Practical Geostatistics For Resource Estimation: Last Update: July 2014

Practical Geostatistics for
Resource Estimation
Overview
Georges
g Verly
y
Last Update: July 2014

© 2014
1
Resource Modeling is
Creating a resource block model …
21400 21600 21800 22000 22200 22400

2
200
200
0
4 g/t
100
100
Au
2.0
0.5
0
0
-100
-100
Wst
-200
200
0.0
-2
PARADISE INC.
DDH
-300
SECTION: 9200 N
-300
Ultimate Pit
CO & CM
MEAS/IND RESOURCES
-400
-400
lc_013a
21400 21600 21800 22000 22200 22400
2
Mar-2010
[email protected] September 2014

Resource Modeling is
When information consists of drill hole samples
that have been logged and assayed.
21400 21600 21800 22000 22200 22400
2
200
200
0
4 g/t
100
100
Au
2.0
0.5
0
0
-100
-100
Wst
-200
200
0.0
-2
0
PARADISE INC.
DDH
-300
-300
SECTION: 9200 N Topo

DDH SAMPLES
-400
-400
lc_011
21400 21600 21800 22000 22200 22400
3
Mar-2010
Content
Resource Modelling Steps
• Data Collection
• Geological Modelling
• Block Grade Model Estimation
– EDA: Exploratory Data Analysis
– Variogram: Spatial Variability
– Block grade Distribution
– Estimation / Simulation
– M d lV
Model Validation
lid ti
– Classification
Other
• Course / Notes Format
• Appendix 1: Exercise solutions 4
May-2014

Content
• Data Collection
– Drilling, logging
– Sampling, assaying, QA/QC
– Database
– EDA: Exploratory
p y Data Analysis
y
– Model Validation
– Classification
5
May-2013
Data Collection
• Data collection is most important
because:
– It is the first step in resource
modeling
– There is a significant risk
associated to data collection
– Poor data collection practice may
result in huge loss of value
• We have to collect the right data
and the data must be of good
quality!
Sampling and QA/QC

not part of this course
6
Oct-2013

Sampling – QA/QC
f163
Accurate Accurate Not Accurate

Precise Not Precise Precise
• Sample values must be accurate and precise

• Poor sampling introduces:
– Random
R d noise
i
• Reflected in variogram nugget effect
• Can be compensated to some degree
– Bias
• Cannot be compensated
• Difficult to identify if data are noisy
7
Jul-2014
Database
• The database must be clean

 Good database management
• Good database management:

– Process for capturing data in place
– Process for prioritizing entries when multiple measurements are made
on same sample
– Avoid manual or repetitive entries
– Verification steps implemented
– Security measures in place
– Documentation (e.g. purpose of data collection)
– Etc.
Data capture and
management not part of
this course.
8
May-2012

Content
• Data Collection
– Database
p y Data Analysis
y
– Classification
9
May-2013
Geological Model
Geological Model 21400 21600 21800 22000 22200 22400
200
200
4 g/t
• The drill hole logs are interpreted

100
100
Au
2.0
and a geological model is 0.5

0
created
created.
-1
-100
100
Wst
-200
-200
• Most important step!

0.0
PARADISE INC.
DDH
-300
-300

– Mineralization controls must be DDH SAMPLES
understood:
-400
-400
lc_011
21400 21600 21800 22000 22200 22400
• Controls must be at DH scale

Samples logged and assayed
• Guides at district scales are of
little use
21400 21600 21800 22000 22200 22400 Overburden
200
200
Ox. Saprolite
Sul. Saprolite
100
100
Saprock
S k
1
Crb Lch Brk
Geological model most important

0
0
Crb Stbl Brk

Dyke
-100
-100
Not part of this course 4 g/t
2.0 Au
-200
-200
PARADISE INC. 0.5

Wst
-300
-300
SECTION: 9200 N
GEOLOGY + DDH SAMPLES 0.0
DDH
-400
-400
lc_012
21400 21600 21800 22000 22200 22400
Geological Model
10
Sep-2012

Content
• Data Collection
– Database
p y Data Analysis
y
– Classification
11
May-2013
EDA – Purpose, Tools

Purpose Tools
• Data familiarization • A few simple:
• Detecting possible errors – Graphics
– Statistics
• Identifying/confirming different
mineralizations
• Lots of results
• Answering questions such as: – Series of results per
– What topcut values ? geology domain
– Ordinary or indicator kriging ?
– Unsure at start which result
– Mean and variance ? might
i ht b
be useful
f l
• Providing information for
model validation
12
Mar-2010

EDA – Graphs
Freq Histogram of Au grades Log-Probability Plot
100
(%)
30
P(2. < Au < 4.) 10
20
10 f. 3a 1.0
0
0 4 8 12 16 Au (g/t) f.115
0.1
99.99 99.8 99 98 95 90 80 70 6050 40 30 20 10 5 2 1 0.5 0.1 0.01
Scattergram
f. 89
x
13
Mar-2010
EDA – Statistics
Freq
• Location (%)
30
– Mean or average Dispersion
20
– Median (middle)
10 f. 3f
0
0 4 8 12 16 Au (g/t)
• Dispersion (Spread) Location
– Variance
– Standard deviation = Variance
14
Sep-2012

EDA – Statistics
Coefficient of Correlation
Smoothed Regression
S
LC - TRENCHES
100. ORIGINAL AU VERSUS REJECT
NB. OF DATA 477
X VAR: MEAN 5.814
Good STD. DEV. 6.110
Y VAR: MEAN 6.021

Au STD. DEV. 7.264
REJECT AU
10. CORRELATION 0.935
S
Poor
f107
1.
Au 1. 10. 100.
33b
ORIGINAL AU
15
Mar-2010
EDA – Some Issues
• Sampling quality
– Look at QA/QC results
• Geological domains
– Different mineralization characteristics?
– Checking character of boundaries (firm?, soft?)
• Compositing
• Declustering
• Top
Top-cutting
cutting
16
Apr-2013

EDA – Quiz 1
2
8 6 2
1
1
f.184
6 Samples values (g/t Au)

Question: what is the average grade?
17
Mar-2010
EDA – Quiz 2
2
1
1 8 2
1
f.184a

18
Mar-2010

EDA – Declustering
• Potential problem:
Clusters of samples are common – Clusters often located within
high-grade zones.
N
– Impact could be serious:
overestimation of the average
grade and variability
• Solution  Declustering:
– Reduce the “weight” of each
f.116a
Clusters clustered data ±

proportionally to the cluster
sampling density
– Calculate weighted
(declustered) statistics.
• Declustering impact
– Often a lower average grade.
19
Apr-2014
EDA – Top-cutting
• Potential problem:
Outliers are “common” – Overestimation of mean grade
0.3 y
and of variability.
• Solution (top-cutting):
– Restrict high-grade zone of
0.2 No. Data 99688
influence during estimation; or
mean 2.17
std. dev 4.95 – Top-cut high assay values
0.1
Max. 330.0 ? • Combination of tools
Min. 0.100
– Histogram, Probability plot
– Indicator correlation plot
f62
f62a – Metal at risk
Metal-at-risk
0.0
? – Decile analysis
0.0 4.0 8.0 12. 16. 20.
BH Au (g/t) • Impact
– high-grade smearing reduced
– Lower average grade
Important to record how much
metal removed by top-cutting 20
Oct-2010

Content
• Data Collection
– Database
p y Data Analysis
y
– Classification
21
May-2013
Quiz – 3
• Suppose a vertical section through a mineralized zone as following:
9m
3m
Saprolite A B C
Zone 3m
D f05_01
• Questions
1) Which pair of samples probably shows the smallest grade difference?
• “A-B”, “A-C”, or “A-D” ?
2) Same question for …
Note: Grades such as Fe, MgO, SiO2, Ni

22
Sep-2012

Variogram
The variogram is a function that quantifies the notion of
geological variability or spatial variability.
• Why is it useful?
– To improve accuracy of block grade estimates
– To analyze block model selectivity (“in situ  (dist)
dilution”)
– To assess risk
– For simulation
Sill
• In fact, the variogram is essential 0
f.78
0 Distance
Range
No variogram  No geostat
No guts  No glory
(H. Parker)
23
May -2013
24

Contents
• Data Collection
• Block vs. Sample grade distribution
• Dilution Misclassification
Dilution,
– Classification
25
May-2013
Block Grade Distribution

• Average grade
– 1.57 g/t
Small Blocks
• Spread
– Min/Max = 0.1/5.7 g/t
1.2 0.2 0.9 1.2 – Variance = 1.96
3.5 2.4 0.8 2.0
• Cut-off grade = 1.5 g/t
2.3 0.2 0.1 2.1
– 6 blocks (2x4m)  1,5 g/t Au
2m 08
0.8 07
0.7 10
1.0 57
5.7
– Proportion off ore
≥ 1.5 g/t f06_03a = 6/16 = 37%
4m
– Grade
= 3.00 g/t

WAS
• Average grade
– 1.57 g/t 1.57 g/t
Large Blocks
• Spread
0.1 / 5.7
– Min/Max = 1.00/2.23
1.2 0.2 0.9 1.2
– Variance = 0.24 1.96
1.83 1.23
3.5 2.4 0.8 2.0
2.3 0.2 0.1 2.1

• Cut-off grade = 1.5 g/t
4m 1.00 2.23 – 2 blocks (4x8m)  1,5 g/t Au
0.8 0.7 1.0 5.7
– Proportion off ore
≥ 1.5 g/t f06_03b
= 50% 37%
8m
– Grade
3.00 g/t
= 2.03 g/t

Grade/Tonnage Curves
1 6
5
0.8
4
0.6
Tonnage
Grade
3
0.4
2
0.2
1
Tonnes (SB) Tonnes (LB)
Tonnes (SB)
Tonnes (SB) Grade (SB)
Grade (SB)
Grade (SB) Grade (LB)
0 0
0 0.5 1 1.5 2 2.5
Cut‐off Grade
SB = Small Blocks
LB = Large Blocks
28
Feb-2014

Contents
• Data Collection
• Conceptt
C Only brief overview of
• Classical Estimation Methods
• Error of Estimation
simulation in this
• Kriging course
– Classification
29
May-2013
Estimation - Purpose
• Estimate grade at a given location using surrounding
information
• Choice of method depends on deposit shape, mineralization
variability, data availability, etc.
• Usually a weighted average of surrounding sample grades
Au1 Au2
Samples:
Au1 = 2.3 g/t
• Estimation methods:
Au3 Au2 = 6.7 g/t – Polygonal
Etc. – Inverse distance
Block – Sectional
f.183a
Estimated AuBlock – Kriging

= w1Au1 + w2Au2 + w3Au3 – Etc.
Surrounding grades should have similar grade values

to that expected in block
 Concept of “stationarity” in the selection neighbourhood
30
Apr-2013

Estimation Error
• Every time there is an
estimate, there is an error
Error = True Value –

Estimated Value  Er
0 Error
• The spread of the error (Er) can be

calculated if the variogram is known.
Jun-2013
Estimation Error
Cu2= 0.2 Best Continuity
5m f.141a
? Cu1= 0.5
5m
Cu0*  w1  Cu1   w 2  Cu2 
Different estimation methods

 Different spreads of errors
W t
Worst A bit Even
better better
f.152c
 0   0   0 
NN2 NN1 ID
Cu0* = 0 x Cu1 + 1 x Cu2 Cu0* = 1 x Cu1 + 0 x Cu2 Cu0* = 0.5 x Cu1 + 0.5 x Cu2
Jun-2013

Estimation Error
The variogram gives access to
the spread of the estimation error
1.20 STDV(Error)
1.15 "NN 2"
1.10
1.05
1.00 Best estimate is that one that

0.95
"NN 1" minimizes spread of the error
"Best"
0.90
"I t iti ”
"Intuition”
ID  Kriging.
g g
0.85
f.152b
0.80
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 W2
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 W1
Estimation Weights (w1 and w2)
Spread: Standard deviation in this case 33

May-2009
Estimation: Kriging
• Kriging picks the set of weights such that the spread of the error is
minimized.
• Kriging is:
– Best  Minimize the spread (variance) of the error
– Linear  Weighted average of surrounding sample grades
– Unbiased  Mean of the error is 0
– Estimator
 BLUE
34
Feb-2012

Estimation: Kriging
Most common krigings

• Ordinary
O di k i i (OK)
kriging
– Mineralization “well” behaved
– Top-cutting data is enough to control high-grades
• Indicator kriging (IK)

– Mineralization very variable
– Continuity directions vary with cut-off
• Co-Kriging
– More than one variable
35
Mar-2010
Estimation: Kriging
21400 21600 21800 22000 22200 22400

200
200
4 g/t
100
100
Au
2.0
0.5
0
0
-100
-100
Wst
-200
-200
0.0
PARADISE INC.
DDH
-300
SECTION: 9200 N
300
Ultimate Pit
-3
CO & CM
MEAS/IND RESOURCES
-400
-400
lc_013a
21400 21600 21800 22000 22200 22400
Output from kriging is a grade block model that needs to be validated.
36
Mar-2010

Contents
• Data Collection
– Classification
37
May-2013
Validation
Why
• Estimated block model must be reasonable
Tools
• Visual checks
• Statistical checks
– Block model VS. sample mean grades
– Block model variability  Selectivity
– Local accuracy (kriging efficiency)
– Grade profiles across geological contacts
• Cross-validation
• Reconciliation
– With previous model
– With production
38
Mar-2010

Validation – Visual Validation
Kriged Grade Map + Sample Values
22400 22500
11400
11400
Au (g/t)
Look at the trends,
0. 1 2 3 4 5.
geological contacts,
xval_map_ok
22400 22500
estimated grades vs.
sample values
39
Feb-2012
40

Contents
• Data Collection
– Classification
41
May-2013
Regulation / Classification
• Resources must have reasonable prospects for

economic extraction.
– Some
S assessmentt off economics,
i selectivity,
l ti it cutoff-grade
t ff d
• Resources must be categorized into Inferred,

Indicated, and Measured categories.
– Depends on confidence in geological and grade
continuity.
– Depends also on quality of information.
• Resources must be signed off by a Qualified/

Competent Person.
42
Jun-2013

Summary (1/4)
Resource modeling is creating a resource block model …
21400 21600 21800 22000 22200 22400
2
200
200
0
4 g/t
100
100
Au
2.0
0.5
0
0
-100
-100
Wst
-200
200
0.0
-2
0
PARADISE INC.
DDH
-300
SECTION: 9200 N
-300
Ultimate Pit
CO & CM
MEAS/IND RESOURCES
-400
-400
lc_013a
21400 21600 21800 22000 22200 22400
43
Sep-2012
Summary (2/4)
When information consists of drill hole samples
that have been logged and assayed.
21400 21600 21800 22000 22200 22400

2
200
200
0
4 g/t
100
100
Au
2.0
0.5
0
0
-100
-100
Wst
-200
200
0.0
-2
PARADISE INC.
DDH
-300
-300

DDH SAMPLES
-400
-400
lc_011
21400 21600 21800 22000 22200 22400
44
Sep-2012

Summary (3/4)
Database
• Resource estimation steps

Sampling
1. Data collection
 Clean and complete database
 Good sampling and QA/QC Geology
2. Good geological model

• Mineralization controls understood
3. EDA: Exploratory Data Analysis

• Data checking EDA
• Data statistics Freq
(%)
– Including analysis of geological boundaries 30
• Information gathering used for the: 20

Variance
– Estimation (topcuts, ...) 10 f. 3d
– Model validation (mean, variance, ...)

0
0 4 8 12 16 Au (g/t)
Mean
45
Feb-2014
Summary (4/4)

Variogram
4. Variogram: Spatial variability analysis
• Quantify grade spatial correlation GT Curves
5. Block grade Distribution

• Grade distribution depends on block size Selectivity
• Selectivity, dilution, recovery issues
6. Estimation
• Error of Estimation BLUE
• Kriging Validation
7. Model Validation 8.2

3.7 0.9
• Estimated model must be thoroughly checked
2.1
?
8. Classification 3.4
2.8
46
Jun-2013

Contents
1. Database
2. Sampling
p g
3. Geological Model
4. EDA: Exploratory Data Analysis
5. Variogram: Spatial Variability
6. Block grade Distribution
7. Estimation
8. Model Validation
9. Classification
• Course / Notes Format
• Appendix 1: Exercise solutions
47
May-2012
The Course Check

Will follow ± the resource estimation steps as inSlide
this
introduction.
Content
5. Block Grade Distribution
Content • Sample vs. block grade
distribution
• Selective Mining Unit
1. Univariate statistics (SMU)
• Theory 6. Estimation
• EDA practice – Estimation methods
– Error of Estimation
2. Bivariate Statistics
7. Kriging
• Theoryy
– Ordinary Kriging
• EDA practice
– Other krigings
3. Sampling, QA/QC 8. Model Validation
4. Spatial Variability 9. Classification
• Variograms 10.Simulation
48
Apr-2013

The Geostat Learning Curve
Monday Tuesday Wednesday Thursday Friday
Kriging Model Simulation

EDA Variogram Validation
Sampling Error of Block Grade Classification
QA/QC Estimation Distribution
49
May-2013
50

The Notes – Table of Content
51
Oct-2010
The Notes – Usual Format

Will follow ± the following format
Example
E l off sections:
ti
Format – Variography
– Estimation
Usually four parts per section

1. Preliminaries 3. Appendices
• Content • Demonstrations
• Brief overview of material • Exercise solutions
covered
• What you need to know 4 Review
4.
2. The section itself
• With eventually some
recalls
• Some exercises
52
Oct-2010

The Notes – Special Section
One special section to show:
• Link between course and estimation
process
• Progress made during the course
• Different kinds of variances
• Formulas
Appendix 3 – Formulas
Appendix 2 – Variances (1/6)
Univariate & Bivariate Statistics
Calculating un-weighted (naïve) statistics

Variance Freq
(%)
Mean
• The variance is a measure of dispersion of Covariance
Dispersion 1
values around the average m X   xi
1N
X
Y
xi xi
m
yi
mY
C
O
V
,
,
• For the same element (e
(e.g.
g Cu)
Cu), there are
N
           
X
f. 3f
several variances
1N
Variance
sX
yi
m
mY
Values  
1 1        
X
  xi  mX    N  xi2   mX2
2
sX2   
N
Variance of sample grade values Note:

• Sample variance: /(N-1) Correlation Coefficient
X
Y
C
Os
V X
, sY
• Section “Univariate Statistics”
X
Y
rX
C
O
R
,
• Population variance: /N ︵︶
  
Y
VAR(Z ) Other notations: sZ2 ,  Z2 
Standard Deviation Coefficient of Variation
• The sample grade variance is sX  Variance s
CVX  X
obtained at the stage of EDA statistics mX
21
Nov 2012
13
Oct-2009
53
Apr-2013
54

Appendix 1
Quizzes and Solutions
55
Mar-2010
Appendix 1 – Quiz 1
2
8 6 2
1
1
f.184

56
Mar-2010

Appendix 1 – Quiz 1 Solution
• Naïve average 2
6
8 2
1 2 + 8 + 6 + 2 + 1 1
 3.3 1
6
f.184
• Declustered average 6 Samples values (g/t Au)

8+6
1 2 + +2+1
2  2.6
26
5
• About -20% change from naive to declustered!
57
Mar-2010
Appendix 1 – Quiz 2
2
1
1 8 2
1
f.184a

58
Mar-2010

• Average (all data) 2

1
8 2
1 2 + 8 + 1 + 2 + 1 1
 2.5 1
6
f.184a
• Average (outlier deleted) 6 Samples values (g/t Au)

1 2 + 1 + 2 + 1
 1.4
5
• About -40% change when highest value is deleted!
59
Mar-2010
Quiz – 3
• Suppose a vertical section through a mineralized zone as following:
9m
3m
Saprolite A B C
Zone 3m
D f05_01
• Questions
1) Which pair of samples probably shows the smallest grade difference?
• “A-B”, “A-C”, or “A-D” ?
2) Same question for …
Note: Grades such as Fe, MgO, SiO2, Ni

60
Sep-2012

Situation
9m A B C
3m
Saprolite A B C
Zone 3m
D f05_01
A D
Likely grade
profiles
Q1) Which pair of samples probably shows the smallest grade difference?
– “A B” “A-C”,
“A-B”, “A C” or “A-D”
“A D” ?
A1) “A-B” because separating distance is:
– Smaller than “A-C”
– Same as “A-D”, but “A-B” is the direction of better continuity
61
Oct-2013
Variogram1––Solution
Appendix Quiz 3 1Solution
(3/3)
Situation
9m
3m
Saprolite A B C
Zone 3m
D f05_01
Q2) Which pair is next?

– “A-C”
“A C” or “A
“A-D”
D” ?
A2) Probably “A-C” if grade continuity is much better along “A-C” direction
than along “A-D”
– Otherwise it could be “A-D”
– Depends on mineralization anisotropy.
62
May-2014

Practical Geostatistics For Resource Estimation: Last Update: July 2014

Uploaded by

Copyright:

Available Formats

Practical Geostatistics For Resource Estimation: Last Update: July 2014

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Practical Geostatistics For Resource Estimation: Last Update: July 2014

Uploaded by

Copyright:

Available Formats

Practical Geostatistics for

Last Update: July 2014

21400 21600 21800 22000 22200 22400

21400 21600 21800 22000 22200 22400

[email protected] September 2014

21400 21600 21800 22000 22200 22400

SECTION: 9200 N Topo

21400 21600 21800 22000 22200 22400

[email protected] September 2014

Sampling and QA/QC

[email protected] September 2014

Accurate Accurate Not Accurate

• Sample values must be accurate and precise

• The database must be clean

• Good database management:

[email protected] September 2014

• The drill hole logs are interpreted

and a geological model is 0.5

• Most important step!

SECTION: 9200 N Topo

21400 21600 21800 22000 22200 22400

• Controls must be at DH scale

Crb Lch Brk

Geological model most important

Crb Stbl Brk

Not part of this course 4 g/t

PARADISE INC. 0.5

21400 21600 21800 22000 22200 22400

[email protected] September 2014

EDA – Purpose, Tools

[email protected] September 2014

[email protected] September 2014

Y VAR: MEAN 6.021

EDA – Some Issues

[email protected] September 2014

6 Samples values (g/t Au)

6 Samples values (g/t Au)

[email protected] September 2014

Clusters clustered data ±

[email protected] September 2014

• Suppose a vertical section through a mineralized zone as following:

Note: Grades such as Fe, MgO, SiO2, Ni

[email protected] September 2014

[email protected] September 2014

Block Grade Distribution

[email protected] September 2014

2.3 0.2 0.1 2.1

Block Grade Distribution

[email protected] September 2014

Estimated AuBlock – Kriging

Surrounding grades should have similar grade values

[email protected] September 2014

Error = True Value –

• The spread of the error (Er) can be

Cu0*  w1  Cu1   w 2  Cu2 

Different estimation methods

[email protected] September 2014

1.00 Best estimate is that one that

Spread: Standard deviation in this case 33