APPLIED MINING GEOSTATISTICS

APPLIED MINING GEOSTATISTICS

MODULE I
Workshop
Newmont Gold Corporation
Boca Ratn, Florida, 24-28 September 2007
2 2
INTRODUCTION
INTRODUCTION

Goals of this workshop:
Goals of this workshop:

Discuss and understand when and where

Discuss and understand when and where
geostatistics
geostatistics
can add value to your product.
can add value to your product.

Understand data issues when applying

Understand data issues when applying
geostatistics
geostatistics
.
.

Introduce or deepen (as the case may be) the

Introduce or deepen (as the case may be) the
ability to use GSLib or other programs to
ability to use GSLib or other programs to
develop resource models.
develop resource models.

Step through major aspects of resource

Step through major aspects of resource
modeling.
modeling.

Understand the limitations of the techniques

Understand the limitations of the techniques
discussed and the resulting models.
discussed and the resulting models.
3 3
INTRODUCTION
INTRODUCTION
(cont)
(cont)

Understand the objectives of a Resource
Understand the objectives of a Resource
Model.
Model.

Different types of resource models for
Different types of resource models for
different objectives. Long
different objectives. Long
-
-
, medium
, medium
-
-
, and
, and
short
short
-
-
term models.
term models.

Conversion of Resources into Reserves.
Conversion of Resources into Reserves.

Conditional Simulation (CS) studies, the
Conditional Simulation (CS) studies, the
concept.
concept.

Discussion of basic geostatistical concepts.
Discussion of basic geostatistical concepts.

Emphasis on practical aspects and
Emphasis on practical aspects and
applications.
applications.
4 4
INTRODUCTION
INTRODUCTION (cont) (cont)

Sampling and data quality issues.
Sampling and data quality issues.

Dilution.
Dilution.

Variability vs. smoothing.
Variability vs. smoothing.

Uncertainty and Risk.
Uncertainty and Risk.

How do all this fit in resource and reserve
How do all this fit in resource and reserve
modeling work, and in mine development
modeling work, and in mine development
and mine planning tasks?
and mine planning tasks?

Some geostatistical knowledge assumed!
Some geostatistical knowledge assumed!
5 5
INTRODUCTION
INTRODUCTION (cont) (cont)

Brief Table of Contents:
Brief Table of Contents:

Random Variables and Random Functions.

Random Variables and Random Functions.

Statistical Analysis (EDA), including clustering,

Statistical Analysis (EDA), including clustering,
dealing with outlier data, and geologic and
dealing with outlier data, and geologic and
estimation domain definition.
estimation domain definition.

Variography.
Variography.

Recoverable Resources
Recoverable Resources

Various forms of Kriging.

Various forms of Kriging.

And minor discussions, time permitting, on

And minor discussions, time permitting, on
model validation and reconciliation, conversion
model validation and reconciliation, conversion
of resource into reserves, and geostatistical
of resource into reserves, and geostatistical
Conditional Simulations.
Conditional Simulations.
6 6
A LITTLE HISTORY
A LITTLE HISTORY

Formal modern Probability Theory dates
Formal modern Probability Theory dates
back to the 1600
back to the 1600

s (
s (
Blaise
Blaise
Pascal and
Pascal and
Fermat).
Fermat).

The foundation for
The foundation for
geostatistics
geostatistics
was laid
was laid
out in the early to mid 1900
out in the early to mid 1900

s by
s by
Kolmogorov
Kolmogorov
,
,
Matern
Matern
, Weiner, and
, Weiner, and
Gandin
Gandin
.
.

Geostatistics started in the late 1950
Geostatistics started in the late 1950

s and
s and
1960
1960

s with Krige and

s with Krige and
Sichel
Sichel
in South
in South
Africa; and formalized by Matheron in
Africa; and formalized by Matheron in
France
France

Theory of Regionalized Variables

Theory of Regionalized Variables

.
.
7 7
A LITTLE HISTORY
A LITTLE HISTORY (cont) (cont)

Two of Matheron students (Journel and
Two of Matheron students (Journel and
David) moved to North America and
David) moved to North America and
established geostatistical centers.
established geostatistical centers.

Popular first in mining and meteorology,
Popular first in mining and meteorology,
geostatistics
geostatistics
in now used in many fields:
in now used in many fields:
forestry, petroleum industry,
forestry, petroleum industry,
environmental assessment and
environmental assessment and
remediation, fisheries, image processing,
remediation, fisheries, image processing,
alternative energy sources, etc.
alternative energy sources, etc.

Current main centers for
Current main centers for
geostatistics
geostatistics
are
are
Fountainbleau
Fountainbleau
, Stanford University,
, Stanford University,
University of Alberta, and others.
University of Alberta, and others.
8 8
Concept and slide courtesy of DAT
Mine Consulting.
MODELING AND PLANNING WORK IS AN
MODELING AND PLANNING WORK IS AN
ITERATIVE PROCESS
ITERATIVE PROCESS
SCHEDULING
ANALYSIS
PIT LIMIT
MODEL
RESOURCE
AND COSTING
EQUIPMENT SIZING
THROUGHPUT
MINE SIZE/
ANALYSIS
CUT-OFF GRADE
SEQUENCING
PHASING AND
9 9
MODELING AND PLANNING WORK IS
MODELING AND PLANNING WORK IS
AN ITERATIVE PROCESS
AN ITERATIVE PROCESS
Long Term Plan Short Term Plan
Operations
Production Drilling Haulage and Transport
Finances
Marketing
Geotech Geology
Surveying
Laboratory. Metallurgy
Concept and slide courtesy of DAT
Mine Consulting.
10 10
CLASSIC STATISTICS OR
CLASSIC STATISTICS OR
GEOSTATISTICS?
GEOSTATISTICS?

Most of classical statistics is based on the
Most of classical statistics is based on the
Central Limit Theorem (Gaussian theory).
Central Limit Theorem (Gaussian theory).
Limited applications to spatially correlated
Limited applications to spatially correlated
variables.
variables.

Geostatistics also has been developed in
Geostatistics also has been developed in
large measure based on Gaussian theory,
large measure based on Gaussian theory,
but it has been adapted to deal with
but it has been adapted to deal with
spatially correlated variables.
spatially correlated variables.

Geostatistics was born out of practical
Geostatistics was born out of practical
needs, and is likely to remain an applied
needs, and is likely to remain an applied
science, a specialized branch of classical
science, a specialized branch of classical
statistics.
statistics.
13 13
ADVANTAGES OF USING
ADVANTAGES OF USING
GEOSTATISTICS
GEOSTATISTICS

Uses the spatial continuity of the data,
Uses the spatial continuity of the data,
which is a reflection of the physics of the
which is a reflection of the physics of the
underlying processes.
underlying processes.

The modeling of the spatial continuity
The modeling of the spatial continuity
accounts for quantitative data (assays,
accounts for quantitative data (assays,
core samples, fuzzy seismic), and also for
core samples, fuzzy seismic), and also for
the qualitative information
the qualitative information
(geologist/expert information).
(geologist/expert information).

The physical/geological interpretation and
The physical/geological interpretation and
modeling of the variables allows to go
modeling of the variables allows to go
beyond actual data.
beyond actual data.
14 14
ADVANTAGES OF USING
ADVANTAGES OF USING
GEOSTATISTICS
GEOSTATISTICS
(Cont)
(Cont)

The physical/geological interpretation and
The physical/geological interpretation and
modeling of the variables allows to go
modeling of the variables allows to go
beyond actual data.
beyond actual data.

Provides a framework for
Provides a framework for

transporting
transporting

geologic/physical interpretations further

geologic/physical interpretations further
downstream.
downstream.

Mathematical models (and the
Mathematical models (and the
incorporated geology) are to be used in
incorporated geology) are to be used in
the process design. There should be
the process design. There should be
consistency between the two.
consistency between the two.

Provides the tool for assessing modeling
Provides the tool for assessing modeling
uncertainty.
uncertainty.
15 15
DISADVANTAGES OF
DISADVANTAGES OF
USING GEOSTATISTICS
USING GEOSTATISTICS

Does not replace or create actual data!
Does not replace or create actual data!

Does not interpret or model by itself. Only
Does not interpret or model by itself. Only
provides tools for interpretation and
provides tools for interpretation and
modeling.
modeling.

Does not replace expert decisions:
Does not replace expert decisions:
modeling cannot be done by a computer.
modeling cannot be done by a computer.

Forces the geologists,
Forces the geologists,
geostatisticians
geostatisticians
, and
, and
mining engineers to work harder! Most
mining engineers to work harder! Most
importantly, forces the experts to make
importantly, forces the experts to make
decisions consistent with the data
decisions consistent with the data

we
we
are data
are data
-
-
driven!
driven!
16 16
SOME BASIC CONCEPTS
SOME BASIC CONCEPTS
1. 1.
Modeling Scale:
Modeling Scale:
mining does not occur at core
mining does not occur at core
(drill hole data) scale, but at a much larger
(drill hole data) scale, but at a much larger
scale. Keep in mind also the time factor in the
scale. Keep in mind also the time factor in the
definition of the scale of interest.
definition of the scale of interest.
2. 2.
Numerical Modeling: numbers replace
Numerical Modeling: numbers replace
subjective concepts to some extent. There is
subjective concepts to some extent. There is
only
only
one
one
true value that we will never know; so
true value that we will never know; so
our probabilistic models are our best guesses.
our probabilistic models are our best guesses.
3. 3.
Stationarity, how it
Stationarity, how it
relats
relats
to domaining.
to domaining.
4. 4.
Uncertainty: it is not a property of the
Uncertainty: it is not a property of the
mineralization. It stems from our incomplete
mineralization. It stems from our incomplete
knowledge (and sometimes incorrect use of
knowledge (and sometimes incorrect use of
numerical models.
numerical models.
17 17
SOME BASIC CONCEPTS
SOME BASIC CONCEPTS
(Cont
(Cont
)
)
5. 5.
Smoothing: most of the methods we use
Smoothing: most of the methods we use
sacrifice representing the true variability of the
sacrifice representing the true variability of the
information for a better guess of the average
information for a better guess of the average
value. Not suited for modeling extreme of data
value. Not suited for modeling extreme of data
distributions.
distributions.
6. 6.
Dilution, its relation to smoothing, and how to
Dilution, its relation to smoothing, and how to
control/model it.
control/model it.
7. 7.
Where does conditional bias fit into all of this?
Where does conditional bias fit into all of this?
8. 8.
Impact of the mistakes we make: differences
Impact of the mistakes we make: differences
between uncertainty and risk. The concept of
between uncertainty and risk. The concept of
Loss and its use in modeling.
Loss and its use in modeling.
18 18
BREAK BEFORE WE GO INTO
BREAK BEFORE WE GO INTO
DETAILS?
DETAILS?
19 19
REGIONALIZED VARIABLES
REGIONALIZED VARIABLES
AND RANDOM FUNCTIONS
AND RANDOM FUNCTIONS

Consider an attribute (mineral

Consider an attribute (mineral
grades, seam thicknesses, pH, or
grades, seam thicknesses, pH, or
just about anything!) varying in the
just about anything!) varying in the
space
space
A:
A:
z(
z(
x
x
),
),
x
x

A.
A.

The value
The value
z(
z(
x
x
) is interpreted as a
) is interpreted as a
realization of
realization of
Z(
Z(
x
x
).
).

z(
z(
x
x
) is a regionalized variable, while
) is a regionalized variable, while
Z(
Z(
x
x
) is the random variable.
) is the random variable.
20 20
REGIONALIZED VARIABLES
REGIONALIZED VARIABLES
AND RANDOM FUNCTIONS
AND RANDOM FUNCTIONS (Cont) (Cont)

The approach is then to randomize the variables
The approach is then to randomize the variables
{
{
z(
z(
x
x
),
),
x
x

A},
A},
interpreting them as realizations of
interpreting them as realizations of
the set
the set
{
{
Z(
Z(
x
x
),
),
x
x

A}.
A}.

The random variables
The random variables
{
{
Z(
Z(
x
x
),
),
x
x

A}
A}
have all
have all
the
the
same probability distribution function:
same probability distribution function:
F
F
(
(
x
x
1
,...,
,...,
x
x
N
;
;
z
z
1
,...,
,...,z
N
) =
) =
Prob{
Prob{
Z
Z
(
(
x
x
1
)
)

z
z
1
,...,
,...,
Z
Z
(
(
x
x
N
)
)

z
N
}, also written as
}, also written as
Prob{Z(x
Prob{Z(x
)
)

z} =
z} =
F(z
F(z
),
),
independent of
independent of
x
x
.
.
21 21
REGIONALIZED VARIABLES
REGIONALIZED VARIABLES
AND RANDOM FUNCTIONS
AND RANDOM FUNCTIONS (Cont) (Cont)

The univariate cdf of the RV Z(x) is used
The univariate cdf of the RV Z(x) is used
to characterize uncertainty about the
to characterize uncertainty about the
value z(x); the multivariate cdf is used to
value z(x); the multivariate cdf is used to
characterize joint uncertainty about the N
characterize joint uncertainty about the N
values z(
values z(
x
x
1 1
) , ... , z(
) , ... , z(
x
x
N N
).
).

The bivariate (N=2) cdf of any two RVs
The bivariate (N=2) cdf of any two RVs
Z(
Z(
x
x
1 1
), Z(
), Z(
x
x
2 2
), is particularly important
), is particularly important
since conventional geostatistical
since conventional geostatistical
procedures are restricted to univariate
procedures are restricted to univariate
(F(x;z)) and bivariate distributions:
(F(x;z)) and bivariate distributions:
F(
F(
x
x
1 1
,
,
x
x
2 2
;z
;z
1 1
,z
,z
2 2
)=Prob{Z(
)=Prob{Z(
x
x
1 1
)
)

z
z
1 1
,Z(
,Z(
x
x
2 2
)
)

z
z
2 2
}
}
22 22
REGIONALIZED VARIABLES
REGIONALIZED VARIABLES
AND RANDOM FUNCTIONS
AND RANDOM FUNCTIONS
(cont)
(cont)

The description just made is the definition
The description just made is the definition
of
of
stationarity.
stationarity.

It is an expert
It is an expert

s decision, who
s decision, who
decides
decides
what area
what area
A
A
is statistically homogeneous
is statistically homogeneous
(stationary)
(stationary)
.
.

Stationarity is necessary because it is the
Stationarity is necessary because it is the
only way we can pool data values coming
only way we can pool data values coming
from different locations (and/or times)
from different locations (and/or times)
into a single population.
into a single population.

The RV
The RV
Z(
Z(
x
x
)
)
are not independent from
are not independent from
each other.
each other.
Their set
Their set
is called
is called
the
the
Random Function
Random Function
{
{
Z(
Z(
x
x
),
),
x
x

A}.
A}.
23 23
REGIONALIZED VARIABLES
REGIONALIZED VARIABLES
AND RANDOM FUNCTIONS
AND RANDOM FUNCTIONS
(cont)
(cont)

It is important to always keep in mind that
It is important to always keep in mind that
the only real attributes are the
the only real attributes are the
regionalized variables,
regionalized variables,
z(
z(
x
x
),
),
i.e., the
i.e., the
actual samples. The rest is only a
actual samples. The rest is only a
construction necessary for geostatistical
construction necessary for geostatistical
modeling.
modeling.

Matheron defined stationarity based on
Matheron defined stationarity based on
the existence of moments of the Random
the existence of moments of the Random
Function
Function
{
{
Z(
Z(
x
x
),
),
x
x

A}.
A}.
The concepts may
The concepts may
be more obscure, but familiar at the same
be more obscure, but familiar at the same
time.
time.
24 24
CONVENTIONAL DEFINITION
CONVENTIONAL DEFINITION
OF STATIONARITY
OF STATIONARITY

Strict Stationarity: all the

Strict Stationarity: all the
multivariate statistical moments of
multivariate statistical moments of
the RF defined are the same. This is
the RF defined are the same. This is
also called
also called

Invariance under
Invariance under
Translation
Translation

, meaning every RV at
, meaning every RV at
every location
every location
x
x
has the same
has the same
distribution function as the others.
distribution function as the others.
25 25
CONVENTIONAL DEFINITION
CONVENTIONAL DEFINITION
OF STATIONARITY
OF STATIONARITY
(cont)
(cont)

Stationarity of Order 2: only the first two
Stationarity of Order 2: only the first two
moments (Expected value and Covariance) of the
moments (Expected value and Covariance) of the
RF are invariant. If the covariance exists
RF are invariant. If the covariance exists
(stationarity of Order 2), then the variance and
(stationarity of Order 2), then the variance and
the covariance will necessarily exist.
the covariance will necessarily exist.
{ } ( ) ( ) E Z m zdF z zf z dz
+ +

= = =

{ , } {[ ][ ]} { }
( )( ) ( , )
X Y X Y
X Y XY
Cov X Y E X m Y m E XY m m
E dx x m y m f x y dy
+ +

= =
=

26 26
CONVENTIONAL DEFINITION
CONVENTIONAL DEFINITION
OF STATIONARITY
OF STATIONARITY
(cont)
(cont)

Intrinsic
Intrinsic

stationarity is when the

stationarity is when the
expected value of the RF does not
expected value of the RF does not
depend on the location (as before),
depend on the location (as before),
and the variance of all increments
and the variance of all increments
h
h
is finite and does not depend on its
is finite and does not depend on its
location
location
x
x
:
:
( ) ( ) ( ) ( ) ( )
{ }
2
2 Var Y Y E Y Y = + = +

h u u h u u h
27 27
CONVENTIONAL DEFINITION
CONVENTIONAL DEFINITION
OF STATIONARITY
OF STATIONARITY
(cont)
(cont)

Quasi
Quasi
-
-
stationarity is when the
stationarity is when the
moments of the RF are inferred only
moments of the RF are inferred only
for some local areas within the
for some local areas within the
overall volume. This is in practice
overall volume. This is in practice
what is used most in
what is used most in
geostatistics
geostatistics
.
.
This is also called
This is also called

local
local

stationarity,
stationarity,
and commonly found when we have
and commonly found when we have
large volumes (estimation domains)
large volumes (estimation domains)
and many data to work with.
and many data to work with.
28 28
STATIONARITY: HYPOTHESES
STATIONARITY: HYPOTHESES
OR DECISION?
OR DECISION?

The most important decision is to define
The most important decision is to define
what part of the physical space
what part of the physical space
A
A
is
is
statistically
statistically

homogeneous
homogeneous

.
.

Can this decision be tested
Can this decision be tested
a priori
a priori
?
?

Stationarity implies that all data within
Stationarity implies that all data within
A
A
can be pooled together. Statistical
can be pooled together. Statistical
moments can be obtained, and all other
moments can be obtained, and all other
population properties are appropriate
population properties are appropriate
descriptors of the Random Function
descriptors of the Random Function
.
.
29 29
STATIONARITY: HYPOTHESES
STATIONARITY: HYPOTHESES
OR DECISION?
OR DECISION?
(Cont.)
(Cont.)

Any statistical test would be performed on
Any statistical test would be performed on
the model,
the model,
not
not
on the real space
on the real space
A
A
.
.

The decision of
The decision of

homogeneity
homogeneity

can only be
can only be
assessed
assessed
a posteriori
a posteriori
.
.

In addition, in most cases the decision is
In addition, in most cases the decision is
affected by practical reasoning: lack of
affected by practical reasoning: lack of
data, knowledge about the physics of the
data, knowledge about the physics of the
phenomenon, consequences or impact of
phenomenon, consequences or impact of
the variables being modeled, etc.
the variables being modeled, etc.
31 31
STATISTICAL ANALYSES
STATISTICAL ANALYSES
Provides a first look at the data. Also called Exploratory Provides a first look at the data. Also called Exploratory
Data Analysis (EDA). Data Analysis (EDA). Should include: Should include:
Histograms; Histograms;
Scattergrams Scattergrams and correlation plots; and correlation plots;
Q Q- -Q and P Q and P- -P plots; P plots;
Contour Maps; Contour Maps;
Moving Averages; Moving Averages;
Declustering techniques; Declustering techniques;
Etc. Etc.
The objective is to identify spatial and global features that The objective is to identify spatial and global features that
would allow for a more informed stationarity decision, such would allow for a more informed stationarity decision, such
as: as:
Trends; Trends;
Sampling errors; Sampling errors;
Presence of multiple populations; Presence of multiple populations;
Impact of outliers; Impact of outliers;
Etc. Etc.