minerals

Article
New Outlier Top-Cut Method for Mineral Resource
Estimation via 3D Hot Spot Analysis of Borehole Data
Sung-Min Kim 1 , Yosoon Choi 2, * ID
and Hyeong-Dong Park 3
1 Division of Graduate Education for Sustainability of Foundation Energy, Seoul National University,
Seoul 08826, Korea; snuhyrule@hanmail.net
2 Department of Energy Resources Engineering, Pukyong National University, Busan 48513, Korea
3 Department of Energy Resources Engineering, Seoul National University, Seoul 08826, Korea;
hpark@snu.ac.kr
* Correspondence: energy@pknu.ac.kr; Tel.: +82-51-629-6562; Fax: +82-51-629-6553




Received: 18 July 2018; Accepted: 9 August 2018; Published: 11 August 2018


Abstract: Three-dimensional (3D) analysis of borehole data is very important for effective mineral
exploration. It can be used not only to understand the geological structure of the underground, but to
estimate the amount of the resource. In the mining industry, the geostatistical interpolation, such as
kriging, is widely used to predict the value of a whole section using this borehole data. In order to
obtain reasonable prediction results, it is firstly necessary to verify assay and geological databases.
In addition, if the assayed grade data deviates significantly from the average value, it is necessary to
perform the prediction including the outlier top-cut because it may excessively affect the predicted
value. However, the existing top-cut methods of determining a specific threshold value may cause an
error by excluding significant data. In this study, to minimize the loss of such data, we developed a
3D hot spot analysis technique to analyze statistically significant outliers. In addition, it was applied
to borehole data analysis of the Au deposit. As a result, we confirmed that the proposed method can
mitigate the overestimation or underestimation that might occur when applying the existing methods.

Keywords: hot spot analysis; resource estimation; top-cut; kriging; outlier

1. Introduction
Resource estimation based on computerized methods has been widely used in the mining industry.
This method is fast and editable compared to the traditional manual methods. Today, three-dimensional
(3D) terrain modeling software, such as Micromine (Micromine, Perth, Australia), GOCAD (Paradigm,
Houston, TX, United States), GEMS (Dassault Systèmes, Vélizy-Villacoublay, France) and Surpac
(Dassault Systèmes, Vélizy-Villacoublay, France), can integrate two-dimensional (2D) geographic
information systems (GIS), databases, statistical methods, and 3D visual technologies. Complex and
irregular geological objects in this software can be modeled in a 3D environment using a variety of
data sources such as topographic maps, geophysical data, and geochemical data [1,2].
For the application of computerized methods, 3D modeling of the ore body as a geological
basis should be done through stratum analysis and critical review of existing sample data. Then,
these envelopes can be divided into block models. The grade of the block is estimated by mathematical
methods, such as geostatistical interpolation. However, the computerized method itself does not
guarantee the quality of the results, and numerous mining projects modeled and designed using
computerized methods have failed [3]. Because the geostatistical method is only a tool for estimating
grades, it cannot solve the existing problems of assay databases or geological interpretations. Therefore,
assay and geological databases should be verified before they are used for resource estimation.

Minerals 2018, 8, 348; doi:10.3390/min8080348 www.mdpi.com/journal/minerals

Minerals 2018, 8, 348 2 of 14

In addition, unbiased models are important for well-established geological resource assessments
because biased model can result in resource overestimation.
To avoid the problem of overestimation, the outlier samples that do not conform to the distribution
of the analysis should be appropriately capped. This process is called top-cut. The determination of
the values considered as outliers is subjective. The high grade sample should be limited enough
to prevent overestimation, but not too conservatively. There are some commonly used top-cut
methods for determining the capping grade [4–6]. One way is to use a breakpoint on the log-normal
cumulative frequency plot. If the assay sample follows a log-normal distribution, the data points of
the probability plot must follow a straight line. In some cases, outliers are defined as values outside
of a specific interval, such as 2 or 3 standard deviations from the mean or median of the distribution.
Another common approach is to use the 95th percentile of cumulative frequency.
However, these methods cannot manage different types of outliers properly. Isolated outliers cause
a problem in estimation because they tend to have different continuity than most of the data. Therefore,
they must be treated differently. More abundant, clustered outliers may define a sufficiently large
domain, which has spatial connectivity, and it can be considered as a significant mineralized area [7].
The presence of outliers is particularly problematic if they have minimal spatial connectivity [8].
In other words, there is a large potential impact of outliers on overestimation if they are located
within a small, spatially restricted volume. Because clustered outliers can be a significant data source
for mineralization, unlike isolated outliers, they should be managed differently. However, existing
methods cannot separate them because they consider only the absolute value of the assayed data.
To overcome this limitation in defining outliers, this study proposed a new method by using
hot spot analysis, which calculates z-scores based on the Getis–Ord Gi* statistic [9], identifying a
statistically significant hot spot sample. To be a statistically significant hot spot, a feature will have
a high value and be surrounded by other features with high values. This method has been widely
used in applied geographic research to identify clustering of species populations, diseases, and crime
incidence. Recently, there have been some studies to assess the soil contamination using hot spot
analysis. McClintock [10] evaluated the risk of lead contamination at post-industrial landscapes in
Oakland, California using hot spot analysis. Griffith et al. [11] considered location error impacts on
local spatial autocorrelation by simulating significant heavy metal clusters using local Moran’s I and
Getis–Ord Gi*. In addition, Kim et al. [12] supported soil survey planning by using hot spot analysis.
However, previous studies of hot spot analysis were applied only to 2D data and there was no case
applied to 3D spatial data, such as borehole data.
In this study, we developed a 3D hot spot analysis algorithm that can be used for borehole data,
and separates clustered outliers and isolated outliers. In addition, kriging was applied for resource
estimation considering the outliers. Kriging is a widely used geostatistical interpolation technique in
the mining industry. The gold (Au) assay data from boreholes of Au deposits in the Cobre Panama
district were used for analysis. The methods for handling outliers in estimating resources are also
diverse and subjective, like the methods for outlier determination. For example, in order to reduce the
influence of outlier data, a method can be used that reduces a value above a certain threshold value to
the threshold value. Another method excludes outlier data from the analysis altogether. A method
which excludes outliers in the variogram modeling process for analyzing the autocorrelation and
uses them in the kriging prediction step can be an alternative. In this study, outliers are excluded
from the analysis in all processes for resource estimation. However, unlike the isolated outliers,
the clustered outliers may represent important mineralization, so the data are preserved and the results
are compared to other outlier determination methods.
Minerals 2018, 8, 348 3 of 14

2. Materials and Methods

Minerals 2018, 8, x FOR PEER REVIEW 3 of 13
2.1. Study Area and Data
In order to apply the proposed method, the part of Colina Au deposit in the Cobre Panama
district order
In to apply In
was targeted. thethe proposed
area of method,
the Cobre thePanamá
part of deposits,
Colina Authe deposit
oldestinrocksthe Cobre Panama
are submarine
district was targeted. In the area of the Cobre Panamá deposits, the oldest
andesite and basalt flows and tuffs, intercalated with clastic sedimentary rocks and reef limestone, of rocks are submarine
andesite
probable and basalt
Eocene flows Oligocene
to early and tuffs, age.intercalated
Miocenewith and clastic
youngersedimentary
rocks comprise rocksthe andbulk
reefoflimestone,
volcanic
of probable Eocene to early Oligocene age. Miocene and younger rocks comprise
rocks in western Panamá and consist of terrestrial and marine volcanic and volcanic-derived rocks of the bulk of volcanic
rocks in western
progressively morePanamá
felsicand consist of terrestrial
composition. and marine volcanic
Copper-gold-molybdenum and volcanic-derived
porphyry style mineralizationrocks
of progressively
occurs more felsic
at the southern margin composition. Copper-gold-molybdenum
of a large granodiorite porphyry style
batholith of mid-Oligocene mineralization
age. The three main
occurs at the southern margin of a large granodiorite batholith of
deposits of the Cobre Panamá deposits are Botija, Colina, and Valle Grande [13]. Hydrothermal mid-Oligocene age. The three
main deposits of the Cobre Panamá deposits are Botija, Colina, and Valle Grande
alteration is primarily silica-chlorite, which is interpreted to be a form of propylitic alteration [14,15]. [13]. Hydrothermal
alteration is primarily
The Colina deposit silica-chlorite,
is a 2.5 km whichlong andis interpreted
1 km widetofeldspar
be a form of propylitic
quartz hornblende alteration
porphyry[14,15].
sill
and dike complex. An irregular L-shaped andesite roof pendant with the long axis orientated to sill
The Colina deposit is a 2.5 km long and 1 km wide feldspar quartz hornblende porphyry the
and dike complex.
east-southeast and Anthe irregular
short axisL-shaped
orientated andesite roof pendant with
to the south-southwest the longinaxis
is located theorientated
southern part to the
of
east-southeast and the short axis orientated to the south-southwest is located
the deposit. The majority of the feldspar quartz hornblende porphyry comprises 50 m to 200 m thick in the southern part of
the
sillsdeposit.
that dipThe majority
shallowly to of
thethe feldspar
north quartz
and are oftenhornblende porphyry
interconnected comprises
by dikes [13]. 50 m to 200 m thick
sills that
The dip shallowly
study area isto athe north
part of and are often
Colina interconnected
deposit. Approximately by dikes
1400[13].Au grade data from 52
The study area is a part of Colina deposit. Approximately
boreholes were used, and this area is a region with a quartz vein with distributed 1400 Au grade data from 52Au, boreholes
and a
were used, and this area is a region with a quartz vein with distributed
north-south direction displacement occurs adjacent to a fault. The 3D spatial database constructed in Au, and a north-south
direction
this studydisplacement
includes the occursborehole adjacent to a fault.
assay data, The 3Dlines
the contour spatial database
showing theconstructed
topographical in this study
relief, the
includes the borehole assay data, the contour lines showing the topographical
fault, and the quartz vein. These data are visualized using ArcScene software (version 10.4, relief, the fault, and the
quartz vein. These
Environmental data are
Systems visualized
Research using Redlands,
Institute, ArcScene software
CA, United(version 10.4,
States), asEnvironmental
shown in Figure Systems
1. The
Research Institute, Redlands, CA, United States), as shown in Figure
vein, including Au, is in the form of plate quartz, and develops in the east-west direction with 1. The vein, including Au,a
is in the form of plate quartz, and develops in the east-west direction with a slope in the 60–70 ◦ north
slope in the 60°–70° north direction. The domain of quartz vein was defined by the Korea Resources
direction.
Corporation Theconsidering
domain of the quartzfaultvein
andwas defined
assayed by thedata.
borehole Korea TheResources Corporation
near vertical considering
fault is developed in
the fault and assayed borehole data. The near vertical fault is developed
the north-south direction. The 52 boreholes consisted of 10 vertical boreholes and 42 inclined in the north-south direction.
The 52 boreholes
boreholes (dip 60°, consisted
azimuthof180°). 10 vertical boreholes and 42 inclined boreholes (dip 60◦ , azimuth 180◦ ).

Figure 1. Cont.
Minerals 2018, 8, 348 4 of 14
Minerals 2018, 8, x FOR PEER REVIEW 4 of 13

Minerals 2018, 8, x FOR PEER REVIEW 4 of 13

Figure 1. (a) 2D map and (b) 3D model of quartz vein, fault, boreholes, and composited Au grade
Figure 1. (a) 2D map and (b) 3D model of quartz vein, fault, boreholes, and composited Au grade
sample of the Au deposit.
sample of the Au deposit.
Figure 1. (a) 2D map and (b) 3D model of quartz vein, fault, boreholes, and composited Au grade
The histogram from the Au grade assay data of the boreholes reveals a clear difference in Au
sample of the Au deposit.
gradeThe histogram
between thefrom the Au
quartz veingradeandassay
the data of the boreholes
surrounding gangue reveals
(Figurea clear
2). difference
In other in Au grade
words, Au
between the
mineralization quartzis vein and
dominated the surrounding
mainly by the gangue
quartz (Figure
vein. 2).
SinceIn other
there words,
is littleAu mineralization
spatial correlation is
The histogram from the Au grade assay data of the boreholes reveals a clear difference in Au
dominated
between mainly
the quartz byvein
the quartz vein. Since there
and other isstudy
little spatial correlation between the quartz vein and
grade between the quartz vein regions,
and the this
surrounding only analyzed
gangue (Figure data2). that corresponded
In other words, Auto the
other
quartz regions, this study
vein. Data compositing
mineralization only analyzed
is dominated was mainly data
performedthat corresponded
as preprocessing
by the quartz to
vein. Since there the quartz
of the vein.
data.spatial
is little Data compositing
Compositing
correlationis the
was
process of averaging data within a specific length and converting it into samples of thedata
performed
between the as preprocessing
quartz vein and of
other the data.
regions, Compositing
this study only is the
analyzed process
data of
that averaging
corresponded towithin
same the
length.a
It hasquartz
specificthe vein.
length and
advantage Data compositing
converting
of reducing wassamples
it into
the performed
amount of as
of the preprocessing
same length.
analytical data and ofItthe
has data. Compositing
the advantage
providing isreducing
the dataofat the even
an
the process of averaging data within a specific length and converting it
resolution in the assessment of resources [16,17]. In this study, compositing was performed of
amount of analytical data and providing the data at an even into samples
resolution ofinthe same
the length.
assessment at
It has [16,17].
resources the advantage In thisof study,
reducing the amount of
compositing wasanalytical
performeddata andat providingofthe
intervals 3 data
m at an even the
considering
intervals of 3 m considering the intervals of assay data. Since the displacement occurred about 100 m
resolution
intervals in data.
the assessment of resources [16,17]. In this study, mcompositing wasafter
performed at
from the of assay
fault after theSince the displacement
formation of the vein,occurred
data on aboutthe east100side from
of thethefault
fault the formation
was moved for hot
of intervals
theanalysis of
vein, data on 3 m considering the
the east side ofby intervals of
the fault wasassay data.
moved Since the displacement
for hot spottoanalysis occurred
and about 100 m
spot from the andfault kriging
after theanalysis
formation ofmovingthe vein,the relative
data on theposition
east side of the the north
fault wasbykriging
100 m. analysis
moved for hot
by
moving the relative position to the north by 100 m.
spot analysis and kriging analysis by moving the relative position to the north by 100 m.

Figure
Figure 2. Sample
2. Sample survey
survey of Au
of Au gradefor
grade for(a)
(a)all
alldata
data and
and (b)
(b)data
dataininquartz
quartzvein from
vein boreholes
from in thein the
boreholes
Figure
Au Au2.deposit.
Sample survey of Au grade for (a) all data and (b) data in quartz vein from boreholes in the
deposit.
Au deposit.
2.2. Hot Spot Analysis for Determining the Capping Grade
2.2. Hot Spot Analysis for Determining the Capping Grade
Minerals 2018, 8, 348 5 of 14

2.2. Hot Spot Analysis for Determining the Capping Grade

In order to distinguish between clustered outliers and isolated outliers, it is necessary to analyze
the spatial correlation between the borehole data. There are several ways to test and measure spatial
autocorrelation. Moran’s I, semivariance, and spatial autocorrelation coefficients are very popular
methods. Although they are applied globally to the complete region of the study area, it is often
required to examine the pattern at a more local scale. Getis and Ord [9] focused upon local effects and
introduced a measure of spatial association called the Getis–Ord Gi* statistic.
In this study, the Getis–Ord Gi* statistic was calculated for each Au sample value in a borehole
dataset by considering the 3D spatial coordinates of the borehole data. Since existing analysis tools,
such as the hot spot analysis tool of ArcGIS, can only calculate values in 2D space, we developed a 3D
hot spot analysis tool using Visual Basic 2013 (Microsoft Corporation, Washington, WA, United States)
for 3D space analysis. The Getis–Ord Gi* statistic is calculated by comparing the local sum for a feature
and its neighbors to the total sum as follows:

∑nj=1 wi,j x j − X ∑nj=1 wi,j

Gi∗ = rh i (1)
2 −( n w )2
n ∑nj=1 wi,j ∑ j=1 i,j
S n −1

where xj is the attribute value for feature j, wi,j is the spatial weight between feature i and j, n is equal
to the total number of feature and

∑nj=1 x j
X = , and (2)
n
s
∑nj=1 x2j 2
S= − (X) . (3)
n
The Gi∗ resultant statistic is a z-score. For statistically significant positive z-scores, the larger the
z-score, the more intense the clustering of high values (i.e., hot spot). When the local sum is very
different from the expected local sum, and that difference is too large to be the result of random chance,
a z-score is statistically significant. In other words, a feature with a high value surrounded by other
high value features is a statistically significant hot spot feature.
To determine the spatial weight wi,j between borehole data, four methods were applied in the
developed 3D hot spot analysis tool. They are the inverse distance method, the inverse distance squared
method, the fixed distance band method, and the zone of indifference method according to how spatial
relationships among features are conceptualized. It is generally known that the fixed distance band
method is a great option in choosing the right conceptualization of spatial relationships [18]. In the
case of a fixed distance band method, the scale of analysis does not change across the study area
because it uses a critical fixed distance to decide which neighbors to include. In this study, the hot spot
analysis results using the fixed distance band method are presented for determining the outliers that
should be excluded in the kriging prediction process.
To assess statistically significant clustered outliers, composited Au sample data were classified as
four groups according to the Au grade and z-scores of the Gi∗ statistics. The four groups are as follows:
high grade value with high z-score (HH), high grade value with low z-score (HL), low grade value
with high z-score (LH), and low grade value with low z-score (LL). The grade value normalized by
standard deviation and Gi∗ z-score were utilized as the x variable and y variable in a scatter plot to
classify the dataset into four groups. According to the existing top-cut methods, higher values above a
certain threshold value should all be excluded, but in this study, we proposed a method named hot
spot top-cut that does not exclude clustered outliers when the z-score is high. A schematic diagram of
the application of the proposed method for borehole data is shown in Figure 3.
Minerals 2018, 8, 348 6 of 14
Minerals 2018, 8, x FOR PEER REVIEW 6 of 13

Figure 3.
Figure 3. A
A schematic
schematic diagram
diagram of
of the
the 3D
3D hot
hot spot
spot analysis
analysis for
for borehole
borehole data.
data.

2.3. Variography
2.3. Variography and
and Grade
Grade Estimation
Estimation
Variography,ororspatial
Variography, spatial statistics,
statistics, is the
is the determination
determination of theofspatial
the spatial correlation
correlation of sampleof sample
values.
values.
It It isimportant
is a very a very important
study in study in the computerized
the computerized resource estimation
resource estimation of the
of the mining mining Ifindustry.
industry. kriging
If used
is kriging
forisgrade
used estimation,
for grade estimation,
experimental experimental
variogramsvariograms are very important
are very important as a preliminary
as a preliminary analysis.
analysis.
The The experimental
experimental variogramvariogram
γ(h) measures measures
𝛾(ℎ)the the of
variability variability
two dataof two[zdata
pairs ( xi ),pairs h)] ),
z( xi +[𝑧(𝑥 𝑧(𝑥are
that +
ℎ)] that are
separated byseparated
lag vectorbyh, lag vector ℎ, as follows:
as follows:

11 N
γ(h) ==2𝑁 ∑
𝛾(ℎ) [z( x)i −
[𝑧(𝑥 ) −𝑧(𝑥
z( x+ ℎ)] 2 (4)
i + h )] (4)
2N i=1
where N is the number of data pairs for a particular distance and direction. After comprehending the
where N is the number of data pairs for a particular distance and direction. After comprehending the
spatial correlations of the data from the experimental variogram, the theoretical variogram should
spatial correlations of the data from the experimental variogram, the theoretical variogram should be
be modeled well as a representative of the experimental variogram to compute these correlations
modeled well as a representative of the experimental variogram to compute these correlations over
over all distances.
all distances.
One of the most common problems in variogram modeling occurs when a variogram
One of the most common problems in variogram modeling occurs when a variogram calculation
calculation is not limited to mineralized samples. If the assay samples are mixed with mineralized
is not limited to mineralized samples. If the assay samples are mixed with mineralized and waste
and waste samples, they are strongly influenced by low grade waste samples in the variogram
samples, they are strongly influenced by low grade waste samples in the variogram calculation [3].
calculation [3]. That is, a seemingly good and smooth variogram model with a long variogram range
That is, a seemingly good and smooth variogram model with a long variogram range can be computed,
can be computed, but it yields inadequate conclusions with irrational long distance correlations. To
but it yields inadequate conclusions with irrational long distance correlations. To prevent this, it is
prevent this, it is necessary to perform calculations by distinguishing geological domains. Another
necessary to perform calculations by distinguishing geological domains. Another method is to set the
method is to set the cut-off grade and exclude the value below the grade for the variogram
cut-off grade and exclude the value below the grade for the variogram calculation.
calculation.
One of the most popular methods in geostatistics is kriging. Kriging is a technique for predicting
One of the most popular methods in geostatistics is kriging. Kriging is a technique for
values for unknown points using a weighted linear combination of known values (Equation (5)).
predicting values for unknown points using a weighted linear combination of known values
(Equation (5)). N
z ∗ = ∑ λi zi (5)
i =1
𝑧∗ = 𝜆𝑧 (5)
z∗
Here, denotes the kriging prediction value for the prediction point, and zi denotes the known
surrounding ∗values. λi represents the weight of each data, and N represents the total number of
Here, 𝑧 denotes the kriging prediction value for the prediction point, and 𝑧 denotes the
data used for kriging prediction. There are several types of kriging according to the difference in the
known surrounding values. 𝜆 represents the weight of each data, and N represents the total
process of determining the weights. Among them, ordinary kriging and indicator kriging are the most
number of data used for kriging prediction. There are several types of kriging according to the
commonly used methods in the mining industry grade estimation.
difference in the process of determining the weights. Among them, ordinary kriging and indicator
kriging are the most commonly used methods in the mining industry grade estimation.
Ordinary kriging does not bias the prediction equation and minimizes the error variance. At
this time, the sum of the weights of the samples equals 1. It has been proved that the result is stable
Minerals 2018, 8, 348 7 of 14

Ordinary kriging does not bias the prediction equation and minimizes the error variance. At this
Minerals 2018, 8, x FOR PEER REVIEW 7 of 13
time, the sum of the weights of the samples equals 1. It has been proved that the result is stable in the
grade
in the estimation and widely
grade estimation used in used
and widely the mining
in the industry [19–21]. Indicator
mining industry kriging converts
[19–21]. Indicator krigingthe grade
converts
of the sample to the indicators of a 0 or 1 based on the cut-off grade prior to the variogram
the grade of the sample to the indicators of a 0 or 1 based on the cut-off grade prior to the variogram modeling.
This method
modeling. does
This not predict
method thepredict
does not grade the
of the block,
grade butblock,
of the evaluates the possibility
but evaluates that the grade
the possibility of
that the
the block
grade is higher
of the block isthan the selected
higher than the cut-off
selectedgrade [22,23].
cut-off gradeThe purpose
[22,23]. The of this study
purpose is to
of this evaluate
study is to
the applicability of the proposed hot spot top-cut method for grade estimation. Therefore,
evaluate the applicability of the proposed hot spot top-cut method for grade estimation. Therefore, ordinary
kriging was
ordinary chosen—the
kriging most popular
was chosen—the mostspatial
popularinterpolation method inmethod
spatial interpolation grade estimation.
in grade estimation.
3. Results and Discussion
3. Results and Discussion
3.1. Outlier Determination
3.1. Outlier Determination
We applied the existing methods for determining outliers and compared them to the proposed
We applied the existing methods for determining outliers and compared them to the proposed
method. Figure 4 shows the method of determining the outliers by using the probability plot,
method. Figure 4 shows the method of determining the outliers by using the probability plot, and
and the Au grade of 12.15 g/t, in which the slope change is determined as the threshold value
the Au grade of 12.15 g/t, in which the slope change is determined as the threshold value of the
of the outlier. In addition, when the upper 5% value is used as the threshold value, the grade of
outlier. In addition, when the upper 5% value is used as the threshold value, the grade of 13.37 g/t is
13.37 g/t is determined as the top-cut value. When +2 standard deviations of the mean is applied,
determined as the top-cut value. When +2 standard deviations of the mean is applied, the grade of
the grade of 14.29 g/t is determined as the threshold value. In the following, each top-cut method
14.29 g/t is determined as the threshold value. In the following, each top-cut method in the text is
in the text is called prob (probability) top-cut, perc (95% percentile) top-cut, and std (+2 standard
called prob (probability) top-cut, perc (95% percentile) top-cut, and std (+2 standard deviations)
deviations) top-cut. In these methods, if the absolute value is above the threshold for top-cut, they are
top-cut. In these methods, if the absolute value is above the threshold for top-cut, they are all
all evaluated as the same outlier, and the values determined by the respective methods are shown
evaluated as the same outlier, and the values determined by the respective methods are shown on
on the probability plot (Figure 4). In order to overcome the limitations of these methods, the data
the probability plot (Figure 4). In order to overcome the limitations of these methods, the data
classification by applying the proposed method is shown in Figure 5a. The large red circle represents
classification by applying the proposed method is shown in Figure 5a. The large red circle represents
a clustered outlier where the Au grade is very high as the top 5%, and the z-score is also high as the
a clustered outlier where the Au grade is very high as the top 5%, and the z-score is also high as the
top 10%. On the other hand, a large yellow circle represents a relatively isolated outlier where the
top 10%. On the other hand, a large yellow circle represents a relatively isolated outlier where the Au
Au grade is high as the top 5%, but the z-score is less than the top 10%. The classification results are
grade is high as the top 5%, but the z-score is less than the top 10%. The classification results are
displayed in three dimensions in Figure 5b. Large yellow circles are scattered relatively sporadically,
displayed in three dimensions in Figure 5b. Large yellow circles are scattered relatively sporadically,
whereas large red circles are densely distributed.
whereas large red circles are densely distributed.

Figure
Figure 4.
4. Probability
Probability plot
plot of
of Au
Au grade
grade and
and threshold
threshold values
values of
of existing
existing top-cut
top-cut methods.
methods.
Minerals
Minerals 2018,
2018, 8,
8, x348
FOR PEER REVIEW 88of
of 13
14

Figure 5. (a) Classification of outliers by considering hot spot analysis and (b) the result displayed in
Figure 5. (a) Classification of outliers by considering hot spot analysis and (b) the result displayed in
three dimensions.
three dimensions.

3.2. Variogram
3.2. Variogram Modeling
Modeling and
and Kriging
Kriging Analysis
Analysis
Figure 66 shows
Figure shows the the results
results of
of the
the experimental
experimental variogram
variogram analysis
analysis andand theoretical
theoretical variogram
variogram
modeling for all borehole data or capped data by the respective top-cut
modeling for all borehole data or capped data by the respective top-cut method, which is required method, which to is
required to perform spatial interpolation by using ordinary kriging. The
perform spatial interpolation by using ordinary kriging. The variogram modeling was performed in variogram modeling was
performed
SGeMS in SGeMS
software software
(version 2.1b, (version
Stanford 2.1b, Stanford
University, University,
Stanford, CA,Stanford, CA, United
United States), whichStates),
is an
which is an open-source computer package for solving problems
open-source computer package for solving problems involving spatially related variables. An involving spatially related
variables. Anmodel
exponential exponential model as
was selected wasa selected as a fitting
fitting function for function for the variogram
the theoretical theoretical variogram
model. Since model.
the
Since the remaining data are the same, except for the capped data, the overall
remaining data are the same, except for the capped data, the overall trend of each variogram is trend of each variogram
is similar.
similar. TheThe nugget
nugget value
value andand
thethe range
range werewere all applied
all applied equally,
equally, andandthe the
rangerange
valuevalue
waswas set100
set to to
100 m. When all data are used (Figure 6a), the sill value is the largest, and the
m. When all data are used (Figure 6a), the sill value is the largest, and the lower the threshold value lower the threshold
value
of the of the outlier,
outlier, the smaller
the smaller the sillthe sill value
value becausebecause the variation
the variation of Auofgrade
Au gradevaluesvalues is minimal.
is minimal. The
The distribution of mineral deposits and Au grades can be anisotropic and variograms
distribution of mineral deposits and Au grades can be anisotropic and variograms can be different, can be different,
depending on
depending on dips
dips and
and directions.
directions. However,
However, the the purpose
purpose of of this
this study
study isis to
to evaluate
evaluate thethe difference
difference
between the proposed method and the existing top-cut methods for grade
between the proposed method and the existing top-cut methods for grade estimation. Therefore, an estimation. Therefore,
an omnidirectional
omnidirectional model
model waswas considered
considered assuming
assuming thethe isotropy
isotropy forfor
thethe control
control of the
of the variables.
variables.
Minerals 2018, 8, 348 9 of 14
Minerals 2018, 8, x FOR PEER REVIEW 9 of 13

Minerals 2018, 8, x FOR PEER REVIEW 9 of 13

Figure
Figure 6. 6.
TheThe experimentalvariogram
experimental variogramand
and theoretical
theoretical variogram
variogramfor for(a)
(a)all
allborehole
boreholedata
dataoror
capped
capped
data by (b) prob top-cut, (c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut.
data by (b) prob top-cut, (c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut.
Figure 6. The experimental variogram and theoretical variogram for (a) all borehole data or capped
Ordinary
data by (b) kriging was (c)
prob top-cut, applied for estimating
perc top-cut, the and
(d) std top-cut, Au (e)
grades of top-cut.
hot spot the quartz vein. The highest
Ordinary
predicted kriging
values are was
widelyapplied for estimating
distributed in the case theofAu grades
using of the
all data quartzallvein.
because data The highest
including
predicted Ordinary
outliers are
values kriging
used forwidely
are thewas applied
kriging for estimating
prediction
distributed in (Figure
the case the
7a).ofAu grades
Inusing
the all of
case of the
data quartz
top-cut
because vein.
usingallthe The
data highest
probability
including
predicted
outliers are values
used forare
the widely
kriging distributed
prediction in the
(Figure case7a).of using
In the all
case data
plot, the outliers are capped the most because of the lowest threshold value. It can be seen that the
of because
top-cut all
using data
the including
probability
outliers
plot, are used
distribution
the outliers forvalues
of high
are the kriging
capped at the prediction
bothmost
ends because (Figure
and center 7a).
ofofthe In
thelowestthe case
quartz veinof top-cut using
is considerably
threshold value. the probability
Itreduced
can be (Figure
seen that
plot,
the7b). the outliers are
In the caseofofhigh
distribution capped
applying
values the
top-cutmost
at both because
using standard
ends of the
and center lowest
deviation, threshold
of thethe value.
threshold
quartz It
vein value can be
is theseen
is considerably that
highest, the
so
reduced
distribution
the number
(Figure 7b). In ofthe
of highcase
values
outliers at both
excluded
of applying is ends and center
thetop-cut
lowest and
using of standard
there theisquartz vein is considerably
no significant
deviation, difference inreduced
the threshold (Figure
the distribution
value is the
7b).
highest, Insothethe
of predicted case of applying
values
number with top-cut
all
of outliers data using standard
(Figure
excluded is7d).
theWhendeviation,
lowest the
andupperthe threshold
there 5%
is novalue value is the
is capped,
significant highest, insothe
the result
difference
the number
corresponds
distribution of outliers
to the middle
of predicted excluded
valuesof withis the
the two lowest
top-cut
all data and there
methods
(Figure is no significant
mentioned
7d). When difference
above5%
the upper (Figure in
value7c). the distribution
The predicted
is capped, the result
of predicted
values at both values
ends ofwith all data
the of
quartz (Figure
vein 7d). When
were lower compared the upper 5% value
to the case is capped,
of (Figure
using all7c).
data. the result
Finally, in
corresponds to the middle the two top-cut methods mentioned above The predicted
corresponds
the case of to the middle
applying the hot of thetop-cut
spot two top-cut
proposed methods
in thismentioned
study, the above (Figure
clustered 7c). The
western anomaliespredicted
are
values at both ends of the quartz vein were lower compared to the case of using all data. Finally,
values
maintainedat both
to ends
some of the quartzthe
extent vein were lower compared to the case of using all data. Finally, in
in the case of applying the hotand predicted
spot top-cut values
proposed inofthis
the study,
isolated theeastern
clusteredanomalies
westernare predicted
anomalies are
the case
to be lowof(Figure
applying7e).the hot spot when
Therefore, top-cut proposed
applying theinproposed
this study, hotthe clustered
spot top-cutwestern
method,anomalies are
it is possible
maintained to some extent and the predicted values of the isolated eastern anomalies are predicted to
maintained
to preservetothe some extent and of
information the important
predicted values of the isolated
mineralized areas ineastern anomaliesofarethe
consideration predicted
spatial
be to
low
be
(Figure
low
7e). Therefore,
(Figure 7e).
when
Therefore, when
applying
applying
thethe
proposed
proposed
hothot
spot
spot
top-cut
top-cut
method,
method,
ititisispossible
possible
to
distribution, not merely by excluding the values with high absolute values.
preserve the information of important mineralized areas in consideration
to preserve the information of important mineralized areas in consideration of the spatial of the spatial distribution,
notdistribution,
merely by excluding
not merelythe by values
excluding withthehigh
valuesabsolute
with high values.
absolute values.

Figure 7. The kriging results of Au grade estimation using (a) all data or capped data by (b) prob
top-cut, (c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut.
Figure 7. The kriging results of Au grade estimation using (a) all data or capped data by (b) prob
Figure 7. The kriging results of Au grade estimation using (a) all data or capped data by (b) prob
3.3. Validation
top-cut, (c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut.
top-cut, (c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut.
3.3. Validation
Minerals 2018, 8, 348 10 of 14

3.3. Validation
Minerals 2018, 8, x FOR PEER REVIEW 10 of 13
To validate the estimation, it is necessary to evaluate whether the global mean of the estimated
block grades To validate
and the thedata
estimation, it is necessary
are identical. For to evaluate
this whether
purpose, the global mean
a histogram can be of useful
the estimated
(Figure 8).
block grades and the data are identical. For this purpose, a histogram can be
Figure 8a shows the results of the kriging prediction using all the assay data, and Figure 8b–d showsuseful (Figure 8). Figure
8a shows the
the prediction results
results of the
using kriging
the existingprediction
top-cutusing all theFigure
methods. assay data, and Figures 8b–d
8e corresponds to theshows the
histogram of
prediction
kriging results
predictions using
using thethe
hotexisting top-cutmethod
spot top-cut methods. Figure 8e
proposed in corresponds
this study. Each to the of histogram of
these histograms
is thekriging
result predictions
for an estimatedusing the hot blocks,
26,676 spot top-cut
whereasmethod proposed
Figure 8f shows in this
the study.
histogram Each for
of these
132 assay
histograms is the result for an estimated 26,676 blocks, whereas Figure 8f shows the histogram for
samples corresponding to the quartz vein. The overall trend of the histogram of kriging predictions is
132 assay samples corresponding to the quartz vein. The overall trend of the histogram of kriging
similar to the histogram of assay data, but both ends of the distribution are reduced by the smoothing
predictions is similar to the histogram of assay data, but both ends of the distribution are reduced by
effectthe
duesmoothing
to the natureeffectofdue
the to
interpolation
the nature of method [17]. Especially
the interpolation method when[17].the threshold
Especially value
when theof the
outlier is decreased by using the existing top-cut methods, it can be seen that
threshold value of the outlier is decreased by using the existing top-cut methods, it can be seen that the higher predicted
valuesthearehigher predicted
decreased. values
In the are decreased.
std top-cut method In(Figure
the std 8d)
top-cut
withmethod
a large (Figure
threshold 8d)value
with aand large
the hot
thresholdmethod
spot top-cut value and the hot8e)
(Figure spot top-cut method
considering (Figure
clustered 8e) considering
outliers, relativelyclustered outliers,are
high values relatively
maintained
in thehigh values are
prediction maintained
process. Tablein 1the prediction
shows process. Table
the minimum, 1 shows average,
maximum, the minimum, maximum,
skewness, and total
average, skewness, and total Au amount for each kriging prediction
Au amount for each kriging prediction and assay data. The average value is most similar to the and assay data. The average
value is most similar to the sample data when all data are utilized. On the other hand, the skewness
sample data when all data are utilized. On the other hand, the skewness is most different from the
is most different from the sample data. The total amount of Au in the vein was calculated by
sample data. The total amount of Au in 3the vein was calculated by considering the volume of block
considering
3 the volume of block (27 m ), number of blocks (26,676), 3 density of rock (2.5 g/cm3), and
(27 maverage
), number of blocks (26,676), density of rock (2.5 g/cm ), and average of predicted Au grade.
of predicted Au grade. Because one prediction method may not have the best results in all
Because one prediction
respects, it is necessary method may notand
to approach have the best
review results
various in all respects,
methods. it is the
In addition, necessary
validitytoofapproach
the
prediction
and review cannot
various be verified
methods. by the global
In addition, estimation
the validity ofresult alone. Therefore,
the prediction cannotthe be verification
verified byof theglobal
the
local estimation
estimation is alsoTherefore,
result alone. necessary. the verification of the local estimation is also necessary.

Figure 8. Histograms of Au grade estimation using (a) all data or capped data by (b) prob top-cut, (c)
Figure 8. Histograms of Au grade estimation using (a) all data or capped data by (b) prob top-cut,
perc top-cut, (d) std top-cut, and (e) hot spot top-cut, and (f) histogram of assayed Au grade data.
(c) perc top-cut, (d) std top-cut, and (e) hot spot top-cut, and (f) histogram of assayed Au grade data.
Table 1. Minimum, maximum, average, skewness, and total amount of Au in the vein for each
1. Minimum,
Tablekriging maximum,
prediction and assay average,
data. skewness, and total amount of Au in the vein for each kriging
prediction and assay data.
Total
Minimum Maximum Average
Skewness Amount of
(g/t)
Minimum (g/t)
Maximum (g/t)
Average Total Amount of
Skewness Au (t)
(g/t) (g/t) (g/t) Au (t)
All data 6.25 11.76 8.63 0.44 15.54
All data 6.25 11.76 8.63 0.44 15.54
Prob top-cut 6.37 10.39 8.30 0.14 14.95
Ordinary Prob top-cut 6.37 10.39 8.30 0.14 14.95
Perc top-cut 6.38 10.86 8.44 0.21 15.13
kriging Perc top-cut
kriging Std top-cut
6.38 10.86 8.44 0.21 15.13
Ordinary
Std top-cut 6.29
6.29 11.15
11.15 8.55
8.55 0.26
0.26 15.40
15.40
Hot spot
Hot top-cut
spot 6.31 11.31 8.53 0.34 15.36
6.31 11.31 8.53 0.34 15.36
Assayed borehole data
top-cut 1.79 14.63 8.80 −0.50 -
Assayed borehole data 1.79 14.63 8.80 −0.50 -
Minerals 2018, 8, x FOR PEER REVIEW 11 of 13
Minerals 2018, 8, 348 11 of 14
Minerals 2018, 8, x FOR PEER REVIEW 11 of 13
In order to verify the local predicted grade, a method of comparing the predicted value of the
In
blockInand order
order the toborehole
to verifythe
verify the local
data
local forpredicted
the panels
predicted grade,
grade, a method
traversing
a method of comparing
thecomparing
of entire orethe body the can
predicted
predicted bevalue value
of theof
generally the
used.
block
block
These
and and
thepanels the
borehole borehole
are data
typically data
for thedrawnfor
panels the
along panels traversing
the strike
traversing of the ore
the entire the entire
ore body
body can ore
andbe body
contain can
generally be
oneused. generally
or more These used.
boreholes
panels
These panels are typically drawn along the strike of the ore body
are typically drawn along the strike of the ore body and contain one or more boreholes [24,25]. Indata
[24,25]. In this study, 12 panels were constructed, as shown in and
Figure contain
9, to one
include or more
all boreholes
borehole this
[24,25].
study, 12Inpanels
and average thisgrades
study,
werewere12 panels
constructed, were
calculated constructed,
for
as shown each in panel.
Figure asTable
9,shown in Figure
2 shows
to include allthe 9,mean
boreholeto include
values
data andall borehole
ofaverage
borehole data
data
grades
and
and average
were kriging
calculated grades
predictions were
for each calculated
for
panel.eachTable 2for
panel. eachthe
They
shows panel.
are Table
plotted
mean on2 ashows
values line
of the mean
diagram
borehole values
against
data of borehole
each
and kriging panel data
(Figure
predictions
and
for kriging
10).each
According
panel.predictions
to theare
They for each panel.
smoothing
plotted effect
on They
a line theare
of diagram plotted
kriging on aeachline panel
prediction,
against diagram
the value against
(Figure 10).each
is predicted panel
to be(Figure
According low
to thein
10). According
the part where
smoothing effecttheto the smoothing
of value effect
of theprediction,
the kriging of
borehole data the
the is kriging
highisand
value prediction,
the value
predicted the
to be value
is low
predicted is predicted
in the to bewhere
part to
high in be low
the
the in
part
value
the
where
of part
the where
the valuethe
borehole of value
datathe of the
is borehole
high andborehole
data data
is low
the value is high
is[17]. Thisand
predicted tothe
be value
smoothing is the
higheffect
in predicted
is part
a generalto be
where highvalue
phenomenon,
the in theofpart
but
the
where
borehole the
the smaller value
data of the
theisdegree,
low [17]. borehole
the better
This data is low
the prediction
smoothing [17]. This
effect isresult. smoothing
a general Although effect is
the result
phenomenon, a general
butofthe
usingphenomenon,
all data
smaller the is but
most
degree,
the
the smaller
similar
betterto the
thethe degree,
result
prediction the
of the better
borehole
result. thedata,
Although prediction
there
the isresult.
result aofrisk Although
all datathe
of overestimation.
using resultsimilar
is most of using
Existing all
thedata
totop-cut isofmost
methods
result the
similar
can to the
underestimate result of
panelsthe borehole
2–4 as the data, there
threshold is a
value risk of of
the overestimation.
outlier
borehole data, there is a risk of overestimation. Existing top-cut methods can underestimate panels 2–4 decreases. Existing
In top-cut
particular, methods
the prob
can
as underestimate
top-cut
the method value
threshold panels
with the
of the2–4outlier
lowest as the threshold
threshold
decreases. value value
In tends of the
particular, outlier
to underestimate
the prob decreases.
top-cut In
Aumethod particular,
grade as
with thelowest
a whole.
the prob
For
top-cut method
panel 11, value
threshold with
overestimation the lowest threshold
is prominent
tends to underestimate Auwhen value
gradeusing tends
as a whole. to underestimate
all data
For andpanelthe 11,std Au grade as
top-cut method.
overestimation a whole. For
It was
is prominent
panel
confirmed
when 11, overestimation
using that
all datatheandhotthe isstd
spot prominent
top-cut when using
method
top-cut method. can
It was all
mitigate datathe
confirmed and the
thestdhot top-cut
overestimation
that of method.
spot top-cut panel 11Itwhile
method was
can
confirmed
evaluating
mitigate that
thepanels the hot spot top-cut
2–4 as a significant
overestimation method
of panel 11mineralized can
while evaluating mitigate the overestimation of
zone. panels 2–4 as a significant mineralized zone. panel 11 while
evaluating panels 2–4 as a significant mineralized zone.

Figure 9. Constructed 12 panels which contain boreholes for validating local estimation.
Figure 9.
Figure Constructed 12
9. Constructed 12 panels
panels which
which contain
contain boreholes
boreholes for
for validating
validating local
local estimation.
estimation.

Figure 10. Validation of the Au grade estimation for each panel using the line-diagram.
Figure
Figure 10.
10. Validation
Validation of
of the
the Au
Au grade
grade estimation
estimation for each panel
for each panel using
using the
the line-diagram.
line-diagram.
Table 2. The average of kriging predictions and borehole data for each panel.
Table 2. The average of kriging predictions and borehole data for each panel.
Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel
Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel
Minerals 2018, 8, 348 12 of 14

Table 2. The average of kriging predictions and borehole data for each panel.

Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel Panel
01 02 03 04 05 06 07 08 09 10 11 12
A1 7.70 9.75 10.64 9.41 8.40 7.63 8.43 9.48 7.51 8.41 8.93 9.46
B2 7.59 8.76 9.79 8.65 8.35 7.70 8.42 9.12 7.51 7.94 8.31 8.56
C3 7.56 8.77 10.06 8.68 8.26 7.68 8.45 9.48 7.57 8.39 8.43 9.26
D4 7.52 9.23 10.37 8.97 8.42 7.64 8.44 9.52 7.53 8.42 8.93 9.47
E5 7.73 9.47 10.52 8.97 8.34 7.65 8.44 9.48 7.54 8.38 8.43 9.26
F6 7.53 9.95 11.04 9.40 8.37 7.92 8.49 9.62 6.57 8.64 8.49 9.59
Ordinary kriging using 1 all data, 2 prob top-cut method, 3 perc top-cut method, 4 std top-cut method, 5 hot spot
top-cut method and 6 assayed borehole data.

4. Conclusions
In this study, a 3D hot spot analysis technique for quantifying statistically significant data was
developed and applied to borehole data analysis of the Au deposit. The resource estimation using
kriging is very useful with the advancement of computer technology. However, the presence of
outliers, which are much higher than other values, can lead to overestimation in the kriging prediction
process, which can cause great problems in assessing the resources. In order to prevent the effect
of overestimation, various top-cut methods are used to manage outliers in the analysis. However,
existing methods have a limitation that they can consider only a certain threshold value.
Even if the outliers have the same grade value, the actual importance may be different depending
on whether they are clustered outliers that are spatially connected or isolated outliers that do not have
spatial connectivity. The isolated outliers may be the main cause of overestimation, while the clustered
outliers may represent the important mineralization. Therefore, it is necessary to evaluate spatial
density, that is, spatial statistical significance. The density of Au grade in 3D space is evaluated as the
Getis–Ord Gi* statistic by applying the hot spot analysis method proposed in this study. In this study,
a new top-cut method was proposed and applied to the Au grade data of the Au deposit. When using
the existing top-cut methods, such as probe top-cut, perc top-cut, and std top-cut, Au grade data higher
than a specific value was excluded in the kriging analysis. On the other hand, statistically significant
hot spot data could be included by hot spot top-cut in the kriging analysis. As a result, we confirmed
that it is possible to mitigate the overestimation or underestimation of a local area that might occur
when applying the existing methods.
The results of this study can contribute to securing significant data in the prediction of resources
using kriging. In the assessment of resources, the top-cut process has no established criteria and
is subject to analyst discretion. The proposed hot spot top-cut method is also subject to evaluation
depending on how the outlier threshold is determined and how the threshold value of the z-score is
determined. However, it is possible to add an option for the analyst’s judgment, in that it provides
a new field of view for areas that cannot be found by conventional methods. In this study, only
the isotropy was considered for evaluating the correlation between data, but it is expected that the
analytical application and accuracy will be improved if further anisotropy is considered in the future.
Also, if the results are verified and the analysis method is improved by comparing it to the production
amount, the proposed method will be more useful. The proposed method is expected to be applicable
not only to underground data, but also to atmospheric and underwater data.

Author Contributions: Y.C. conceived and designed the experiments; S.-M.K. and Y.C. performed the experiments;
S.-M.K. developed the software and analyzed the data; H.-D.P. contributed reagents/materials/analysis tools;
and S.-M.K. and Y.C. wrote the paper.
Funding: This work was supported by (1) Basic Science Research Program through the National Research
Foundation of Korea (NRF) funded by the Ministry of Education (2015R1D1A1A01061290), (2) the Brain Korea 21
Project, and (3) the Research Institute of Energy and Resources, Seoul National University, Korea.
Conflicts of Interest: The authors declare no conflicts of interest.
Minerals 2018, 8, 348 13 of 14

References
1. Mallet, J.L. Geomodeling Applied Geostatistics Serie; Oxford University Press: New York, NY, USA, 2002.
2. Zanchi, A.; Francescac, S.; Stefano, Z.; Simone, S.; Graziano, G. 3D reconstruction of complex geological
bodies: Examples from the Alps. Comput. Geosci. 2009, 35, 49–69. [CrossRef]
3. Bullock, R.L.; Mernitz, S. Mineral Property Evaluation: Handbook for Feasibility Studies and Due Diligence; Society
for Mining, Metallurgy & Exploration: Englewood, NJ, USA, 2017.
4. Valley, M.; David, M.; Dagbert, M.; Desrochers, C. Guide to the Evaluation of Gold Deposits; Geological Society
of CIM: Calgary, AB, Canada, 1992.
5. Sinclair, A.J. A fundamental approach to threshold estimation in exploration geochemistry: Probability
graphs revisited. J. Geochem. Explor. 1991, 41, 1–22. [CrossRef]
6. Parker, H.M. Statistical treatment of outlier data in epithermal gold deposit reserve estimation. Math. Geol.
1991, 23, 175–199. [CrossRef]
7. Sinclair, A.J.; Garston, H.B. Applied Mineral Inventory Estimation; Cambridge University Press: Cambridge,
UK, 2006.
8. Rossi, M.E.; Clayton, V.D. Mineral Resource Estimation; Springer: Berlin, Germany, 2013.
9. Getis, A.; Ord, J.K. The analysis of spatial association by use of distance statistics. Geogr. Anal. 1992, 24,
189–206. [CrossRef]
10. McClintock, N. Assessing soil lead contamination at multiple scales in Oakland, California: Implications for
urban agriculture and environmental justice. Appl. Geogr. 2012, 35, 460–473. [CrossRef]
11. Griffith, D.A.; Chun, Y.; Lee, M. Locational Error Impacts on Local Spatial Autocorrelation Indices: A Syracuse
Soil Sample Pb-Level Data Case Study. In Proceedings of the Spatial Accuracy Assessment in Natural
Resources and Environmental Sciences, Montpellier, France, 5–8 July 2016.
12. Kim, S.M.; Choi, Y. Assessing statistically significant heavy-metal concentrations in abandoned mine areas
via Hot Spot Analysis of portable XRF data. Int. J. Environ. Res. Public Health 2017, 14, 654. [CrossRef]
[PubMed]
13. Rose, W.L.; Burge, C.; Davis, B.; Kozak, A.; Sim, R.; Wells, G.S. Mina De Cobre Panamá Project, Panamá, NI
43-101 Technical Report; Inmet Mining Corporation: Toronto, ON, Canada, 2010.
14. Sepp, M.D.; John, H.D. Structural evolution, vein orientation, and paragenesis of the botija porphyry
Cu-Mo-(Au) deposit, Panama. Econ. Geol. 2018, 113, 857–890. [CrossRef]
15. Gray, D.; Lawlor, M.; Stone, R. Cobre Panamá Project, Colón Province, Republic of Panamá, NI 43-101 Technical
Report; First Quantum Minerals Limited: Vancouver, BC, Canada, 2015.
16. Pan, G. Practical issues of geostatistical reserve estimation in the mining industry. CIM Bull. 1995, 88, 31–37.
17. Abzalov, M. Applied Mining Geology; Springer: Berlin, Germany, 2016.
18. Environmental Systems Research Institute (ESRI). ArcGIS Desktop Help 10.2 How Hot Spot Analysis
(Getis–Ord Gi*) works. Available online: http://resources.arcgis.com/en/help/main/10.2/index.html
(accessed on 1 July 2018).
19. Wang, G.; Huang, L. 3D geological modeling for mineral resource assessment of the Tongshan Cu deposit,
Heilongjiang Province, China. Geosci. Front. 2012, 3, 483–491. [CrossRef]
20. Shahbeik, S.; Afzal, P.; Moarefvand, P.; Qumarsy, M. Comparison between ordinary kriging (OK) and
inverse distance weighted (IDW) based on estimation error. Case study: Dardevey iron ore deposit, NE Iran.
Arab. J. Geosci. 2014, 7, 3693–3704. [CrossRef]
21. Yamamoto, J.K. Quantification of uncertainty in ore-reserve estimation: Applications to Chapada copper
deposit, State of Goiás, Brazil. Nat. Resour. Res. 1999, 8, 153–163. [CrossRef]
22. Badel, M.; Angorani, S.; Panahi, M.S. The application of median indicator kriging and neural network in
modeling mixed population in an iron ore deposit. Comput. Geosci. 2011, 37, 530–540. [CrossRef]
23. Lemmer, I.C. Estimating local recoverable reserves via indicator kriging. In Geostatistics for Natural Resources
Characterization; Springer: Dordrecht, The Netherlands, 1984; pp. 349–364.
Minerals 2018, 8, 348 14 of 14

24. Abzalov, M. Quality control of assay data: A review of procedures for measuring and monitoring precision
and accuracy. Explor. Min. Geol. 2008, 17, 131–144. [CrossRef]
25. Abzalov, M.; Bower, J. Geology of bauxite deposits and their resource estimation practices. Appl. Earth Sci.
2014, 123, 118–134. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).