figure 10).
Minerals 2023, 13, x FOR PEER REVIEW 13 of 22

Minerals 2023, 13, 1205 13 of 22

of the Main orientation is aligned with the orientation of the orebody, while the direction
of the Secondary orientation corresponds to the dip angle of the orebody. The Vertical
orientation is perpendicular
Table 4. Parameters to thevariogram
of experimental plane formed by therange
in different Mainorientations.
orientation and the Second-
ary orientation (Table 4; Figure 10).
Orientation Lag (m) Angular Tolerance Number of Point Pairs Nugget Cumulative Sill Range (m)
Main Table 4. Parameters of experimental variogram in different range orientations.
10 22.5◦ 60 1.65 1.89 230
orientation
Orientation
Secondary Lag (m) Angular Tolerance Number of Point Pairs Nugget Cumulative Sill Range (m)
10 22.5◦ 60 2.26 1.87 220
orientation
Main orientation 10 22.5° 60 1.65 1.89 230
Vertical orientation
Secondary 10 ◦ 22.5° 60 2.26 1.87 220
10 22.5 60 2.12 1.88 110
orientation
Vertical orientation 10 22.5° 60 2.12 1.88 110

Figure
Figure 10.
10. Variogram
Variogramfitting
fittingand search
and ellipsoid
search designing.
ellipsoid (a) Main
designing. range
(a) Main orientation
range fitting.
orientation (b)
fitting.
Secondary range orientation fitting. (c) Vertical range orientation fitting; (d) Search ellipsoid.
(b) Secondary range orientation fitting. (c) Vertical range orientation fitting; (d) Search ellipsoid.

4.3.4. The
The 3D
3D Geochemical
Geochemical Data
Data Volume Model and Element Spatial Distribution Patterns
Patterns
The results of the 3D spatial interpolation of 12 elements provide an intuitional pre-
The results
sentation of the distribution
of the element 3D spatial interpolation
pattern (Figure of11).
12 elements provide
The variation an intuitional
of elemental content
presentation of the
values in the data element
volume distribution
model pattern
is represented (Figurecolors
by 9-level 11). The variation
according of cumulative
to the elemental
content
frequency values in natural
of the the datalogarithm
volume modelof the is represented
element contentbydata
9-level colors
(Table 5). according to the
cumulative frequency of the natural logarithm of the element content data (Table 5).
5. Cumulative
TableThe frequencythat
analysis indicates of the
thenatural logarithm
indices of elementAu,
of the elements content data.Sb increase pro-
As, and
gressively from the surface to the depth, and their change patterns are similar. This sug-
Element gests that these elements Cumulative
are associatedFrequency
with gold mineralization, and their enrichment
5% 15% 25% 40% 60% 75% 85% 95% 100%
capacity increases with depth. Hg has two strong anomalies at elevations 1700 m and 3200
ln(Au) −0.11 0.69 1.19
m. As a low-temperature 1.93element 3.00that is easy to4.22
diffuse, Hg 5.27 6.89
usually appears 11.24
on the head
ln(As) 2.23 2.91 3.38 The enrichment
of the orebody. 4.04 5.14 m reflects
at 3100 6.07the orebody
6.91 below, while
8.09 enrichment
10.13 at
ln(Sb) 0.26 1.17 1.80 2.57 3.43 4.07 4.66 6.02 11.78
1700 m may indicate potential orebodies at deeper locations. Co and Mo anomalies are
ln(Hg) 2.18 2.53 2.73 2.97 3.29 3.61 3.94 4.63 11.46
ln(Cu) 2.08 found at 2300
2.54 2.86 m, indicating
3.21 the tail3.49of orebodies. W and Bi 3.86
3.67 have similar4.28
anomaly distribu-
8.14
ln(Pb) 2.01 tion, and have
2.51 2.73 negative correlation3.10
2.91 with Au, As, and Sb (Figure
3.25 3.38 11). 3.74 5.94
ln(Zn) 2.69 3.33 Based3.94 on the above 4.20analysis, and
4.45 comparing 4.57 the spatial
4.65distribution
4.77of the elements
5.76
ln(Co) 2.07 2.32 2.53
with the spatial 2.75 of the orebodies,
position 2.91 it3.03
can be seen 3.35 4.42 halo element
that the front 4.96
ln(Ag) −2.92 −2.76 −2.66 −2.54 −2.33 −2.12 −1.97 −1.05 2.28
ln(Mo) −1.61 −1.61 −1.61 −1.11 −0.56 0.00 0.69 1.89 5.72
ln(W) 0.41 0.83 1.12 1.59 2.29 2.83 3.26 4.03 7.40
ln(Bi) −2.18 −1.54 −1.22 −0.92 −0.63 −0.26 0.19 1.20 5.82
 Minerals 2023, 13, x FOR PEER REVIEW 14

Minerals 2023, 13, 1205

association of the Zaozigou gold deposit is As-Sb-Hg; the near-ore halo element as
14 of 22
tion is Au-Ag-Cu-Pb-Zn; and the tail halo element association is W-Mo-Bi.

Figure
Figure 11. The 3D 11. The 3D
geochemical geochemical
model model of elements.
of elements.
The analysis indicates that the indices of the elements Au, As, and Sb increase progres-
sively from the surface to the depth, and their change patterns are similar. This suggests
that these elements are associated with gold mineralization, and their enrichment capacity
increases with depth. Hg has two strong anomalies at elevations 1700 m and 3200 m. As
a low-temperature element that is easy to diffuse, Hg usually appears on the head of the
orebody. The enrichment at 3100 m reflects the orebody below, while enrichment at 1700 m
 ln(Sb) 0.26 1.17 1.80 2.57 3.43 4.07 4.66 6.02
ln(Hg) 2.18 2.53 2.73 2.97 3.29 3.61 3.94 4.63
ln(Cu) 2.08 2.54 2.86 3.21 3.49 3.67 3.86 4.28
ln(Pb) 2.01 2.51 2.73 2.91 3.10 3.25 3.38 3.74
Minerals
ln(Zn) 2023, 13, 12052.69 3.33 3.94 4.20 4.45 4.57 4.65 15 of 22
4.77
ln(Co) 2.07 2.32 2.53 2.75 2.91 3.03 3.35 4.42
ln(Ag) −2.92 −2.76 −2.66 −2.54 −2.33 −2.12 −1.97 −1.05
may indicate potential orebodies at deeper locations. Co and Mo anomalies are found at
ln(Mo) −1.61 −1.61
2300 −1.61
m, indicating the tail of−1.11
orebodies. W−0.56 0.00 anomaly
and Bi have similar 0.69
distribution,1.89
and
ln(W) 0.41 0.83negative correlation
have 1.12 1.59
with Au, As, and2.29 2.83
Sb (Figure 11). 3.26 4.03
Based on the above analysis, and comparing the spatial distribution of the elements
ln(Bi) −2.18 −1.54 −1.22 −0.92 −0.63 −0.26 0.19 1.20
with the spatial position of the orebodies, it can be seen that the front halo element associ-
ation of the Zaozigou gold deposit is As-Sb-Hg; the near-ore halo element association is
5. Results and and
Au-Ag-Cu-Pb-Zn; Discussion
the tail halo element association is W-Mo-Bi.
Building
5. Results upon the geological entity model and the geochemical orebody mo
and Discussion
tioned above,
Building uponthis
theprediction primarily
geological entity focuses
model and on deep-seated
the geochemical orebodyforecasting
model men- within
ing area,
tioned above,covering an approximate
this prediction area ofon2.16
primarily focuses km . Theforecasting
2
deep-seated predictedwithin
depththeis determ
2 . The predicted depth is determined
referencing the maximum depth borehole (SDZK8314) within the mining area, re
mining area, covering an approximate area of 2.16 km
by referencing
depth the maximum
of 2001.15 m. Thedepth borehole
predicted (SDZK8314)
depth is set within
at 3000themmining
below area,
thereach-
surface. Th
ing a depth of 2001.15 m. The predicted depth is set at 3000 m below the surface. The
dimensional prediction scope is illustrated in Figure 12.
three-dimensional prediction scope is illustrated in Figure 12.

Figure
Figure 12.12. Three-dimensional
Three-dimensional prediction
prediction scope. scope.

5.1. Geological Geochemical Prospecting Model

5.1.Mineral
Geological Geochemical Prospecting Model
prospecting models are comprehensive representations in mineralization
Mineral
prediction prospecting
practices models
that encompass arereflecting
features comprehensive representations
potential indicators in miner
of orebodies,
deposits, mining districts, and even mineralization belts. These
prediction practices that encompass features reflecting potential models are indicators
typically of or
summarized using forms such as text, illustrations, and tables. Establishing mineral
deposits, mining districts, and even mineralization belts. These models are
prospecting models is not only a fundamental aspect of scientific mineral exploration, but
also an effective approach to discovering latent ore deposits. They hold significant guiding
significance in ore prediction [90].
In the Zaozigou gold mining region, orebodies are primarily influenced by faulting and
vein intrusions. Consequently, three-dimensional fracture models and three-dimensional
Minerals 2023, 13, 1205 16 of 22

rock body models can serve as indicators for mineral prospecting. The distribution, zoning,
and combinations of geochemical elements serve as favorable indicators for mineralization.
They can reveal the sources of materials during ore deposit formation, the mineraliza-
tion process, and the mineralization environment, thereby aiding in the identification of
prospective mineralization areas.
Three-dimensional mineralization prediction involves establishing quantitative indica-
tors for favorable ore-controlling factors (mineralization conditions) under the guidance of
mineral prospecting models. This process, coupled with solid and block models, facilitates
the extraction of three-dimensional cubic representations for various favorable mineraliza-
tion conditions. This study employs a rock body model, a structural buffer zone model,
and a combination of the geochemical elements, Au, As, and Sb, along with elements from
the proximal halo, as well as the geochemical parameters (ratio of front halo to distal halo)
as the predictive factors for mineralization. In total, seven variables (the three-dimensional
model of Quartz diorite porphyrite, the 30 m structural buffer zone model, Au, As, and Sb,
the combination of Au-Ag-Cu-Pb-Zn in the near-ore halo, the ratio of front halo elements to
tail halo elements (As-Sb-Hg)/(W-Bi-Co-Mo)) are utilized for the mineralization prediction
(Table 6).

Table 6. Prospecting prediction elements.

Predictive Factors Description of Variables Variables

Concealed intermediate acidic intrusion,
The three-dimensional model
Magmatic rocks quartz diorite (porphyrite), high
of quartz diorite porphyrite
Geology potassium calcium alkaline series
Fault surface and fracture surface; The 30 m structural buffer
Structures
lithologic interface zone model
Au, As, and Sb combination; Au content
is above 15 × 10−9 , Sb content is above
Element Combinations Au, As, and Sb
30 × 10−6 , which are direct geochemical
Geochemistry indicators for gold exploration
As-Sb-Hg (front halo) →
Au-Ag-Cu-Pb-Zn
Primary Halo Au-Ag-Cu-Pb-Zn (near-ore halo) →
(As-Sb-Hg)/(W-Bi-Co-Mo)
W-Bi-Co-Mo (tail halo)

5.2. Convolutional-Neural-Network-Based Ore Deposit Prediction

Prior to the incorporation of the 3D-CNN model, these seven variables underwent a trans-
formation into cubic representations, with each small cube measuring 10 m × 10 m × 10 m.
Binary coding was independently applied to the rock body model and the structural buffer
zone model. Specifically, if a small cube fell within the mineralized rock body or fracture
buffer zone, it received a binary value of 1; conversely, if it lay outside these areas, it received
a binary value of 0. The geochemical parameters were seamlessly integrated by attributing
their content values directly to the corresponding small cubes. In essence, each small cube
encapsulates these seven attributes, facilitating a comprehensive mineralization prediction.
Next, regarding the selection of positive and negative samples, positive samples
were chosen from the small cubes within the scope of the orebody model as described in
Section 4.2. Negative samples, on the other hand, were randomly selected in positions
within the drilling model where no orebody was present, matching the quantity of positive
samples. Together, these positive and negative samples formed the training dataset.
The convolutional neural network employs a binary cross-entropy loss function [91]
for binary classification, aimed at balancing the loss between positive and negative samples
while smoothly adjusting network parameters. The optimizer chosen is Adam [92] with a
learning rate of 0.1 and trained over 100 epochs. The resulting binary classification loss and
accuracy are depicted in Figure 13a.
 Together, these positive and negative samples formed the training dataset.
The convolutional neural network employs a binary cross-entropy loss function [91]
for binary classification, aimed at balancing the loss between positive and negative
samples while smoothly adjusting network parameters. The optimizer chosen is Adam
Minerals 2023, 13, 1205 17 of 22
[92] with a learning rate of 0.1 and trained over 100 epochs. The resulting binary
classification loss and accuracy are depicted in Figure 13a.

Figure 13. (a) Training process

Figure 13. (a)binary classification
Training loss
process binary and accuracy;
classification (b) accuracy;
loss and ROC curve (b) graph.
ROC curve graph.

The parameterisconfiguration
The parameter configuration presented inisTable
presented in Table
7, where the7,number
where the
of number
channelsof channels
in the “Input”
in the “Input” signifies signifies
the count the count
of input of input
attributes, andattributes, and the
the number number ofin
of channels channels
the in the
“Output” represents the final prediction outcome denoting
“Output” represents the final prediction outcome denoting the confidence of mineralthe confidence of mineral
presence. The Sigmoid activation function is employed to constrain the output values
presence. The Sigmoid activation function is employed to constrain the output values
within the range of (0, 1), wherein values closer to 1 indicate higher confidence in the
within the rangepresence
of (0, 1),ofwherein values closer to 1 indicate higher confidence in the
minerals, while lower values imply reduced confidence.
presence of minerals, while lower values imply reduced confidence.
Table 7. Cumulative frequency of the natural logarithm of element content data.
Table 7. Cumulative frequency of the natural logarithm of element content data.
Architecture Feature Maps Kernel Size Strides Padding Channels
Architecture Feature Maps
Input Kernel
9, 9, 9 Size Strides Padding Channels 7
Input 9, 9,
Conv_19 5, 5, 5 3, 3, 3 2, 2, 2 same7 14
Conv_1 5, 5, 5
Conv_2 3,5,3,5 3
5, 2,3,2,3 2
3, same
2, 2, 2 14
same 21
Conv_2 TConv_1
5, 5, 5 5, 5, 5
3, 3, 3 3, 3, 3
2, 2, 2 1, 1,
same 1 valid
21 14
TConv_2 9, 9, 9 3, 3, 3 2, 2, 2 same 7
TConv_1 5, 5, 5
Output 3,9,3,9 3
9, 1,3,1,3 1
3, valid
1, 1, 1 14
same 1
TConv_2 9, 9, 9 3, 3, 3 2, 2, 2 same 7
Output 9,The
9, 9ROC curves depicted
3, 3, 3 in Figure1,13b
1, for
1 the trained
sameconvolutional
1 neural network
model on both the training and validation datasets reveal a notably robust fit, surpassing the
The ROC curves depicted
performance ofin Figure mixture
Gaussian 13b for models
the trained convolutional
and maximum entropyneural network
predictive models [93,94]
The construction of the convolutional neural network model,
model on both the training and validation datasets reveal a notably robust fit, surpassing precision assessment
the performance of Gaussian mixture models and maximum entropy predictive modelsunderscore
through training, and the ROC curves of the training and testing sets collectively
[93,94] the accuracy and reliability of the model. The visual representation of three-dimensional
ore deposit prediction, as illustrated in Figure 14, notably exhibits a high likelihood of
The construction of the convolutional neural network model, precision assessment
deep-seated mineralization, thereby suggesting an extension of mineral bodies into the
through training,deeper
and strata
the ROC curves oftwo
and delineating theprospective
training target
and testing
areas. sets collectively
underscore the accuracy and reliability of the model. The visual
Comparing the distribution of orebodies and fractures representation
(Figuresof5three-
and 6), a close
dimensional ore relationship
deposit prediction, as illustrated
can be found between themin Figure
and the14, notably
high exhibits
probability areaainhigh
Figure 14. As
likelihood of deep-seated mineralization,
for delineating the most likelythereby suggesting
metallogenic an at
position extension
depth, theofmineral
mineral exploration
targets are delineated by considering the factors of
bodies into the deeper strata and delineating two prospective target areas.the extension of orebodies, fractures,
and the result of CNN. Finally, two target areas are identified. Target I lies in the southeast
direction at an elevation of 1700–2000 m. This position represents the extension end of the
Au1 orebody. Notably, the Au and Sb elements show significantly strong anomalies at this
location. The ratio of elements between the front and distal halos has been consistently
rising. Convolutional neural network modeling has computed a relatively high probability
of mineralization in this region. Therefore, it is predicted that the Au1 orebody will extend
to depths beyond 1800 m or even give rise to new mineral bodies. Target II is situated in the
northwest direction within an elevation range of 2350–2650 m. This region exhibits a higher
level of structural complexity. There is a notable vertical overlap of the distal halo element
 Minerals 2023, 13, 1205 18 of 22

Co with the front halo elements Sb and As. Compared to the orebodies in the southeast
direction, the control over orebodies on the northwestern side is slightly less profound in
terms of depth. However, the deeper extension of the orebody has not been constrained,
OR PEER REVIEW particularly concerning stable and thick orebodies like Au9 and Au126. The18geochemical
of 22
parameters starting from an elevation of 2600 m and deeper suggest the potential presence
of further extensions or even new blind orebodies at greater depths.

Figure 14. Deep mineral prediction

Figure map based
14. Deep mineral on CNN.
prediction (a) Results
map based on CNN.without deepest
(a) Results withoutdrilling data ofdata of
deepest drilling
SDZK8314; (b) ResultsSDZK8314;
with the (b)
data of SDZK8314).
Results with the data of SDZK8314).

6. Conclusions
Comparing the distribution of orebodies and fractures (Figures 5 and 6), a close
The integration of 3D geological and geochemical modeling, coupled with the in-
relationship can be found
novation between them and the
of the convolutional high
neural probability
network areamarks
algorithm, in Figure 14. As for
a significant stride in
delineating the mostdeep-seated
likely metallogenic position
mineralization at depth, the mineral exploration targets
prediction.
are delineated by considering theprocess,
During this factorsa of the extension
fracture buffer of 30 of orebodies,
m serves fractures,
as the 3D modelingand
rangethe
because
result of CNN. Finally, two target areas are identified. Target Ⅰ lies in the southeastSimul-
of the metallogenic process controlled by the fracture in the Zaozigou deposit.
taneously, the development of 3D geochemical models has illuminated the subsurface
direction at an elevation of 1700–2000 m. This position represents the extension end of the
distribution of critical mineralizing elements. The element distribution and element asso-
Au1 orebody. Notably, thedistribution
ciation Au and Sbofelements
front halo,show significantly
near-ore strong
halo, and tail halo areanomalies at important
extracted as this
location. The ratio of elements
prediction between
indicators the front
appropriate and
for the distal
gold halos has
metallogenic been In
regularity. consistently
particular, the use
of convolutional neural network helps us gain a new
rising. Convolutional neural network modeling has computed a relatively high insight into deep-seated mineraliza-
tion, and the exploration targets proposed by this method deserve more attention.
probability of mineralization in this region. Therefore, it is predicted that the Au1 orebody
In summary, these advancements contribute not only to the scientific understanding
will extend to depths beyond 1800 processes
of mineralization m or even butgive
also rise
to thetopractical
new mineral bodies.
exploration Target
strategies Ⅱ is in
employed
situated in the northwest
unlockingdirection withinriches.
hidden mineral an elevation range of 2350–2650 m. This region
exhibits a higher level of structural complexity. There is a notable vertical overlap of the
distal halo element Co with the front halo elements Sb and As. Compared to the orebodies
in the southeast direction, the control over orebodies on the northwestern side is slightly
less profound in terms of depth. However, the deeper extension of the orebody has not
been constrained, particularly concerning stable and thick orebodies like Au9 and Au126.
Minerals 2023, 13, 1205 19 of 22

Author Contributions: Methodology, C.L., K.X. and B.L.; software, M.X. and C.L.; writing—original
draft preparation, C.L., B.L. and L.W.; writing—review and editing, C.L. and B.L.; visualization, Y.K.,
M.X., Y.W. and R.T.; supervision, B.L.; funding acquisition, B.L. All authors have read and agreed to
the published version of the manuscript.
Funding: This research was funded by the National Key Research and Development Program of
China (Grants 2017YFC0601505) and the National Natural Science Foundation of China (Grants
41602334; 42072322).
Data Availability Statement: Data is not available.
Acknowledgments: The authors thank the anonymous reviewers and the editors for their hard work
on this paper. We are grateful to the Development Research Center of China Geological Survey and
No.3 Geological and Mineral Exploration team, Gansu Provincial Bureau of Geology and Mineral
Exploration and Development for their data support.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Houlding, S.W. 3D Geoscience Modeling: Computer Techniques for Geological Characterization; Springer: London, UK, 1994.
2. Hou, W.S.; Wu, X.C.; Liu, X.G.; Chen, G.N. A complex fault modeling method based on geological plane map. Rock Soil Mech.
2007, 28, 169–172, (In Chinese with English Abstract).
3. Guo, J.T.; Wu, L.X.; Zhou, W.H. Automatic ore body implicit 3D modeling based on radial basis function surface. J. China Coal Soc.
2016, 41, 2130–2135, (In Chinese with English Abstract).
4. Mao, X.C.; Zhang, M.M.; Deng, H.; Zou, Y.H.; Chen, J. 3D Visualization Prediction Method for Concealed Ore Bodies in the Deep
Mining Area. J. Geol. 2016, 40, 363–371, (In Chinese with English Abstract).
5. Xiao, K.Y.; Li, N.; Alok, P.; Holden, E.J.; Leon, B. GIS-based 3D prospectivity mapping: A case study of Jiama copper-polymetallic
deposit in Tibet, China. Ore Geol. Rev. 2015, 71, 611–632. [CrossRef]
6. Li, N.; Xiao, K.Y.; Yin, J.N.; Fan, J.F.; Wang, K. A Method of 3D Buffer Analysis of Boundary Representation. J. Comput.-Aided Des.
Comput. Graph. 2015, 27, 1625–1636.
7. Nielsen, S.H.H.; Partigton, G.A.; Franey, D.; Dwight, T. 3D mineral potential modelling of gold distribution at the Tampia gold
deposit. Ore Geol. Rev. 2019, 109, 276–289. [CrossRef]
8. Farahbakhsh, E.; Hezarkhani, A.; Eslamkish, T.; Bahroudi, A.; Chandra, R. Three-dimensional weights of evidence modelling of a
deep-seated porphyry Cu deposit. Geochem. Explor. Environ. Anal. 2020, 20, 480–495. [CrossRef]
9. Mao, X.C.; Deng, H.; Chen, J. Research Progress on Three-Dimensional Analysis and Modeling Prediction of Deep Metallogenic
Structure. In Proceedings of the Annual Meeting of Chinese Geoscience Union, Beijing, China, 21 October 2018.
10. Mao, X.C.; Wang, Q.; Chen, J.; Deng, H.; Liu, Z.K.; Wang, J.L.; Chen, J.P.; Xiao, K.Y. Three-dimensional Modeling of Deep
Metallogenic Structure in Northwestern Jiaodong Peninsula and Its Gold Prospecting Significance. Acta Geol. Sin. 2020,
41, 166–178.
11. Agterberg, F.P. Automatic contouring of geological maps to detect target areas for mineral exploration. J. Int. Assoc. Math. Geol.
1974, 6, 373–375. [CrossRef]
12. Zhao, P.D.; Meng, X.G. Geological Anomaly and Mineral Prediction. Earth Sci. 1993, 18, 682–684, (In Chinese with English Abstract).
13. Cheng, Q.M. Singularity-Generalized Self-Similarity-Fractal Spectrum (3S) Models. Earth Sci. 2006, 31, 337–348, (In Chinese with
English Abstract).
14. Chen, J.P.; Lv, P.; Wu, W.; Zhao, J.; Hu, Q. A 3D method for predicting blind orebodies, based on a 3D visualization model and its
application. Earth Sci. Front. 2007, 14, 54–62, (In Chinese with English Abstract). [CrossRef]
15. Song, M.; Li, S.; Zheng, J.; Wang, B.; Fan, J.; Yang, Z.; Wen, G.; Liu, H.; He, C.; Zhang, L.; et al. A 3D Predictive Method for
Deep-Seated Gold Deposits in the Northwest Jiaodong Peninsula and Predicted Results of Main Metallogenic Belts. Minerals
2022, 12, 935. [CrossRef]
16. Xiao, K.; Xiang, J.; Fan, M.; Xu, Y. 3D mineral prospectivity mapping based on deep metallogenic prediction theory: A case study
of the Lala Copper Mine, Sichuan, China. J. Earth Sci. 2021, 32, 348–357. [CrossRef]
17. Mao, X.C.; Zhou, Y.H.; Chen, J.; Lai, J.Q.; Peng, S.L.; Shao, Y.J.; Shu, Z.M.; Lv, J.W.; Lv, Z.Y. Three-Dimension Visual Prediction Of
Concealed Ore Bodies in the Deep and Marginal Parts of crisis Mines: A case Study of the Fenghuangshan Ore Field in Tongling,
Anhui, China. Geol. Bull. China 2010, 29, 401–413, (In Chinese with English Abstract).
18. Liu, Y.; Zhou, K.F. Gold Anomaly Identification and Its Uncertainty Analysis in the West Junggar Belt, Xinjiang. Earth Sci. 2018,
43, 3186–3199, (In Chinese with English Abstract).
19. Zhang, Q.; Chen, J.; Xu, H.; Jia, Y.; Chen, X.; Jia, Z.; Liu, H. Three-dimensional mineral prospectivity mapping by XGBoost
modeling: A case study of the Lannigou gold deposit, China. Nat. Resour. Res. 2022, 31, 1135–1156. [CrossRef]
20. Mao, X.C.; Ren, J.; Liu, Z.K.; Chen, J.; Tang, L.; Deng, H.; Bayless, R.C.; Yang, B.; Wang, M.J.; Liu, C.M. Three-dimensional
prospectivity modeling of the Jiaojia-type gold deposit, Jiaodong Peninsula, Eastern China: A case study of the Dayingezhuang
deposit. J. Geochem. Explor. 2019, 203, 27–44, (In Chinese with English Abstract). [CrossRef]
Minerals 2023, 13, 1205 20 of 22

21. Yousefi, M.; Kreuzer, O.P.; Nykanen, V.; Hronsky, J.M.A. Exploration information systems-A proposal for the future use of GIS in
mineral exploration targeting. Ore Geol. Rev. 2019, 111, 103005. [CrossRef]
22. Liu, G.; Wu, C.L.; He, Z.W.; Wong, Z.P.; Zhu, Q.; Zhang, Y.T.; Li, X.M. Design and Application of 3D Spatial Database Model for
Integrated Management of Aboveground and Underground Features. Earth. Sci. 2011, 36, 367–374, (In Chinese with English
Abstract).
23. Sun, L.; Xiao, K.Y.; Tang, J.X.; Zou, W.; Li, N.; Sun, Y. 3D Geologic Modeling of Jiama Cu Deposit Based on Minexplorer System.
J. Chengdu Univ. Technol. Sci. Technol. Ed. 2011, 38, 291–297, (In Chinese with English Abstract).
24. Xiao, K.Y.; Li, N.; Sun, L.; Zou, W.; Li, Y. Large scale 3D mineral prediction methods and channels based on 3D information
technology. J. Geol. 2012, 36, 229–236, (In Chinese with English Abstract).
25. Chen, J.P.; Yu, M.; Yu, P.P.; Shang, B.C.; Zheng, X.; Wang, L.M. Method and Practice of 3D Geological Modeling at Key Metallogenic
Belt with Large and Medium Scale. Acta Geol. Sin. 2014, 88, 1187–1195, (In Chinese with English Abstract).
26. Wang, K.; Xiao, K.Y.; Yan, Y.Y. Application of 3 D modeling technology in Lead-zinc prospecting in Huayuan, Hunan Province.
J. Geol. 2014, 38, 441–444, (In Chinese with English Abstract).
27. Zhang, M.M.; Li, X.H.; Yuan, F.; Hu, X.Y.; Wang, X.H. Comparative Study Between Three-Dimensional Straya Modeling and
Surface model Interpolation. Anhui Geol. 2015, 25, 182–186, (In Chinese with English Abstract).
28. Gao, L.; Liu, Q.Y.; Xv, S.T.; Yang, T.; Yv, P.P.; Zhou, Y.Z. Three-dimensional geological modeling of deposits and reserve
estimation: A case study of the Jingkou ore block of the Fengcun Pb-Zn deposit. Geol. Explor. 2016, 52, 956–965, (In Chinese with
English Abstract).
29. Carranza, E.J.M.; Sadeghi, M. Primary geochemical characteristics of mineral deposits—Implications for exploration. Ore Geol.
Rev. 2012, 45, 1–4. [CrossRef]
30. Safronov, N.I. Dispersion haloes of ore deposits and their use in exploration. Probl. Sov. Geol. 1936, 4, 41–53.
31. Distler, V.V.; Yudovskaya, M.A.; Mitrofanov, G.L.; Prokof’Ev, V.Y.; Lishnevskii, E.N. Geology, composition, and genesis of the
Sukhoi Log noble metals deposit, Russia. Ore Geol. Rev. 2004, 24, 7–44. [CrossRef]
32. Gundobin, G.M. Peculiarities in the zoning of primary halos. J. Geochem. Explor. 1984, 21, 193–200. [CrossRef]
33. Ziaii, M.; Carranza, E.J.M.; Ziaei, M. Application of geochemical zonality coefficients in mineral prospectivity mapping. Comput.
Geosci. 2011, 37, 1935–1945. [CrossRef]
34. Goodell, P.C.; Petersen, U. Julcani mining district, Peru: A study of metal ratios. Econ. Geol. 1974, 69, 347–361. [CrossRef]
35. Jones, B.K. Application of metal zoning to gold exploration in porphyry copper systems. J. Geochem. Explor. 1992, 43, 127–155.
[CrossRef]
36. Pirajno, F.; Smithies, R.H. The FeO/(FeO+MgO) ratio of tourmaline: A useful indicator of spatial variations in granite-related
hydrothermal mineral deposits. J. Geochem. Explor. 1992, 42, 371–381. [CrossRef]
37. Kreı̆n, M.G.; Melik-Adamjan, F.È. An integrated lithogeochemical approach to detecting and interpreting cryptic alteration
around the Elura Zn-Pb-Ag deposit, New South Wales, Australia. Geochem. Explor. Environ. Anal. 2011, 11, 233–246.
38. Pearce, T.H. A contribution to the theory of variation diagrams. Contrib. Mineral. Petrol. 1968, 19, 142–157. [CrossRef]
39. Stanley, C.R.; Madeisky, H.E. Lithogeochemical exploration for hydrothermal ore deposits using molar element ratio analysis. In
Alteration and Alteration Processes Associated with Oreforming Systems. Geological Association of Canada, Short Course Handbook; Lentz,
D.R., Ed.; Geological Association of Canada: Labrador, NL, Canada, 1994; Volume 11, pp. 193–211.
40. Urqueta, E.; Kyser, T.K.; Clark, A.H.; Stanley, C.R.; Oates, C.J. Lithogeochemistry of the Collahuasi porphyry Cu-Mo and
epithermal Cu-Ag (−Au) cluster, northern Chile: Pearce element ratio vectors to ore. Geochem. Explor. Environ. Anal. 2009, 9, 9–17.
[CrossRef]
41. Barrie, C.T. Petrochemistry of shoshonitic rocks associated with porphyry copper-gold deposits of central Quesnellia, British
Columbia, Canada. J. Geochem. Explor. 1993, 48, 225–258. [CrossRef]
42. Date, J.; Watanabe, T.; Saeki, Y. Zonal alteration around the Fukazawa kuroko deposits, Akita prefecture, northern Japan. Econ.
Geol. 1983, 5, 365–386.
43. Ishikawa, Y.; Sawaguchi, T.; Iwaya, S.; Horiuchi, M. Delineation of prospecting targets for Kuroko deposits based on modes of
volcanism of underlying dacite and alteration haloes. Min. Geol. 1976, 26, 105–117.
44. Kishida, A.; Kerrich, R. Hydrothermal alteration zoning and gold concentration at the Kerr–Addison Archean lode gold deposit,
Kirkland Lake, Ontario. Econ. Geol. 1987, 82, 649–690. [CrossRef]
45. Large, R.R.; Mcgoldrick, P.J. Lithogeochemical halos and geochemical vectors to stratiform sediment hosted Zn–Pb–Ag deposits,
1. Lady Loretta Deposit, Queensland. J. Geochem. Explor. 2000, 68, 105–126. [CrossRef]
46. Large, R.R.; McPhie, J.; Gemmell, J.B.; Herrmann, W.; Davidson, G.J. The spectrum of ore deposit types, volcanic environments,
alteration halos, and related exploration vectors in submarine volcanic successions: Some examples from Australia. Econ. Geol.
2001, 96, 913–938. [CrossRef]
47. Piché, M.; Jébrak, M. Normative minerals and alteration indices developed for mineral exploration. J. Geochem. Explor. 2004,
82, 59–77. [CrossRef]
48. Prendergast, K. Application of lithogeochemistry to gold exploration in the St Ives Goldfield, Western Australia. Geochem. Explor.
Environ. Anal. 2007, 7, 99–108. [CrossRef]
Minerals 2023, 13, 1205 21 of 22

49. Wang, L.; Percival, J.; Hedenquist, J.W.; Hattori, K.; Qin, K.Z. Alteration mineralogy of the Zhengguang Au-Zn deposit, Northeast
China: Interpretation of shortwave infrared analyses during mineral exploration and assessment. Econ. Geol. 2021, 116, 389–406.
[CrossRef]
50. Xie, X.J.; Chen, H.C. The role of the native halo method in the geological survey and exploration. Acta Geol. Sin. 1961, 41, 273–284.
(In Chinese)
51. Shao, Y. Several problems with the native halo approach. Geol. Explor. 1964, 03, 21–24. (In Chinese)
52. Beus, A.A.; Grigorian, S.V. Geochemical Exploration Methods for Mineral Deposits; Applied Publishing Ltd.: Wilmette, IL, USA, 1977.
53. Li, H. An I deal Pattern of Primary Superimposed Halo of Hydrothermal Gold Deposits. Geol. Explor. 1993, 03, 46–51, (In Chinese
with English Abstract).
54. Qing, C.S.; Peng, X.H.; Xu, B.; Yang, H.; Zhang, J.S.; Hu, Y.; Zhong, Y.L.; Mu, X.H.; Wang, T. Research progress of native halo
prospecting method. Acta Miner. Sin. 2011, 31, 828–829. (In Chinese)
55. Li, H.; Yu, B.; Wei, J.; Li, Y.C.; Wang, J.; Wei, Z.X.; Zhang, H.R.; Ren, L.L.; Wang, X.; Sun, F.Z.; et al. A new practical ideal model of
structural superimposed halos for prediction of deep blind hydrothermal deposits and its significance. Geol. Explor. 2020, 56,
889–897, (In Chinese with English Abstract).
56. Ran, X.; Xue, L.; Zhang, Y.; Liu, Z.; Sang, X.; He, J. Rock Classification from Field Image Patches Analyzed Using a Deep
Convolutional Neural Network. Mathematics 2019, 7, 755. [CrossRef]
57. Sang, X.; Xue, L.; Ran, X.; Li, X.; Liu, J.; Liu, Z. Intelligent High-Resolution Geological Mapping Based on SLIC-CNN. ISPRS Int. J.
Geo-Inf. 2020, 9, 99. [CrossRef]
58. Guo, J.; Li, Y.; Jessell, M.W.; Giraud, J.; Li, C.; Wu, L.; Li, F.; Liu, S. 3D geological structure inversion from Noddy-generated
magnetic data using deep learning methods. Comput. Geosci. 2021, 149, 104701. [CrossRef]
59. Li, T.; Zuo, R.; Xiong, Y.; Peng, Y. Random-drop data augmentation of deep convolutional neural network for mineral prospectivity
mapping. Nat. Resour. Res. 2021, 30, 27–38. [CrossRef]
60. Sun, T.; Li, H.; Wu, K.; Chen, F.; Zhu, Z.; Hu, Z. Data-driven predictive modelling of mineral prospectivity using machine learning
and deep learning methods: A case study from southern Jiangxi Province, China. Minerals 2020, 10, 102. [CrossRef]
61. Li, S.; Chen, J.; Liu, C.; Wang, Y. Mineral prospectivity prediction via convolutional neural networks based on geological big data.
J. Earth Sci. 2021, 32, 327–347. [CrossRef]
62. Zhang, S.; Carranza, E.J.; Wei, H.; Xiao, K.; Yang, F.; Xiang, J.; Zhang, S.; Xu, Y. Data-driven mineral prospectivity mapping by
joint application of unsupervised convolutional auto-encoder network and supervised convolutional neural network. Nat. Resour.
Res. 2021, 30, 1011–1031. [CrossRef]
63. Feng, Y.M.; Cao, X.Z.; Zhang, E.P.; Hu, Y.X.; Pan, X.P.; Yang, J.L.; Jia, Q.Z.; Li, W.M. Tectonic Evolution Framework and Nature of
The West Qinling Orogenic Belt. Northwest. Geol. 2003, 36, 1–10, (In Chinese with English Abstract).
64. Wei, L.X. Tectonic Evolution and Mineralization of Zaozigou Gold Deposit, Gansu Province. Master’s Thesis, China University of
Geosciences, Beijing, China, 2015. (In Chinese with English Abstract).
65. Zeng, J.J.; Li, K.N.; Yan, K.; Wei, L.L.; Huo, X.D.; Zhang, J.P. Tectonic Setting and Provenance characteristics of the Lower Triassic
Jiangligou Formation in West Qinling—Constraints from Geochemistry of Clastic Rock and zircon U-Pb Geochronology of
Detrital Zircon. Geol. Rev. 2021, 67, 1–15, (In Chinese with English Abstract).
66. Li, Z.B.; Liu, Z.Y.; Li, R. Geochemical Characteristics and metallogenic Potential Analysis of Daheba Formation in Ta-Ga Area of
Gansu Province. Contrib. Geol. Miner. Resour. Res. 2021, 36, 187–194, (In Chinese with English Abstract).
67. Li, K.N. Geochemical prospecting perspective of Wengcheng area, Guangdong province. Geol. Miner. Resour. 2014, 30, 138–145.
68. Chen, Y.; Wang, K.J. Geological Features and Ore Prospecting Indicators of Sishangou Silver Deposit. Gansu. Metal. 2015, 37,
108–111, (In Chinese with English Abstract).
69. Di, P.F. Geochemistry and Ore-Forming Mechanism on Zaozigou Gold Deposit in Xiahe-Hezuo, West Qinling, China. Doctor’s
Thesis, Lanzhou University, Lanzhou, China, 2018. (In Chinese with English Abstract).
70. Tang, L.; Lin, C.G.; Cheng, Z.Z.; Jia, R.Y.; Li, H.R.; Li, K.N. 3D Characteristics of Primary Halo and Deep Prospecting
Prediction in The Zaozigou Gold Deposit, Hezuo City, Gansu Province. Geol. Bull. China 2020, 39, 1173–1181, (In Chinese with
English Abstract).
71. Chen, G.Z.; Wang, J.L.; Liang, Z.L.; Li, P.B.; Ma, H.S.; Zhang, Y.N. Analysis of Geological Structures in Zaozigou Gold Deposit of
Gansu Province. Gansu Geol. 2013, 22, 50–57, (In Chinese with English Abstract).
72. Chen, G.Z.; Liang, M.L.; Wang, J.L.; Zhang, Y.N.; Li, P.B. Characteristics and Deep Prediction of Primary Superimposed Halos
in The Zaozigou Gold Deposit of Hezuo, Gansu Province. Geophys. Geochem. Explor. 2014, 38, 268–277, (In Chinese with
English Abstract).
73. Jin, D.G.; Liu, B.C.; Chen, Y.Y.; Liang, Z.L. Spatial Distribution of Gold Bodies in Zaozigou Mine of Gansu Province. Gansu Geol.
2015, 24, 25–30+41, (In Chinese with English Abstract).
74. Zhu, F.; Wang, G.W. Study on Grade Model of Gansu Zaozigou Gold Mine Based on Geological Statistics. Acta Mineral. Sin. 2015,
35, 1065–1066. (In Chinese)
75. Chen, G.Z.; Li, L.N.; Zhang, Y.N.; Ma, H.S.; Liang, Z.L.; Wu, X.M. Characteristics of fluid inclusions and deposit formation in
Zaozigou gold mine. J. Jilin Univ. (Earth Sci. Ed.) 2015, 45, 1–2. (In Chinese)
76. Liu, Y.; Liu, Y.H.; Dong, F.C.; Li, Z.H.; Yv, J.K. Accurate Determination and geological significance of Zaozigou gold deposit in
Gansu province. Adv. Earth Sci. 2012, 33, 10–15.
Minerals 2023, 13, 1205 22 of 22

77. Li, W.; Chen, J.P.; Jia, Y.L.; Zhou, G.Y.; Mao, X.C.; Xiao, K.Y. Three-dimensional Modeling and Comprehensive Metallogenic
Prediction of the Zaozigou Gold Deposit, Gansu Province. Area. J. Geol. 2020, 41, 144–156, (In Chinese with English Abstract).
78. LeCun, Y.; Bengio, Y. Convolutional networks for images, speech, and time series. In The Handbook of Brain Theory and Neural
Network; Arbib, M.A., Ed.; MIT Press: Cambridge, MA, USA, 1995; p. 3361.
79. Zeiler, M.D.; Fergus, R. Visualizing and understanding convolutional networks. In Proceedings of the Computer Vision–ECCV
2014: 13th European Conference, Zurich, Switzerland, 6–12 September 2014; Proceedings, Part I 13. Springer International
Publishing: Berlin/Heidelberg, Germany, 2014; pp. 818–833.
80. Nair, V.; Hinton, G.E. Rectified linear units improve restricted boltzmann machines. In Proceedings of the 27th International
Conference on Machine Learning (ICML-10), Haifa, Israel, 21–24 June 2010; pp. 807–814.
81. Sermanet, P.; Eigen, D.; Zhang, X.; Mathieu, M.; Fergus, R.; LeCun, Y. Overfeat: Integrated recognition, localization and detection
using convolutional networks. arXiv 2013, arXiv:1312.6229.
82. Boureau, Y.L.; Ponce, J.; LeCun, Y. A theoretical analysis of feature pooling in visual recognition. In Proceedings of the 27th
International Conference on Machine Learning (ICML-10), Haifa, Israel, 21–24 June 2010; pp. 111–118.
83. Simonyan, K.; Zisserman, A. Very deep convolutional networks for large-scale image recognition. arXiv 2014, arXiv:1409.1556.
84. Lecun, Y.; Bottou, L.; Bengio, Y.; Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 1998,
86, 2278–2324. [CrossRef]
85. Maturana, D.; Scherer, S. Voxnet: A 3d convolutional neural network for real-time object recognition. In Proceedings of the 2015
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hamburg, Germany, 28 September–3 October 2015;
IEEE: Piscataway, NJ, USA, 2015; pp. 922–928.
86. Abdoli, S.; Cardinal, P.; Koerich, A.L. End-to-end environmental sound classification using a 1D convolutional neural network.
Expert Syst. Appl. 2019, 136, 252–263. [CrossRef]
87. Ieracitano, C.; Mammone, N.; Bramanti, A.; Hussain, A.; Morabito, F.C. A convolutional neural network approach for classification
of dementia stages based on 2D-spectral representation of EEG recordings. Neurocomputing 2019, 323, 96–107. [CrossRef]
88. Li, X.; Xue, C.; Chen, Y.; Yuan, F.; Li, Y.; Zheng, C.; Zhang, M.; Ge, C.; Guo, D.; Lan, X.; et al. 3D Convolutional Neural
Network-based 3D mineral prospectivity modeling for targeting concealed mineralization within Chating area, middle-lower
Yangtze River metallogenic Belt, China. Ore Geol. Rev. 2023, 157, 105444. [CrossRef]
89. Jia, L.J. Research on 3D Geological Modeling Interpolation Algorithm Based on Drilling Data. Master’s Thesis, China University
of Geosciences, Beijing, China, 2018. (In Chinese with English Abstract).
90. Ye, T.Z.; Xiao Ky Yan, G.S. Methodology of deposit modeling and mineral resource potential assessment using integrated
geological information. Earth Sci. Front. 2007, 14, 11–19, (In Chinese with English Abstract).
91. Lin, T.Y.; Goyal, P.; Girshick, R.; He, K.; Dollár, P. Focal Loss for Dense Object Detection. In Proceedings of the IEEE International
Conference on Computer Vision, Venice, Italy, 22–29 October 2017; pp. 2999–3007.
92. Kingma, D.; Ba, J. Adam: A Method for Stochastic Optimization. arXiv 2014, arXiv:1412.6980.
93. Kong, Y.; Chen, G.; Liu, B.; Xie, M.; Yu, Z.; Li, C.; Wu, Y.; Gao, Y.; Zha, S.; Zhang, H.; et al. 3D Mineral Prospectivity Mapping of
Zaozigou Gold Deposit, West Qinling, China: Machine Learning-Based Mineral Prediction. Minerals 2022, 12, 1361. [CrossRef]
94. Yu, Z.; Liu, B.; Xie, M.; Wu, Y.; Kong, Y.; Li, C.; Chen, G.; Gao, Y.; Zha, S.; Zhang, H.; et al. 3D Mineral Prospectivity Mapping of
Zaozigou Gold Deposit, West Qinling, China: Deep Learning-Based Mineral Prediction. Minerals 2022, 12, 1382. [CrossRef]

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.