Coal resource classification and geostatistics

By KEITH 0 WHITCHURCH 1 , Graduate Member, AD STEWART GILLlES 2 , Member and

GEOFFREY 0 JUSr, Fellow

for
uranium
and
thorium
where
Nuclear
Energy
Authority/Intemational Atomic Energy Authority (NEW/IAEA)
guidelines have been adopted. For all other commodities the
approach taken varies from country to country and often from state
to state with different methods used for different commodities.
Most resources are however, classified according to two sets of
criteria, geological assurance and economic facility.
The
relationship between them has been summarised in the well known
McKelvy Box, Figure 1 (McKelvy, 1972).
The principal

ABSTRACT
An internationally recognised and unifonn method for classification,
categorisation and designation of mineral and energy resources is not yet
available. With the increasing need for reliable and comparable coal
resource data it is necessary to standardise the traditional classification
procedures by quantifying the three basic evaluation criteria of economic
feasibility, geologic assurance and recovery. For well documented deposits,
geostatistical methods can considerably improve classification quality. A
review is undertaken of major classification with particular emphasis on
those that incorporate limits on estimation error.
A geostatistically based algorithm for classifying coal resources within the
current Queensland and New South Wales codes has been developed.
Application of the classification algorithm for resources from a number of
seams exhibiting different structural characteristics is assessed. The method
is found to give classification results that closely reflect the error associated
with an estimate of resource quantities based on current sampling densities.
lbis requires a careful geostatistical analysis with an emphasis on
geological awareness. Resource category restrictions although albitrarily
assigned should remain constant for all deposits being compared. This
requires some engineering judgement with additional interpretation also
required for isolated and peripheral block.
Keywords: coal, geostatistics, resource classification.

IDENTifiED

o
z

o
~

...on,
...

...'"
u

:i

o
z
o
u

INTRODUCTION

o
z

..,
u

The economic incentive to assess and classify mineable coal

resources so as to enable the calculation of reserve tonnage and
grade has increased significantly with present international
commodity marketing conditions. Improvements to classification
concepts and defmitions are needed as it is recognised that no
current practice produces results that are free of some degree of
subjectivity and therefore readily reproducable. Quantification of
geological assurance is a major difficulty in most classification
costs. The magnitude of the error associated with an estimate of the
quantity and quality of a resource needs to be understood.
Geostatistics allows calculation of the variance of errors associated
with an estimate and is a potentially valuable tool for classifying
resources on the basis of geological assurance.
In a review of major classification codes particular emphasis was
placed on those addressing geological assurance. Codes which
incorporate limits on estimation error were analysed in conjunction
with a study of the impact that geostatistics may have in developing
an understanding in this area.
A geostatistically based algorithm for classifying coal resources
within the framework of the current Queensland and New South
Wales codes was specifically developed to compare results with
current practice.
Seams exhibiting different structural
characteristics were used as case studies to assess the sensitivity of
the algorithm in classifying resources.

RESOURCE CLASSIFICATION PRINCIPLES

There is no internationally recognised and uniform method for
classification, categorisation an designation of mineral and energy
resources. The one exception to this is the classification systems
1. Tutorial Fellow in Mining Engineering
2. Senior Lecturer in Mining Engineering
3. Reader in Mining Engineering, University of Queensland, St Lucia Qld
4067.
4. Original manuscript received on 11 May 1987

The AusIMM Proceedings

UNDISCOVERED

...:>
'"

...'"o

.."'"'
C>

"'o
C>

on

."'
<t

FIG 1 - McKelvy box scheme (as recommended by the

USBM/USGS) the hatched area represents that portion to which
geostatistical methods may be applicable.
differences between codes lies in the defmition of these criteria.
This paper principally addresses resource classification on the basis
of geological assurance.
The degree of certainty with which an estimate of material in the
ground can be made is dependent on the amount of exploration
undertaken and the nature of the deposit. The different codes
recognise a range of geological assurance from certainty to extreme
speculation. Consequently a diverse body of opinion exists on how
geological assurance should be assigned to an estimate or on how a
resource classification scheme should be subdivided on the basis of
that assurance. Table 1 has been prepared to summarise the
categories of geological assurance for a number of codes. It should
be noted that owing to the diversity of class defmitions a strict
comparison of these groupings is not possible. Table 1 should
therefore be used as a guide only. This does, however, highlight the
difficulty in comparing resource estimates based on different
classification codes.
Categories of assurance are usually defined in general terms only.
For example an extract from the Bureau of Mineral Resources
classification system (BMR-1984) defines:
Measured Resources:

No 2 t990

Those resources for which the sites for

inspection, sampling and measurement
are spaced so closely that their shape
and
mineral content are
well

K D WHITCHURCH, ADS GILLIES AND G D JUST

TABLE 1
Comparison ofcategories ofgeological assurance.

USBM
SGS
AUSIMM
AMIC
GDMB
1959
SSR

OLD
1985
OLD
1978
SOUTH
AUSTRALIA

DE~O

M~5VoRED
MEASURED
PROVED

"s, gal

STANDARD
DRAFT
CA ADA
1975
GDR
1982
AUSTRIAN
COAL

INFERRED

INDICA TED
PROBABL.E

INFERRED
POSSIBLE

ERR lOI
...ss 10-g01

r"~'~~~O'

Cl

MEASLRED
013T <

INDICATED
DIU <

I~

MEASURED

Ol.sr

INDICA TED

MEASURED
015T <
~

< ZJQ.4

PRE ORE RESER/ES

SAGE
1.... 1~~ERALlSAT CNI

0
PROGNOSTIC

ERR 301

AS' 10-)0'

.. ss

)O-~O'

C2

INFE'lRED

f-'''

INFERRED
OI$T <

41(),4

INFERRED

013T < ZJO,(

I~

le,

INDICATED

MEASURED

IN SIGHT PROBABLE

INFERRED

SURMISED

sPEC ...'l..A

U~ r<~

FROS""CT'\E

.1=

CM

SP:C~L~- f~

,NFERRED

INDICA 1E~
B

HYPOTHETICAL

t,; 'CERTAIN

POSSIB E

DEMONSTRATED
MIO~V~oiv

R-3

R-2

R-l

C2

Cl

A
IN SIGHT

PROBABLE

Cl
INDICATED

ASS eO-IDOl

ASS '0801

"'" <'01

established.
Those resources for which the degree of
assurance, although lower than for
resources in the measured category, is
high enough to assume continuity
between points of observation.
Resource classification codes of this type give rise to subjective
estimates that are rarely reproducable, owing to the diversity of
experience and opinion among practitioners. Such estimates are
difficult to compare due to their subjective nature. A number of
organisations have attempted to improve on this situation by
placing more exact requirements on categories of assurance. One
approach has been to place maximum error limits on categories.
For example under the classification systems of the United States
Bureau of Mines/United States Geological Survey, (USBM/USGS,
1971), the Canadian Department of Energy Mines and Resources
(EMR, 1975) and the South Australian Department of Mines and
Energy (Wigglesworth, 1981) estimates of proved or measured
resources must be accurate within 20 per cent of the true value.
None of these codes, however, offer any suggestions as to how a
practitioner is to decide the error of a particular estimate.
Consequently, subjectivity remains an integral part of the estimation
process. Recent trends have seen the removal of error limits from
such codes as the USBM/USGS (1980) and Queensland (Galligan
and Mengle, 1986). Another approach has been to specify either
levels of confidence only (Fettweis, 1979) or a combination of
levels of confidence and error limits such as the Gesselschaft
Deustcher Metallhutten and Bergluete (GDMB system, 1959) for
categories of assurance. Wellmer (1983) pointed out that level of
confidence was originally an estimated geological factor and hence
a subjective figure. It was only later that this was interpreted and
used as exactly dermed levels of confidence in classical statistical
calculations. A geostatistical approach to the GDMB system has
been proposed by Wellmer (1983).
Indicated Resources:

.ro>OT1'1ETlCAl.

C2
INFERRED

IOI'~
'ww

2JO,oI

INDICATED

OI.sT ( II(M
~ ZOI

ESC

BMR

INDICATED

Cl
A
B
IN SIGHT PROBABLE INOICA Er>
ERR 101

I
I SPEc...U r E

STRATED

C2
I~F,,"RED

Some codes such as those of South Australia, Canada and

Queensland recommend maximum sample spacings for various
categories of assurance, while recognising that the level of
assurance attained for a fixed drillhole spacing will vary with
deposit types. This approach sets maximum limits only and does
not remove any subjectivity from estimates.
It is clear that classification on the basis of assurance is important
and difficult and in the opinion of Fettweis (1979), "more energetic
endeavours than hitherto, should be made to determine the degrees
of assurance in figure".

RELATIVE PRECISION AS A DEFINITIO OF

RESOURCE CLASSES

E~or ) as a defmition of the

Esurnate
limits of resource classes has been advocated by a number of
authors including Fettweis (1979), Deihl and David (1982),
Froidevaux (1982, 1983) and Wellmer (1983). It is also the
approach used by all organisations that stipulate error limits for
resource classes in their classification codes (Table 1).
Wellmer (1983) notes, however, that relative precision is
meaningless without a corresponding statement on confidence
limits, such as that in the GDMB system of 1959. Deihl and David
(1982) point out that the first step in applying confidence intervals
is the determination of the distribution of errors involved in an
estimation.
Kriging defines two parameters of the error
distribution.
The mean,
The use of relative precision (

E[e(V)]=O=~

(1)

The variance,
Var [2(V)] =

021990

ok2

(2)

TheAusIMM Proceedings

COAL RESOURCE CLASSIFICATION AND GEOSTATISTICS

However, this does not give information on the form of the

distribution. Comparisons between estimated grades and actual
grades (after mining) have shown the distribution of errors (Ugarte,
1972) to closely approximate a Gaussian distribution with a meanzs
o equal to zero and a variance f:1, equal to the kriging variance 0-,
k
(Figure 2). This approximation has been shown to be closest for a
9<1 per cent confidence interval and for relative standard deviation
(- ) less than 0.5. If a Gaussian distribution is assumed for

estimation errors then it can be shown that the relative precision of

an estimate may be calculated from equation 1.
(3)

where
R

= relative error at a 95 per cent confidence level,

ok =

although this code is, at first glance very convincing, it does in fact
contain some ambiguities that can be explained by the following
example.
A block in a deposit with an error of nine per cent at a level of
confidence of 90 per cent qualifies for the category of proven
resources. A neighbouring block with an error of 12 per cent at the
same confidence level does not qualify for the category of proven
resources. However, at a confidence limit of 80 per cent, (that for
probable resources) an error of only nine per cent is obtained
(Figure 3).
Wellmer (1983) recommended that the level of confidence be
maintained at a constant level for all classes of assurance with only
the error limits varying. Similar approaches have been advocated
by Deihl and David (1982), Froidevaux (1983) and the GDMB
(Wellmer, 1983). Table 3 shows the new GDMB system that
follows this recommendation.

estimation variance (Kriging variance), and

TABLE 3
ClassifICation system ofthe GDMB/1983 (after We//mer, 1983)

= estimated value of the parameter.

Results of this calculation can be used as a quantitative measure

in the resource classification process.

01

>
v

~ 005-

Category

Upper limit of error

Level of confidence

Proven
Probable
Possible 1
Possible 11
Unclassified

10%
20%
30%
50%

90%
90%

>50%

90%
90%
90

SCOPE AND APPLICABILITY OF GEOSTATISTICS

O/oCu
0'

095

FIG 2 - Histogram of experimental kriging errors of blocks from the

copper deposit of Chiquicamata (Chile). The Gaussian distribution
with the respective mean and variance (dotted line) and the 0.95
per cent confidence interval are shown. Overlapping areas of the
two curves are hatched.

CO FIDENCE LIMITS AND RESOURCE

CLASSIFICATIO CODES

The Influence of block size on classification

The earliest resources classification code to make use of confidence

intervals and error limits to define categories of geological
assurance was the 1959 (GDMB Code, Table 2). Under this Code
with decreasing assurance classes, the limits of error become wider
and the level of confidence lower. Wellmer (1983) points out that,
TABLE

Classification system ofthe GBMB (1959)

Category

Upper limit of error

Level of confidence

Proven

10%
20%
30%
30

90%
79 -90%
50 -70%
30 - 60%
10 - 30%

Probable
Indicated
Inferred
Prognostic

The AusIMM Proceedings

The use of geostatistics requires quantitative data and certain

minimum restrictions with respect to the number and positions of
sample information. Theoretical reasoning and practical experience
have shown that a minimum of 30 to 50 sample points evenly
distributed over any field of interest are necessary to obtain a
reliable variogram (Diehl and David, 1982). Hence a resource
classification system based on geostatistics must be restricted to
fairly well documented deposits. As a rule of thumb Deihl and
David (1982) suggest that geostatistical methods are only
applicable to those resource classification categories covered in the
hatched area of the McKelvey Box (Figure 1). This roughly
equates to the measured, indicated and inferred class one categories
of the new Queensland Code.

The estimation variance

0i

of any deposit parameter is a function

of the size of the block being estimated as discussed by Deihl and

David (1982). Any block classification that is dependent in
estimation variance is therefore a function of block size. As an
example, using a set of borehole data for an irregularly drilled
brown coal deposit, the relative error of estimation has been
calculated at, with a 95 per cent confidence limit (Equation 1) for
square blocks of varying sizes. The results of these calculations are
shown in Figure 4 and indicate the influence of block size on
kriging variance and hence on the outcome of resource
classification. The percentage error calculated by equation 1 has
been plotted against the length of the block sides expressed as a
percentage of the seam thickness variogram range.
From a statistical point of view it is clear that classification of
resources without considering the relation between the respective
deposit quantity and the estimation variance is meaningless. The
solution to this problem is neither simple nor straightforward and in
fact Sabourin (1983) noted that there are almost as many proposed
solutions as there are authors who have considered the problem. It

N021990

K D WillTCHURCH, ADS Gll..LIES AND G D JUST

LIMIT

0.3

OF RELATIVE
ERRORS

0.2

0.1

PROVEN

POSSIBLE

PROBABLE
DEGREE OF
ASSURANCE

FIG 3 - Undefined areas (AB) in old GDMB-classification of 1959.

C L = Confidence level
A = at 90 per cent level of confidence (probability) above 10 per cent, at 80 per cent level of confidence (probability) below 10 per cent relative
error.
B = at 80 per cent level of confidence (probability) above 20 per cent, at 70 per cent level of confidence (probability) below 20 per cent relative
error.
is beyond the scope of this paper to review all proposed solutions,
although it is worth noting that in many cases interim solutions only
are suggested such as in the case of the GDMB (Wellmer, 1983).

100

10

A definition of maximum block size based on drill hole spacing

(Figure 5) combines some of the better aspects of a number of the
solutions proposed by different authors, while observing the
restrictions inherent in the current Queensland and New South
Wales codes. In this case the maximum block size allowable for
each class of resource is therefore dependent on the maximum drill
hole spacing for that class. This limit on block size is therefore
suggested for use in the proposed geostatistical solution.

PROPOSED CLASSIFICATION CLASS CRITERIA

-0

00

00

The proposed resource class system for use in this study is defmed
by drill hole spacings and error limits. Each class is based on the
maximum drill hole spacing specified in the current Queensland
code. For class A resources the maximum allowable error is based
on the estimation error specified for the most restrictive resource
class (measured) under the Queensland Coal Reserve code in use up
to 1985. Error limits for the less restrictive classes B and C have

PERCfNTAC, 0" RANC,

FIG 4 - Relative kriging error (at 95 per cent confidence level)

versus block side dimension as a percentage ofvariogram range.

TABLE 4

Proposed resource class limits. Drill hole spacing is based on

Queensland and New South Wales codes.

Resource
Class

Maximum
Illowable

Maximum
drill hole

Maximum
block

Block
iuntion

spacing
km

IreI

area

tm2

tm2

1(_1 2)

O.2S
1.00
<4.00

(95" c.1.)
A
B

20
40
60

1
2

(-~)

16(_42

FIG 5 - Polygonal block showing the relationship between drillhole

spacing and block dimensions
10

No21990

The AuslMM ProcecdUtgs

COAL RESOURCE CLASSIFICATIO AND GEOSTATISTICS

been assigned arbitrarily although they follow recommendations put

forward by Fettweis (1979), Diehl and David (1982) and Wellmer
(1983). This information is summarised in Table 4.
Calculations for this geostatistical classification of resources
follow a series of steps as defined by Diehl and David (1982).
1.
Determination of economically relevant parameters such as
coal thickness or ash content, upon which the classification
will be based.
2.
Review of raw data and preparation of basic data flies
3.
Classical statistical analysis of parameter data and variogram
calculation.
4.
Determination of the outline of presently feasible resources
by geological and technical criteria, such as depth or
oxidation limits.
5.
Further sub-division of the areas defmed in point four, into
blocks that satisfy the predefined constraints of a specific
category of geological assurance with respect to:
(a) dimensions, and
(b) parameter confidence levels.
6.
Calculation of in-situ and recoverable tonnages of each
block defmed in point five and compilation of total
quantities for each class of resource.
The above steps are standard well-defmed procedures in practical
geostatistics with the exception of point five which required the
development of a special algorithm. This presents fundamental
problems in the division and definition of blocks to satisfy resource
class constraints in respect of both area and the precision of the
grade estimates. The algorithm needs to ensure that a maximum
resource quantity is assigned to the category with the highest degree
of geological assurance.
The proposed algorithm uses an iterative method starting with a
small block that is enlarged step by step. After each incremental
increase, the area of the enlarged block and associated kriging
variance are calculated and compared with constraints for the first
and most restrictive resource class. If both area:fld confidence
interval calculated from the kriging variance (Ok ) satisfy the

constraints, the block is classified to this restrictive category, class

A. Should the area of the block under study surpass the upper limit
of class A without obtaining the necessary precision, the procedure
continues with the less restrictive requirements of class B and so no
until the block is fmally classified. In order to maximise the
resource quantities in the upper classes, the i~on procedure
begins at the location with the lowest value for ~ determined by
point kriging the deposit on a regular grid. The stepwise extensions
of any block are constrained by the principle that an increase is
always performed iIlihe direction where the gradient of precision or
rate of change of ~ is least. After fmal classification, the block
area is recorded on file, the results of the estimation printed and the
procedure re-starts at another point with local minimum kriging
variance.
A FORTRAN program has been written to perform the
classification automatically for one selected parameter. A series of
checks within this program ensure that the shape of the developing
block is controlled to avoid intricate block contours.

APPLICATION OF THE GEOSTATISTICALLY BASED

CLASSIFICATIO METHOD
Case Study 1 - Classification of a laterally persistent
deposit
An undeveloped brown coal deposit was analysed according to the
previously mentioned procedures using the limits for various
classes of resource detailed in Table 4. The deposit was sampled on
an irregular grid by 91 drill holes extending over a region 7.5 km

o
o

o
FIG 6(a) - Exploratory drillhole locations for Case Study 1 showing
500 m ranges of influence
The AusIMM Proceedings

l-----J'--_-'

2000M

FIG 6(b) - Exploratory drillhole locations for Case Study 1 showing

250 m ranges of influence.
No21990

11

K D WIDTCHURCH, ADS GILUES AND G D JUST

north-south by 3.5 km east-west. Three major coal plies were

identified. Classification results for the lowest ply only are
discussed.
Figures 6(a) and 6(b) show the locations of exploratory drill
holes; the circles represent arbitrary ranges of influence for each
hole. Using this type of arbitrary assignment it can be seen that
areas of a deposit can be placed in a number of different classes of
resource depending on the range of influence selected. This
highlights the need for an objective method for classifying
resources.

2000'"
CONTOUR

~-ERVA

0 5'"

TABLES

PrelimifUlry classifICation results for the laterally persistent deposit

Ac1w
TX. Ma
1.316 0.25
0.111 1.00
2.371 0.25
2.960 0.25
1.527 O.~
2.~16 0.2.5

Bit
1
7
1
11
12
1~

Vel. Bit
0.329 2
0.111 3
0-'93 ~
,
0.7~0
6
0.7~
0.622 9
13
15
16
17

22
23
2~

31

TocaI

B c1UI
TK Ana
0.67 1.25
0.65 1.25
0.66 1.25
0.99 1.2.5
O.~ 2.2.5
0.69 1.2.5
0.71 1.2.5
0.6' 3.00
0.61 1.25
1.09 1.25
I.U 1.75
1.14 O.SO
1.67 0.2.5
2.69 0.2.5

3.16

----,--

Vel.
0.14
0.11
0.16
l.27
1.0
0.17
0."
1.95
0.77
1.37
3.17
0.57

Bit
10
11
'20
21
2.5
26
27

C elw
TK Ala
0.6.5 2..5
0.59 0-'
0.62 0.75
0.65 0.25
0.75 0-'0
0." 0-'0
0.92 0.25

U ellS.!

Vel. Bit TK
1.61 19 0.~7
0.30 21 0.61
0.~6 29 0.~3
0.16 30 0.'9
0.31 32 0.~4
o.~~ 33 0.57
0.23

NU
1.25
0.75
0.25
2.00
0.25
0.25

Vel.
0.591
0.'13
0.107
1. I 14
0.109
0.1~2

FIG 7 - Contours of raw seam thickness data - laterally persistent

deposit.

0.~2

0.67
15.9

3.6

2.6

thickness over the field of interest.

The deposit was kriged with 500 m by 500 m blocks and the
resulting thickness and estimation error (at a 95 per cent confidence
level) contoured (Figures 9 and 10). At first examination, the
contours of estimation errors indicated that there was litlle
possibility of any significant resources falling in the measured
category.
Classification of the deposit was carried out using the
geostatistically based computer program for square block iterations
with a side dimension of 500 m. The results have been plotted and
are presented in Figure 11 and Table 5. The majority of the regions
assigned to the various classes closely follows the distribution that
would be expected from contours of relative error (Figure 10). As
expected only a very small region has been classified as A class; a
total of 2.5 square kilometres from a deposit covering some 31.5
square kilometres. What is at first glance surprising, is that the A
class region does not correspond to the area of higher drilling

Preliminary statistical analysis of the seam thickness data

suggested a bimodal distribution, and several apparenlly ano~alous
data points were also indicated. Analysis of the seam thickness
contours (Figure 7), indicated that no valid reason existed, based on
current information, for excluding any of the data points and all
further analysis was therefore based on the complete data set.
A large number of attempts were made to calculate an
experimental variogram model that accurately described the ply's
thickness. The final theoretical variogram which has been used
follows an isotropic spherical model with a range of 2300 m, a sill
of 0.29 m2 and no nugget effect (Figure 8). Although the geology
of the deposit indicated possible anisotropy, insufficient data
existed for this step to be undertaken.
Validation of the variogram model by a "jack-knifing" method
indicated that it provides an accurate description of the ply

TABLE 6
Final classification results for the laterally persistent deposit

Bit
1
7
8
11
12
14

A class
TK
Area
1.316
0.811
2.371
2.960
1.521
2.0486

0.15
1.00
0.25
0.15

O.SO
0.15

Vol.
0.329
0.811
0.593
0.7040
0.764
0.622

Bit
2
3
04

S
6
9
13

IS
16
17
22
23
24
31

Total

12

B class
TK
Area
0.67
0.65
0.66
0.99
0.64
0.69
0.71
0.65
0.61
1.09
1.15
1.14
1.67
2.69

3.859

021990

1.25
1.25
1.25
1.25
2.25
1.25
1.25
3.00
1.25

1.25
2.75

O.SO
0.25
0.25

Vol.
0.84
0.81
0.86
1.27
1.043
0.87
0.88
1.95
0.77
1.37
3.17
0.57
0.42
0.67
15.88

Bit
10
18
20
21
15
26
27
19
28
29
30
32
33

Cclass
TK
Area
0.65
2.5
0.59
0..5
0.62
0.75
0.65
0.25
0.7S
0.50
0.88
0.50
0.92
0.25
0.47
1.25
0.68
0.75
0.43
0.25
0.59
2.00
0.44
0.15
0.57
0.25

Vol.
1.61
0.30
0...6
0.16
0.38
0.44
0.23
0.591
0.513
0.107
1.184
0.109
0.142
6.23

The AusIMM Proceedings

COAL RESOURCE CLASSIFlCATION AND GEOSTATISTICS

VARIOGRAM
AZIIoUTH

90

NUGGET

DIP

SILL

o
o.

RANGE

2300

90
350
7000

YIIHOOW

a.ASS SI lE

w.x

01 STAHCE

~AH

O. 81

VARIAHCE

srn

OHI An OH
HO.SAAf'LES

0.2902 0 . 55
0.54
57

1. 8

1\

I
1
I
1
1

DI STAIIC(
o-

350
350 - 700
700 - 1050
1050 - 1400
1400 - 1750
1750 - 2100
21 00 - 2450
2450 - 2800
2800 - 31 50
31 50 - 3500
3500 - 3850
3850 - 4200
4200 - 4550
4550 - 4900
4900 - 5250
5250 - 5600
5600 - 5950
5950 - 6300
6300 - 6650
6650 - 7000

29

UIIM I~

o. 44

;0

po

0.031 25
O. 08033
0.15048
0.14226
0.24586
0.32811
0.32177
O. 44273
o 22594 O. 33
O. 28557
O. 45391
O. 55329
0.43029
O. 30328
0.3603
0.27/14
0.26556
0.24764
O. 37446 o. 22
0.22827

1\

II 0

1
1
1
1
1
1
1

r -- -'4
I

"

\
\
\
\
\
\
\

1
1
1
1

,
,,,

\
\

\
\
\
\

\
'y\2
\
\
\
\
\
\

'lE8

f2

10

,,

,\

/ \

\\

1
1

\
\

\
\
\
\

'14

~8

011

- + - - - -..- - 700

I 400

21 00

---t--2800

-1---

3500

---1---- +
4200
4900

- I -

5600

I -

6300

-----j

7000

FIG 8 - Average seam thickness variogram -laterally persistent deposit.

2000M
CONTOUR INTERVAL = 025M

FIG 10 - Relative kriging error (95 per cent confidence level)

contours based on 500 m x 500 m blocks -laterally persistent
deposit.

FIG 9 - Kriged seam thickness contours based on 500 m x 500 m

blocks - laterally persistent deposit.
The AuslMM Proceodings

No21990

13

K D WHITCHURCH, ADS GILLIES AND G D JUST

FIG 11 - Resource classification results -laterally persistent deposit

FIG 13 - Final classification results for the laterally persistent

deposit.

2000M
_

MEASUREJ

1000M

FIG 12 - Traditionally derived resource classification results.

FIG 14 - Major structural features - structurally disturbed deposit.

density. This may be easily explained by the fact that the

classification is based on relative precision (error/estimate) rather
than error alone. For this reason the class of resource to which a
particular region is assigned, is a function of both the drilling
density and the seam thickness in that region.
A number of regions near the deposit boundary remain
unclassified. This situation occurs due to the constraints placed on
the growth of these regions by previously classified regions and the
deposit boundary. It should be noted, however that the deposit
boundary tends to be a region of high uncertainty. This is reflected

by the fact that, with few exceptions, regions near the boundary are
classified into the least restrictive classes of resource. For this
deposit, and with knowledge of the existence of coal seam
continuity, it is recommended that unclassified areas that fall within
the deposit boundary are considered as falling within the lowest
restrictive class of resource.
Another apparent anomaly is the existence of two isolated C class
regions (blocks 18 and 21, Figure 11), surrounded by B class
regions. The presence of these regions may be attributed to the
restrictions placed on their growth by previously classified regions

14

No 21990

The AusIMM Proceedings

COAL RESOURCE CLASSIFICATION AND GEOSTATISTICS

SUMMARY

STATISTICS FOR UrlTRANSFORMED DATA

tlEAIl
4 (1'~ '" 0 24 '3E + 0 1
I,'AF: I HNCE
0.2990468'?E+'Jl
0.17292971E+Ol
STD. ['E\I.
SI'E:,J~1 E 5:3
0.25723405E+01
I:UF:TOS I 5
\C' .13504897E+02
, HIS T Cr'~ P F1 11 > ..
SOUTHERN AREA
COMPLETE DATA SET

o.

CE: S/
F F.E(~

F:ELA
FRECI

CUtlL
FRE(~

UPPER
CELL L!I'lIT

..
.....

20

+
6

0.018

':'

0,027

4:,
1'" -.

0.136
(1.548

4~

8 , .... "',
0 .042
0 .036

~.

0.018
O. 045
0.181
O. 72'3

f~-:O

14
12
'j

(1

4
4

(1

~1~7

,012

-.

0.012
(1.00E

(1

0.000
8.(11)0

'-

0
1

. (103
(1.00.

(I

.~

'-

0.855
0, :::98
0.934
O. '?61
(1.'0'73
(1.9:::5

.9~1

0.9?1
0.9'H
(, . Sr',' 4
1.0013

o .12'O6E+('l
0.22::2E+(11
0.3178E+01
e. 4134E+('1
0.509(IE+1<11
O. E04t:.E+81
0.7130:E+(11
0.79:1;:E+Ol
1<1.8914E+Ol
o . ';"3 ;:' (, E + (1 1
O. H):33E+(12
(1.117:::[+02
O.1:274~+02

0.1 "3E9E+1<12
IrIF

+>
+

t ..

+~t

~:+-~

+tt

~ tl:~*

+ ..

te

..

... '+

~~*~~.~~~~~

.. +

te

t' t:

+>
+.'
+ ..
+
+
+
+
+
+

:: ~~

(t

332

4
+

FIG 15 - Classical statistical analysis results for seam thickness - Case study 2.

SUMMARY

STATISTICS FOR UNTRAIlSFORMED DATA

1).4(,1951867E+01
'./ fl R I Arl CE
0.30339720E+01
src'. (\EI/.
0.17418381[+81
:.~: E 1...Irl E S5
I) . 2 ~ ~: 18 7 8 6 E + 81
I(URTOS I S
0.13444473E02

m: 'HI

." t ~ I 1ST Cl GF:

CL EFill E0 (, HTA I) E:-:: I,'

FF:EQ

f;:ELFI
Ff;:E(1

cur'lL
FP.EO

ftl,

t' ...

SI) U T H r) F F HU L T
UFo FE P
CELL L I I" I T

::0
+

i'

,Ol'j

(1

':.
1-''':,
, .

') . ~12:::
1) . 133
O. r...' ..c';,
''';'

42
11

0.130
0.034

4":"_.

... .-,

'j

4
4

-.

(t

0
1

-...

0.037
(1 ,0:'::-::

0.01;;
0.046
O. 17'3
0 .:'31
0.851
(1.:::95
0.':'32
0.9(.(1

O.(~lL

O.~72

0.01:
I) . 006
(1 .01<10
1<1 .000
~1 .003
CI.OOE

O.

9~1:"
.: 1

O.9'H
O. '?'H
1
0.'?':'4
1.000
(1 'j'1

o . 1: oS t E + '" 1

o. 2::::2E" ~'1
(. 31 :' ;;: E. +01
~1 413 .. E + (, 1
O. 50'~OE+01
(:.6046E+01
8.7002E+Ol

0,

7'?-:,~:E+01

O. ::;'?14EHJ1
O. ';o::;:'OE+Ol
'" . 1 0:::'3 E +(12
(1.11 r(:E +82
o .1::74E+I)2
13':;'E+02

o.

lr.F

3~4

+ .>
+ t

4
+

.,

+ ......

it' t ... :

+'...

~~.f..~~~~~~~.~+~~

+'t'''1t: ttfr it

t'! t'

+ ..
+T.
+.,

..

+ .,
+

+
+
+
+
+
(:

+
2~t

..
4

FIG 16 - Classical statistical analysis results for seam thickness with data from oxidised regions removed - Case study 2.

The AusIMM Proceedings

No21990

15

K D WHITCHURCH. ADS GILUES AND G D JUST

1I/~

III

40M OJB

L1N~\

60M O/B LINE

---

FAUL T A

, ....u:

MU

40M O/B LINE

100..

FIG 17 (a) - Seam thickness contours for the southern zone.

\
)
I

V "

((0)

---~

'I

('f

FIG 18 (a) - Structural features of the southern region.

--~

')

l
\

\\

/'

LOXLIN[~

/7'

)
/.

J. (

FIG 17(b) - Expanded view of the sample area shown

in Figure 17(a).
FIG 18(b) - Expanded view at the sample area shown in Figure 18(a).
and the fact that they represent isolated zones of high estimation
error. Such locations exist due to isolated thinning of the deposit or
a local decrease in drilling density. Whatever the reason for the
existence of a particular isolated region it is clear that each should
be given close individual attention at the fmal stage in the
classification process. Some subjective judgement is at present
considered unavoidable in the treatment of such regions leading to
the conclusion that further work is needed to address this problem.
For the purpose of this case study these isolated regions were
considered to be of sufficient size to be classed as independent of
the surrounding blocks and to therefore accurately represent the
presence of locations of low relative precision. The two regions
under question, blocks 18 and 21. were consequently left
unchanged from the original computer base classifications. Final
results for this deposit are presented in Table 6.
A comparison of the regions assigned to the previous resource
classes by traditional methods (Figure 12) and by the geostatistical

16

algorithm (Figure 13) reveals little correlation. This may be

explained by the fact that the traditional results were based on drill
hole spacing only. Use of this estimation procedure tends to distort
and increase the resource quantity in categories of greatest
geological assurance in this case. However, this approach is
allowable under the present Queensland and New South Wales
codes although it is no true indication of the level of confidence
associated with the estimate of the resource quantity. In contrast,
the results generated by the geostatistically based algorithm closely
reflect the level of confidence and as such may be considered more
soundly based.
Case Study 2 - Classification of a structurally disturbed
deposit

An anthracitic deposit was chosen as the second case study because

021990

The AusIMM Proceedings

COAL RESOURCE CLASSIFICATION AND GEOSTATISTICS

of the highly structurally-disturbed state of the seams. Complex

faulting in this area gives rise to several fundamental problems not
evident in the analysis of brown coal deposit. Major structural
features of this anthracite deposit are shown in Figure 14.
Preliminary investigations of the sample information indicated
the need to sub-divide the deposit into regions that were accepted as
geologically continuous for the purpose of geostatistical analysis.
Two major regions were identified, the area north of fault A (Figure
14) and the larger area to the south. The southern zone is of greater
interest and will be discussed. Classical statistical analysis of the
raw sample data revealed a standard deviation of almost 50 per cent
of the sample mean of about 4.1 m thickness (Figure 15). The large
value of the standard deviation can be attributed to the presence of a
number of apparently anomalous high values.
In particular it is noted that three samples recorded thicknesses
greater than 13.0 m. Removal of all data lying in oxidised regions
did little to improve the results (Figure 16).
Contouring of raw thickness values, however, revealed the
presence of a large number of isolated high and low values (Figure
17a). This situation is clearly illustrated in Figure 17(b) in which
adjacent sample values range from two m to seven m. Further
drilling showed that these were a reflection of the high degree of
structural disturbances present in the region. Systems of normal
and reverse faults effectively isolate small blocks of the regions into
areas that should be analysed separately (Figures 18(a) and (b)).
These geological characteristics make it extremely difficult to apply
geostatistics, or indeed any estimation method to this region.
Consequently this is considered to be an area where geostatistical
classification is difficult to apply.
Various attempts were made to construct a variogram model for
this region. The experimental variogram exhibited a spherical
model structure with a short range hole effect. The hole effect is a
reflection of the closely spaced seam structural changes and so
distorts any analysis based on a examination of seam thickness.
This emphasises the fact that any geostatistical analysis undertaken
without due emphasis on deposit geology may be misleading and
result in highly erroneous conclusions. In this case since no further
analysis was undertaken, no comparison with traditionally derived
results was possible. It may however, be concluded that in deposits
where geostatistics is not readily applicable for structural reasons
the possibility of A class resources is remote.

CONCLUSIONS
Coal resource classification concepts and definitions have been
studied and a geostatistically based algorithm derived for assigning
mine blocks to classes of varying geological assurance. In use, the
algorithm gives classification results that closely reflect the error
associated with the kriged estimate of resource quantities. The
method provides a consistent basis for comparison of different
deposits. It requires a careful geostatistical analysis with particular
emphasis on details of geological variations. From its use, it is
possible to predict the increase in sampling density required to
attain a higher classification category for a particular area. Results
are readily approachable and require a minimum of subjective
judgement. Resource category restrictions must be arbitrarily
assigned and remain constant for all deposits being compared.

TheAusIMM Proceedings

Finally, it must be emphasised that, as with all classification

approaches, . engineering judgement is needed. In particular,
interpretation may be required with isolated and peripheral blocks.
Further research is warranted on these aspects.

ACKNOWLEDGEMENTS
The support of CSR Ltd during the study is acknowledged. This
paper was prepared within the Department of Mining and
Metallurgical Engineering, University of Queensland. Discussions
with various staff and students assisted in the formulation of the
concepts presented.

REFERENCES
Bureau of Mineral Resources, 1984. BMR refines its resource classification
system, Australian Mineral Industry Quarterly, 36 (3):73-82.
Department of Energy Mines and Resources, 1975. Departmental
tenninology and definitions of resources and reserves, Interim Document,
Ottowa.
Diehl, P. and David, M., 1982a. Classification of ore reserves/resources
based on geostatistical methods, CIM Bull, 75:838, Feb:127-136.
Diehl, P. and David, M., 1982b. Geostatistical concepts for ore reserve
classification, in Froc. 17th Int. APCOM Symp. pp.413-424 (Colorado
School of Mines).
Fettweis, G., 1979. Developments in Economic Geology, Volwne 10: World
Coal Resources, Methods of Assessment and Results, (Elsevier:
Amsterdam)
Froidevaux, R, 1982. Geostatistics and ore reserve classification, CIM Bull,
75:843 pp.77-83
Froidevaux, R, 1983. Precision of estimation of recoverable reserves, in
Proc. NATO Advanced Study Institute on Geostatistics for Natural
Resources Characterisation, Lake Tahoe, pp. 141-164.
Galligan, A.G. and Mengel., D.C., 1986. Code for reporting of identified
coal resources and reserves, Queensland Government Mining Journal,
May, pp.201-203.
GDMB, 1959. Eine Klassifikation der lagerstattenvorrate, Empfohlen vom
Lager Stattenaushub, Erzmetall, 12:55-57.
McKelvy, V.E. Mineral resource estimates and public policy, American
Scientist, 60, (January-February): 32-40.
Sabourin, RL., 1983a. Application of a geostatistical method to
quantitatively define various categories of resources, in Proc NATO
Advanced Study Institute on Geostatistics for Natural Resource
Characterisation, Lake Tahoe, pp201-205.
Sabourin, RL., 1983b. Geostatistics as a tool to define various categories of
resources, Mathematical Geology, 15:131-143.
Ugarte, 1., 1972. Ejemplos de modelos de estimacion a corto y largo plazo,
Boletin de Geostatistica, 4:3-22.
United States Bureau of Mines and US Geological Survey 1976. Principles
of the mineral resource classification system, USBM/USGS, Geological
Survey Bulletin 1450-A.
United States Bureau of Mines and US Geological Survey, 1980. Principles
of a resource/reserve classification system: Geol. Survey Circ.831.
Wellmer, EW., 1983. Classification of ore reserves by geostatistical
methods, Erzmetall., 36, N. 7IB; 315-321.
Wigglesworth, K.E, 1981. Code for assessment and reporting of coal
resources and reserves of South Australian coal deposits, Dept. of Mines
and Energy, South Australia, Report. BK. No 81/86.

No21990

17