Micromine 2014 (Version 15.

0) Training MG 311 – Geostatistics and Kriging

MG 311 – Geostatistics and Kriging

Table of Contents

LESSON 1 – GEOSTATISTICAL ESSENTIALS ..........................................................................................1

INTRODUCT ION ............................................................................................................................................ 1
QUANT IFYING SPAT IAL CORRELAT ION ................................................................................................................ 3
Measuring the Difference between Sample Pairs ..................................................................................... 3
Comparing Sample-pair Differences for all Locations................................................................................ 4
LESSON 2 – SEMI-VARIOGRAMS ............................................................................................................7
INTRODUCT ION ............................................................................................................................................ 7
Review ................................................................................................................................................. 7
THE EXPERIMENTAL VARIOGRAM ...................................................................................................................... 8
Choosing Sample Pairs .......................................................................................................................... 8
What Makes a Good Experimental Variogram? ...................................................................................... 14
DISPLAYING A VARIOGRAM ............................................................................................................................ 15
Choosing a Lag Interval and Number of Intervals .................................................................................. 15
Keeping Everything Synchronised ......................................................................................................... 16
The Gold Project Data ......................................................................................................................... 17
Working with the Chart Display ............................................................................................................ 20
Display Modes ..................................................................................................................................... 20
LESSON 3 – VARIOGRAM MODELLING.................................................................................................23
INTRODUCT ION .......................................................................................................................................... 23
DESCRIBING A VARIOGRAM MODEL ................................................................................................................. 23
Component Types ............................................................................................................................... 24
FITTING A VARIOGRAM MODEL ....................................................................................................................... 25
Alternative Method for Identifying Structures ........................................................................................ 29
Validating the Model............................................................................................................................ 29
WHY F IT A THEORET ICAL VARIOGRAM? ............................................................................................................ 29
Interpreting a Fitted Theoretical Model ................................................................................................. 30
LESSON 4 – THE VARIOGRAM MAP ......................................................................................................33
INTRODUCT ION .......................................................................................................................................... 33
A NISOTROPY ............................................................................................................................................. 33
MEASURING A NISOTROPY ............................................................................................................................. 35
Choosing Initial Parameters ................................................................................................................. 36
Measuring Directions ........................................................................................................................... 41
Passing Map Directions to Other Functions............................................................................................ 44
INTERPRETING T HE VARIOGRAM MAP ............................................................................................................... 46
LESSON 5 – MODELLING 3D VARIOGRAMS .........................................................................................49
THE W ORKFLOW ........................................................................................................................................ 49
Finding Workflow Shortcuts ................................................................................................................. 49
DOWNHOLE: T HE NUGGET EFFECT .................................................................................................................. 51
OMNIDIRECTIONAL: EVALUAT ING DATA ............................................................................................................ 51
Quickly Filling Out the Semi Variograms Form ....................................................................................... 53
VARIOGRAM MAP: A NISOTROPY ..................................................................................................................... 55
DIRECT IONAL: MODELLING THREE A XES ........................................................................................................... 55

© Copyright MICROMINE 2014 Revision 2014-11

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Setting Directions ................................................................................................................................ 56

Choosing Tolerances ........................................................................................................................... 56
Adjusting the Lag Intervals .................................................................................................................. 58
Fitting the Theoretical Models .............................................................................................................. 60
USING A VARIOGRAM CONTROL F ILE ............................................................................................................... 61
LESSON 6 – OTHER VARIOGRAM TYPES.............................................................................................. 68
INTRODUCT ION .......................................................................................................................................... 68
VARIOGRAM T YPES ..................................................................................................................................... 68
Pair-wise Relative (Proportional Effect Present) ..................................................................................... 68
Natural Log (Positively-skewed Population) ........................................................................................... 69
Indicator (Mixed Populations Present) .................................................................................................. 69
Uniform/Rank (Strongly Skewed Population) ......................................................................................... 71
LESSON 7 VARIOGRAM CROSS-VALIDATION...................................................................................... 73
INTRODUCT ION .......................................................................................................................................... 73
Analysing the Result ............................................................................................................................ 74
Conditional Bias .................................................................................................................................. 74
Limitations .......................................................................................................................................... 74
THE PROCESS ............................................................................................................................................ 75
LESSON 8 – ORDINARY KRIGING ........................................................................................................ 82
INTRODUCT ION .......................................................................................................................................... 82
Kriging Advantages ............................................................................................................................. 82
ORDINARY KRIGING WIT H M ICROMINE ............................................................................................................. 83
Interpolating Multiple Elements or Domains .......................................................................................... 84
Choosing Data for OK Interpolation ...................................................................................................... 84
Using Variograms to Determine Initial Search Radii ............................................................................... 85
The Iron Data ..................................................................................................................................... 86
Preparing for Resource Classification .................................................................................................... 91
WHEN NOT TO USE ORDINARY KRIGING ........................................................................................................... 91
LESSON 9 – KRIGING VARIATIONS (OPTIONAL) ............................................................................... 94
INTRODUCT ION .......................................................................................................................................... 94
Universal Kriging (Kriging with a Trend) ............................................................................................... 94
Log-normal Kriging.............................................................................................................................. 95
Ordinary Kriging (Relative Variograms) ................................................................................................. 95
Rank Kriging ....................................................................................................................................... 95
INDICATOR KRIGING ................................................................................................................................... 95
Non-linear Transformation ................................................................................................................... 96
Grade Estimation ................................................................................................................................ 96
Indicator Kriging Approaches ............................................................................................................... 98
Choosing Data for IK Interpolation ....................................................................................................... 98
UNSUPPORT ED KRIGING MET HODS ................................................................................................................104
Simple Kriging....................................................................................................................................104
Co-kriging..........................................................................................................................................105
Disjunctive Kriging .............................................................................................................................105
A PPENDIX 1: A NSWERS TO S ELECTED EXERCISES ..............................................................................................108
A PPENDIX 2: T HE IRON VARIOGRAMS .............................................................................................................111

Revision 2014-11 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

A PPENDIX 3 – VARIOGRAM FAN W ORKFLOW .................................................................................................... 114

Examples

Example 1.1: Quantify the temperature difference between Perth and nearby sample locations ........................ 3
Example 2.1: Choosing sample pairs ............................................................................................................. 8
Example 2.2: Calculate semi-variance at a lag of 30 m ................................................................................. 11
Example 9.1: IK in a gold deposit using five cutoffs...................................................................................... 97

Tables

Table 1.1: Maximum temperatures within 80 km of Perth for 1st December 2008 ............................................ 4
Table 8.1: Filter logic for the IRON_HG_BX and IRON_LG domains ............................................................... 87
Table 8.2: Model and output files for the first and subsequent runs............................................................... 87
Table 8.3: Search radii and minimum points requirements ............................................................................ 88
Table 9.1: MIK search radii and minimum points requirements ................................................................... 103

Figures

Figure 1.1: Australian maximum temperatures for 1st December 2008 ........................................................... 1
Figure 1.2: Squared temperature difference vs. distance from Perth............................................................... 5
Figure 2.1: Iron ore samples (% Fe) ............................................................................................................. 8
Figure 2.2: E-W sample pairs, lag = 30 m...................................................................................................... 9
Figure 2.3: E-W sample pairs, lag = 60 m...................................................................................................... 9
Figure 2.4: H-scattergrams at lag intervals of 30 m ...................................................................................... 10
Figure 2.5: Lag tolerance ............................................................................................................................ 16
Figure 3.1: Features of a one-component variogram .................................................................................... 24
Figure 3.2: Theoretical variogram model types............................................................................................. 25
Figure 4.1: Geometric (a) and zonal (b) anisotropy ...................................................................................... 34
Figure 4.2: Values falling within each cell .................................................................................................... 35
Figure 4.3: Anisotropy and the direction of greatest continuity...................................................................... 36
Figure 4.4: One-cell direction and lag overlaps include all cells in a 3 × 3 group ............................................ 37
Figure 4.5: Pitch variogram map and resulting axes in their 3D orientations ................................................. 45
Figure 5.1: The variogram modelling workflow............................................................................................. 50
Figure 5.2: Angular tolerance and bandwidth ............................................................................................... 56
Figure 5.3: Axis 3 is sometimes sub-parallel to the drilling ............................................................................ 58
Figure 6.1: Decision tree for choosing variogram types................................................................................. 71
Figure 7.1: Decision tree for variogram validation ........................................................................................ 80
Figure 8.1: IDW and OK grade estimates from redundant data ..................................................................... 82
Figure 8.2: Using proportions of the sill to determine radii ............................................................................ 85
Figure 9.1: Decision tree for choosing a kriging method ............................................................................. 104

© Copyright MICROMINE 2014 Revision 2014-11

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

MG 311 – Geostatistics and Kriging

Table of Contents

Sidebars

Using the temperature data ......................................................................................................................... 3

Exercises

Exercise 2.1: Build an E-W experimental variogram...................................................................................... 12

Exercise 2.2: Display a downhole variogram ................................................................................................ 17
Exercise 3.1: Model a downhole variogram .................................................................................................. 27
Exercise 4.1: Display a variogram map ........................................................................................................ 38
Exercise 4.2: Refine the variogram map ...................................................................................................... 40
Exercise 4.3: Measure the 3D direction of greatest continuity ....................................................................... 41
Exercise 4.4: Create a variogram control file ................................................................................................ 45
Optional Exercise 5.1: Create omni-variograms ............................................................................................ 51
Optional Exercise 5.2: Evaluate the variograms and find an initial lag interval ................................................ 54
Exercise 5.3 (Part 1): Set up variogram directions using a variogram control file ........................................... 57
Exercise 5.4: Fit theoretical models to the three axes ................................................................................... 62
Exercise 5.5: Save and embed a variogram control file ................................................................................. 63
Optional Exercise 6.1: Create median indicator-transformed variograms ........................................................ 70
Exercise 7.1: Cross-validate the variogram model ........................................................................................ 75
Exercise 7.2: Create a scatter-plot of the cross-validation results .................................................................. 76
Exercise 7.3: Display the validation results in Vizex ...................................................................................... 78
Advanced Exercise 7.4: Complete the validation .......................................................................................... 79
Exercise 8.1: Determine initial search radii using distances to reach proportions of the total sill ...................... 86
Exercise 8.2: Complete the first OK interpolation run ................................................................................... 87
Exercise 8.3: Complete the subsequent OK runs ......................................................................................... 89
Optional Exercise 9.1: Calculate mIK grade deciles ...................................................................................... 99
Advanced Exercise 9.2: Calculate mIK metal deciles ..................................................................................... 99
Optional Exercise 9.3: Complete the first mIK interpolation run ...................................................................101
Optional Exercise 9.4: Complete the subsequent mIK runs ..........................................................................103

Revision 2014-11 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 1 – Geostatistical
Essentials
Duration: 20 minutes

This lesson introduces spatial correlation, which is an important assumption

in most grade estimation methods.

After this lesson you’ll be able to:

 Understand the concept of spatial correlation

 Quantify spatial correlation by measuring the squared sample value
differences at different distances.

Introduction
The properties of many natural phenomena become more alike as the dis-
tance between any two measurement locations decreases, and less alike as
the distance between measurement locations increases. For example, in your
daily life you expect the maximum temperature in your neighbourhood to be
about the same as the neighbouring suburb, but you also expect it to be
rather different to a city thousands of kilometres away.

Figure 1.1: Australian maximum temperatures

for 1st December 2008

Figure 1.1 shows contours of the maximum temperature (in degrees Celsius)
across Australia on December 1st, 2008. The map reveals several aspects of
the temperature that day:

© Copyright MICROMINE 2014 1

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:  Values in the southern half of the map appear to be more consistent in
an east-west direction than they do in a north-south direction, with
some contour bands crossing the entire continent.
 Values in the northern half of the map are generally higher than those in
the south and do not show a preferred orientation.

If you concentrate on the south-western corner of the map you can quantify
some temperature-distance relationships using Perth as a reference point:

 The temperatures in Perth and Kalgoorlie, about 540 km to the east-

northeast, differ by around three degrees.
 The temperatures in Perth and Meekatharra, about 635 km to the north-
northeast, differ by about thirteen degrees.

Although the distance between Perth and the two towns varies by less than
20%, the north-south temperature difference is four times the east-west
difference. The map clearly shows this relationship – the contour bands ex-
tend a long way in an east-west direction but show little continuity in a
north-south direction. You see similar results if you consider any other pair of
locations in the southern half of the continent.

The relationship between separation distance and value difference (e.g. tem-
perature or ore grade) is known as spatial correlation. If something is spa-
tially correlated you would expect the difference between nearby measure-
ments to be small, and for the difference to increase as the distance between
measurements increases. You can describe the temperature map as showing
strong east-west spatial correlation, because temperature differences remain
small over large distances in that direction. However, temperatures are less
spatially correlated in a north-south direction, as shown by the rapid change
in temperature with increasing distance between location pairs.

In this example the amount of spatial correlation also varies with direction.
This effect is known as anisotropy and is fully described in Lesson 3. In
comparison, something with properties that change equally in all directions is
said to be isotropic.

A final point to consider is the difference between very closely- or very

widely-separated location pairs. For something with perfect spatial correlation
you would expect the difference between two measurements to decrease to
zero when the distance between them is zero. Although not shown on the
map, the temperature data suggests this is almost the case. The difference
between Perth City and Perth Airport, around 10 km apart (effectively zero at
the scale of the map), is only 0.5 degrees. In spatial terms you can say val-
ues at very short distances are highly correlated. However, real-world differ-
ences are rarely zero at zero distance, owing to measurement errors and
other random effects. In geostatistics this effect is called the nugget effect,
and is described in Lesson 3.

At the other extreme, the difference at very large distances appears to be

almost random; depending on which locations you pair you obtain differ-
ences ranging from zero to nearly 30°C. Thus, you can say temperatures are
not correlated at large distances. In geostatistics the distance at which
spatial correlation ceases to exist is called the range, and is described in
Lesson 3.

2 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Quantifying Spatial Correlation

Notes:

Visually interpreting the spatial correlation in Figure 1.1 was easy because
the contours had already been created for you. However, the main reason
for geostatistical analysis is to calculate the contours in the first place, so you
must now turn your attention to the raw data.

Using the temperature data

If you’d like to work these examples yourself, you’ll find the raw
temperature data for 768 recording stations across Australia within the
training project, in the DATA file Oz_Max_Temps_08_12_01.

Data reproduced courtesy of the National Climate Centre, Australian Bureau

of Meteorology.

Measuring the Difference between Sample Pairs

The conclusions you drew from the visual inspection of Figure 1.1 were:

 Temperatures appeared to be correlated over very large E-W distances

and somewhat shorter N-S distances.
 The temperature difference between measurement pairs approached
zero as the distance between them approached zero.
 The temperature difference between very widely separated
measurement pairs appeared to be random.

However, the assessment was purely qualitative and you made no attempt to
quantify the differences. You will begin to quantify them now.

Example 1.1: Quantify the temperature difference between

Perth and nearby sample locations

Consider the maximum temperatures recorded at mainland stations within 80

km of Perth Metro, listed in order of increasing distance from Perth (Table
1.1). A simple way to quantify the temperature differences might be to sub-
tract each station’s temperature from that at Perth Metro, shown in the
Temp Diff column. Clearly the temperature differences gradually increase
with distance, from 0.5°C at a distance of 9.75 km to 3.1°C at 79.4 km.

However, the Temp Diff column contains both positive and negative values.
You only need to know the size of the differences, not their signs. One way
to eliminate the negatives is to square the numbers, as listed in the Diff
Squared column. A glance at these values shows that, with exception of
some occasional glitches, the squared temperature difference does gradually
increase with distance, confirming that the data is spatially correlated.

© Copyright MICROMINE 2014 3

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Plotting the squared temperature difference between Perth Metro and all re-
cording stations within 1,800 km, and overlaying them with a polynomial
trend line produces the result in Figure 1.2. An interesting property of the
trend line is its gradual levelling-off at a distance of about 1500 km, which
suggests there is no systematic increase in temperature differences beyond
that distance. The increasing level of noise at longer distances also hints that
the relationship becomes more random as the distance increases.

Table 1.1: Maximum temperatures within 80 km of Perth for

1st December 2008
Dist
Max
From Temp Diff
Station Temp
Perth Diff (°C) Squared
(°C)
(km)
Perth Metro 27.2 0.00 0.0 0.00

Perth Airport 27.7 9.75 0.5 0.25

Gooseberry Hill 27.8 16.84 0.6 0.36

Gosnells City 28 17.63 0.8 0.64

Jandakot Aero 26.5 20.01 -0.7 0.49

Bickley 26.2 26.57 -1.0 1.00

Pearce RAAF 29.5 30.93 2.3 5.29

Medina Research Centre 25.9 33.72 -1.3 1.69

Gingin Aero 29.4 50.15 2.2 4.84

Karnet 26.1 60.29 -1.1 1.21

Mandurah 24.1 67.97 -3.1 9.61

Northam 30.3 79.44 3.1 9.61

Data courtesy of National Climate Centre, Australian Bureau of Meteorology
© Commonwealth of Australia 2008

Comparing Sample-pair Differences for all Locations

Comparing Perth with its neighbours is a useful learning exercise, but the
result hardly represents all of the temperature variations across Australia.
Ideally you should compare every recording station with its neighbours and
then summarise the result by averaging all of the graphs. Each station would
be the reference point for its own graph, and a neighbouring point for all
other graphs. However, constructing 768 graphs would be laborious and they
would be difficult to average. Incorporating anisotropy would also be difficult.

Fortunately, these problems are solved by using a semi-variogram, which

is a more efficient way to calculate and display the average of the squared
differences for all of the data. The construction of a semi-variogram is intro-
duced in the next lesson.

4 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:

Figure 1.2: Squared temperature difference

vs. distance from Perth

Before you continue

A solid understanding of the classical statistical concepts mean, variance, and

standard deviation is essential for the following discussions, so be sure to
complete MG 205 – Statistical Essentials before continuing.

© Copyright MICROMINE 2014 5

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 1 Summary

This lesson introduced the concept of spatial correlation using a familiar

weather map as an example. The topics covered so far are:

Many natural phenomena are spatially correlated, meaning their

properties become more alike as the distance between any two
measurement locations decreases, and less alike as the distance
between measurement locations increases.
Something with properties that vary with direction is said to exhibit
anisotropy. In comparison, something with properties that change
equally in all directions is isotropic.

To quantify the difference between sample locations:

Measure the distance between them, then

Measure the value difference between them, and
Square the difference

Additional Resources and Further Reading

Australian Bureau of Meteorology:

[Link]

6 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 2 – Semi-variograms
Duration: 90 minutes

This lesson will teach you to construct an experimental semi-variogram.

After this lesson you’ll be able to:

 Manually pair samples at set distances and directions

 Interpret a scatter plot of the paired values
 Manually build an experimental semi-variogram
 Understand what makes a good experimental semi-variogram
 Develop a consistent workflow for constructing experimental semi-
variograms in Micromine
 Construct and display an experimental downhole semi-variogram in
Micromine.

Introduction
A semi-variogram (often abbreviated to variogram) is a graph of the aver-
age value differences (and thus the correlation) between sample pairs at dif-
ferent distances, and it allows you to visualise spatial correlation within a
study area. It is similar to, but more versatile than, the simple graph of
Figure 1.2 because it incorporates all possible location pairs instead of using
just one reference point (e.g. Perth Metro).

In geostatistical modelling you use two different variograms. You begin by

constructing an experimental variogram, which displays the value differ-
ences of the raw data at various distances. You then fit a theoretical vario-
gram, which is a best-fit mathematical model that you use to infer any
spatial relationships.

The theoretical variogram is also used to calculate the weighting applied to

each input sample during grade interpolation. Methods that rely on vario-
grams are called kriging interpolators and are described in Lesson 8.

Review
In MG 205 and MG 301 you statistically described the input data, applied a
geological cutoff (by assigning the orebody wireframe) and top-cut to the
assays, and generated a composite file containing regular 2 m intervals. You
also calculated 3D coordinates for the centroids of the composites. In prep-
aration for grade interpolation you created an empty block model, which you
assigned to the orebody wireframe using true sub-blocks; you also created
an initial search ellipsoid and defined your minimum point strategy, and test-
ed for proportional effect.

The work completed so far is sufficient for IDW interpolation, but you must
perform variogram modelling if you wish to use kriging.

© Copyright MICROMINE 2014 7

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Before you continue

Make sure you’ve completed the exercises in MG 205 – Statistical Essentials

and Lessons 1 through 5 of MG 301 – Block Modelling.

The Experimental Variogram

The experimental variogram is a graph of the average squared value differ-
ence versus distance. To simplify the calculation you group the sample pairs
into set distances and then calculate the average squared difference for all
pairs in each group, instead of measuring all possible distances like Example
1.1. The steps needed to calculate an experimental variogram are:

1. Take each sample pair at a given separation distance, or lag (h), and
measure the value difference between them.
2. Square the difference.
3. Sum all of the squares.
4. Divide the sum by twice the number of pairs.

5. The result is the semi-variance, or gamma ( ), at that separation

distance, in the square of the original measurement units.

Choosing Sample Pairs

The trickiest part of calculating an experimental variogram is pairing the
samples, since you must consider both distance and direction from the refer-
ence point when choosing the second sample. The weather data from Lesson
1 is irregularly spaced in distance and direction, so before you construct an
experimental variogram you’ll explore the concept of sample pairing with a
simpler dataset consisting of samples on a regular grid.

Example 2.1: Choosing sample pairs

Consider the iron ore samples (in percent Fe) of Figure 2.1, collected on a
regular grid spacing of 30 m. You can pair samples in many different ways,
but the regular grid spacing suggests you should work in either a north-south
or east-west direction and in distance increments of 30 m.

44 NS 40 42 40 39 37 36 NS

42 NS 43 42 39 39 41 40 38

37 37 37 35 38 37 37 33 34

35 38 NS 35 37 38 36 35 NS

36 35 36 35 34 33 32 29 28

38 37 35 NS 30 NS 29 30 32

Figure 2.1: Iron ore samples (% Fe)

8 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Begin by pairing samples in an east-west direction, separated by the closest Notes:

possible spacing of 30 m (Figure 2.2). The distance between sample pairs is
called the lag and is given the symbol h, the first (reference) point is known
as the tail, and the paired point the head (e.g. Goovaerts, 1997, p.26).
Some workers (e.g. Pyrcz & Deutsch, 2014, p.79) use the opposite head/tail
notation. As you work down the western-most samples you pair 44 (tail) with
NS (head), which is 30 m to the east. However, NS indicates that location
was not sampled so this is not a valid pair. Continuing, you pair 42 with NS
(another invalid pair), and so on down the column until you pair 38 with 37.

Figure 2.2: E-W sample pairs, lag = 30 m

Moving to the next column you then pair NS with 40 (not a valid pair), NS
with 43, and so on until you pair 37 with 35. On average each sample point
is used twice: as the tail of one pair and as the head of a different pair.

A scatter plot of the head versus the tail values for all valid pairs, called an
h-scattergram, produces the result in Figure 2.4(a). Pairs whose head and
tail values are equal fall on the 45° correlation line; the others fall further
from the line as their head/tail differences increase. You can see that alt-
hough the points do follow the trend of correlation line there is still some
scatter either side of it. Given that pairs falling close to the line are strongly
correlated, you can use the width of the scatter cloud around the line to indi-
cate the amount of correlation for pairs separated by that distance.

Once you have assessed the pairs at 30 m E-W spacing, you pair E-W
samples at 60 m spacing (Figure 2.3). At this interval you must skip one
sample; starting with the western-most samples you pair 44 with 40, 42 with
43, and so on until you pair 38 with 35.

Figure 2.3: E-W sample pairs, lag = 60 m

© Copyright MICROMINE 2014 9

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: The second column begins at NS as before, but this time it is paired with 42,
the second NS is also paired with 42, 37 is paired with 35, and so on until
you pair 37 with NS. The h-scattergram in Figure 2.4(b) shows that the width
of the cloud has increased slightly, meaning these points are slightly less
correlated than those at a lag of 30 m.

Figure 2.4: H-scattergrams at lag intervals of 30 m

As you work through the data, pairing samples at increasing E-W distances,
you produce the results shown in Figure 2.4(c) to (h). It is clear that as the
lag increases fewer pairs are created and the width of the cloud increases.

10 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

This result confirms your understanding of spatial correlation: sample values Notes:
become less alike as the distance between them increases.

Although the h-scattergrams show the data becoming less correlated with
increasing lag, using them to model real data would produce reams of graphs
that are all-but impossible to interpret. Instead you construct an experi-
mental variogram, whose value, the semi-variance, at each lag distance rep-
resents the width of the scatter cloud on the corresponding h-scattergram.

Example 2.2: Calculate semi-variance at a lag of 30 m

Using the same iron ore samples as Example 2.1 you can now establish the
degree of spatial correlation in an E-W direction by using an experimental
variogram. In order to fully calculate the variogram you must calculate the
semi-variance at varying distances, starting at 30 m.

You calculate the 30 m variogram value by measuring and squaring the dif-
ference between each pair separated in an E-W direction by 30 metres,
summing the squares, and then dividing by twice the number of pairs. Doing
so produces the following equation, where:

 = semi-variance in units of (% Fe)2

* signifies a measured, not theoretical, value, and
(30) indicates this value corresponds to a lag of 30 m

*(30) =
[(37 - 37)2 + (35 - 38)2 + (36 - 35)2 + (38 - 37)2 + (37 - 37)2 +
(35 - 36)2 + (37 - 35)2 + (40 - 42)2 + (43 - 42)2 + (37 - 35)2 +
(36 - 35)2 + (42 - 40)2 + (42 - 39)2 + (35 - 38)2 + (35 - 37)2 +
(35 - 34)2 + (40 - 39)2 + (39 - 39)2 + (38 - 37)2 + (37 - 38)2 +
(34 - 33)2 + (39 - 37)2 + (39 - 41)2 + (37 - 37)2 + (38 - 36)2 +
(33 - 32)2 + (37 - 36)2 + (41 - 40)2 + (37 - 33)2 + (36 - 35)2 +
(32 - 29)2 + (29 - 30)2 + (40 - 38)2 + (33 - 34)2 + (29 - 28)2 +
(30 - 32)]  (2 × 36)

*(30) = 1.5139(%)2

Thus you can describe the experimental semi-variance at a lag of 30 m as

being 1.51 percent squared Fe. You’ll calculate the semi-variance at 60, 90,
and 120 m, and graph the result, in the following exercise.

© Copyright MICROMINE 2014 11

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Exercise 2.1: Build an E-W experimental variogram

Starting with the result calculated in Example 2.2, calculate the remainder of
the E-W variogram and plot it on the supplied graph (on Page 14). The first
pair has been highlighted for you.

[HINT: Do NOT include pairs with NS values; DO include pairs whose

difference is zero.]

44 NS 40 42 40 39 37 36 NS

42 NS 43 42 39 39 41 40 38

37 37 37 35 38 37 37 33 34

35 38 NS 35 37 38 36 35 NS

36 35 36 35 34 33 32 29 28

38 37 35 NS 30 NS 29 30 32

At 60 m E-W spacing:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 – 40 16

Sum of squares:
Number of pairs × 2:
Semi-variance:

12 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

At 90 m E-W spacing: Notes:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 – 42 4

Sum of squares:
Number of pairs × 2:
Semi-variance:

At 120 m E-W spacing:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 – 40 16

Sum of squares:
Number of pairs × 2:
Semi-variance:

© Copyright MICROMINE 2014 13

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:

What Makes a Good Experimental Variogram?

What should you look for when you create an experimental variogram? From
an ideal orebody with a good coverage of data, you would look for a few fun-
damental features:

 Low values near the origin. This indicates a strong correlation between
closely-spaced data pairs.
 Indication of a plateau (called the sill). This indicates that the data is
more orderly than chaotic. The points should ideally stay below the sill
but this is seldom true for all points.
 Smooth increase in value. The shape of the curve and proximity of
points to the curve give a good impression of data quality.
 A reasonable range of influence. The larger the range, the greater
the distance over which the data is correlated.

Variograms can be hard to use for a few different reasons:

 Too much low-grade material in the domain. This can be remedied

by reviewing and adjusting your interpretations and wireframes.
 Mixed grade populations or proportional effect. You may need to
subdivide the data or use relative variograms to overcome these
problems.
 The drillhole spacing is too large to reveal a useful variogram. A
perfectly regular drill pattern can emphasise this problem. The only
remedy for this is to drill additional holes, but because each sample
contributes to many different pairs only one or two more closely-spaced
holes are needed to improve the variogram.

14 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Displaying a Variogram
Notes:

You display and model a variogram via Stats | Semi Variograms. The
Input group determines how the variogram is calculated and displayed. Al-
ways begin with the Mode set to Calculate. Once you’ve created a variogram
file you can then change to one of the Display Existing modes to avoid un-
necessary recalculations. Use the Direction to switch between Downhole,
Omnidirectional and Directional variograms. If your study area requires you
to use relative variograms you can use the Type to enable the Pair-wise rela-
tive transformation, which is detailed in Lesson 6.

Downhole variograms require Raw Data with Hole and From/To depth
values, whereas the omni- and directional variograms require 3D coordi-
nate values for each input point. The Semi variogram field is the one
whose values you intend to model.

You apply a data Transformation in the Data Values group. Aside from
not altering the data at all, the most common transformations are NATURAL
LOG and INDICATOR. For a NATURAL LOG transform you may optionally
supply an Additive constant, and for an INDICATOR transform you must
supply the Cutoff value. Strategies for selecting these transformations are
discussed in Lesson 9.

The Search Definition group changes according to the variogram (down-

hole, omnidirectional or directional) you’re modelling. Click the Semi Vario-
gram button to display the relevant dialog, where you define the interval
and number of Lags along with their Display properties. For directional
variograms you also define the Direction, Tolerance, and Bandwidth for
Azimuth and Plunge. The dialog contains multiple rows to make it easier to
experiment with different parameter combinations in one pass, with right-
click menu options to auto-fill each column.

In all cases you have the option to Show variance, which overlays the
global variance of the data on the variogram graphs. Although this may pro-
vide a useful tool for gauging the quality of a variogram, you should not
force the total sill to match this value (e.g. Goovaerts, 1997, p.103).

The Semi Variograms group has two output options. For nearly every ap-
plication you should select Show semi variograms; as you’ve seen, de-
scribing a variogram is a visual process. It’s also essential to enable Write
semi variogram file as this is the best way to redisplay existing variograms
without reprocessing the data.

Generally you should not enable Write H Scattergram file from the
H Scattergrams group unless you need to work with every data pair used
to draw the variograms. Selecting this option for this training project would
create a file containing several hundred million records! To use this option,
write the H Scattergram file for just the variogram under investigation.

Choosing a Lag Interval and Number of Intervals

Setting up an experimental variogram requires you to know or estimate sev-
eral parameters in advance, starting with the lag interval and number of
intervals. The lag interval (Δh) is simply the step size used to draw the

© Copyright MICROMINE 2014 15

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: graph (in the preceding example it was 30 m), and the lag (h) is the current
multiple of the lag interval under investigation. The lag interval is obvious
when the samples are regularly spaced, in which case it should be a multiple
of the sample spacing. Some adjustment is usually needed to find the best
lag interval when the sample spacing is irregular, and this is easily performed
using an interactive slider in the variogram Properties window.

Even so-called regularly spaced samples are rarely separated by exactly the
current lag, and a tolerance is normally applied so that the current lag can
find more pairs. Micromine automatically adds a tolerance of one-half of the
lag interval to either side of the current lag, as shown on Figure 2.5.

Figure 2.5: Lag tolerance

Micromine sets the lag tolerance by adding half of the
lag interval to either side of the current lag.

Using a lag interval that is too small will create a noisy and hard-to-interpret
variogram because some of the lag distances will be too short to find enough
pairs. In contrast, using a large lag interval may be desirable when the data
is sparse or irregularly spaced, since it also increases the tolerance and con-
sequently allows each lag to find more pairs.

The number of intervals should be large enough to span the study area at
the chosen lag interval. However, a variogram has 180° symmetry because
the value differences are squared when you calculate the semi-variance. This
just means you get the same result working from east to west as you do by
working from west to east. Because of this symmetry you only need to span
half of the study area. A simple rule of thumb is to choose enough intervals
to cover one-half of the diagonal dimensions of the study area, accounting
for the proportions of the mineralised zone (Coombes, 1997). This parameter
may also be interactively changed via the Properties window.

Keeping Everything Synchronised

A number of different variograms are needed to model even a simple study
area, creating many opportunities to mistakenly overwrite a previous file or
use mismatched form sets. A systematic approach is critical to avoid these
problems, especially if you’re working on a reportable resource. A workflow
that may help minimise problems is:

16 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

1. For each new variogram, set the Mode to Calculate and choose a Type. Notes:
2. If needed, re-select the appropriate Raw Data fields (these change
whenever you switch to or from a downhole variogram).
3. Immediately enter the name of the new Semi variogram File.
4. Immediately Save the Semi Variograms form as a new form set.
5. Click the Semi Variograms button in the Search Definition and
define the lags, angles, and tolerances.
6. Click OK to run and then close the variogram.
7. Change the Mode to Display Existing.
8. Re-save the form set and click OK to apply the change.

Entering the new file name and saving the form set before defining the lags
reduces the risk of forgetting to do it later on, and changing to Display Exist-
ing and re-saving the form set makes it easy to revisit existing (and already
calculated) variograms. Although it is not necessary to use this double-save
workflow it has been proven to minimise file and form set mismatches.

The Gold Project Data

The iron data used for this course does not exhibit all of the features nec-
essary for variography, and because of these limitations you will use the gold
dataset to learn about displaying and modelling variograms. You will then
switch back to the iron data for the remainder of the course, using
variograms that are provided for you. Refer to Page 2 of MG 301 – Block
Modelling for a geological summary of the gold project.

Drillholes are mostly inclined towards the east. Drill lines are spaced at
intervals from 5 m to 100 m, averaging around 15 m, and the hole spacing
varies from 4 m to over 50 m, averaging about 25 m. The average hole
depth is about 70 m. Assay data covers the entire study area and reaches a
maximum grade of 3390 ppm. Grades show a typical log-normal distribution
with some weakly developed mixed populations. Nearly 99% of the intervals
are 1 m long and compositing was not used.

Exercise 2.2: Display a downhole variogram

The gold training dataset provides good data for variogram modelling be-
cause it comprises a log-normal population and produces variogram models
with geometric anisotropy. (These terms are explained at the appropriate
stages in the text.) For simplicity and brevity it contains one element and one
domain, but in the workplace you should perform these steps on a per-
element and per-domain basis.

You’ll begin by viewing the data in Vizex to assess the trend of the grades
and gain an impression of the average downhole depth of the orebody:

1. Load the Gold Drillholes Vizex Saved View and inspect the data.

© Copyright MICROMINE 2014 17

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: The samples are displayed as points sorted by grade, coloured by deciles,
with depth testing disabled. This configuration produces a maximum inten-
sity projection, which makes the highest grades more prominent and as-
sists with the detection of structural trends in the grade data (Cowan, 2014).

2. Switch to Looking North orientation and note that most of the

holes are inclined to the east.
3. Inspect the mineralised intervals (anything shown in colour) and then
use the Measure Tool to obtain a rough downhole thickness of the
mineralised zone, which should be around 60 m.

Half of 60 m is 30 m, so this is the distance you need to span. The 1 m drill-

hole sample length gives you the lag interval. Dividing 30 m by the lag
interval (1 m) produces 30 intervals. Don’t be too specific about this value;
you can always start with too many intervals and reduce them once you
display the graph.

Now that you are familiar with the data you can display the downhole vario-
gram. To do so:

1. Select Stats | Semi Variograms from the main menu.

2. Set the Mode to Calculate, the Direction to Downhole, and the Type
to Pair-Wise Relative Variogram.

Variograms of log-normal data often need a pair-wise relative variogram

or natural log transformation. Although the two methods produce essen-
tially the same result, a relative variogram may appear smoother and is gen-
erally more stable during kriging. They are both detailed in Lesson 6.

3. Select Gold_Assay as the Raw Data File.

4. Ensure the Raw Data field names are set to HOLE , FROM, TO, and
Au1_ppm_CUT.
5. Enable the Ignore blanks numeric exception and ensure the
Transformation is set to NONE .

You may wish to optionally enable Show Variance in the Search Definition
group to display the global variance of the data as a horizontal line. How-
ever, in this exercise it does not contribute any meaningful information.

6. Click the Downhole Semi Variograms button and complete the form
as shown in this screenshot:

The various display modes are described at the end of this lesson.

18 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

7. Back on the main dialog enable Show semi variograms and Write Notes:
semi variograms to file in the Semi Variograms group. Enter the File
name SVG_AuCut_Downhole.
8. Switch to the Chart tab and set the Title to Downhole Au (cut) .
9. Switch to the Axis tab and set the X Axis Title to Lag Distance (m) and
the Y Axis Title to Semi-variance.
10. Save the form as a form set with the Title AuCut Downhole.
11. Click OK to calculate and display the experimental downhole variogram.
Inspect the marker points, which slowly shrink as the lag increases.

The gradual decrease in marker size indicates that fewer pairs are available
at longer lag distances, which is typical for downhole variograms. Your dis-
play should resemble this screenshot:

At this point you might accidentally overwrite these settings with the next
variogram. A good habit for avoiding confusion is to immediately switch the
mode to Display Existing and re-save the form set.

12. Click the Form button at left of the Chart toolbar, change the Mode
to Display Existing, and re-save the form set.
13. Click OK to apply the change.
14. Leave the Vizex view and variogram display open in preparation for the
next lesson.

© Copyright MICROMINE 2014 19

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Working with the Chart Display

Micromine presents the Chart Specific toolbar whenever you display a vari-
ogram. Items in this toolbar include:

 Semi variogram model : Shows or hides a fitted variogram model.

Displays the Chart Controls if they are not already visible; use them to
interactively fit a theoretical model.
 Previous/Next : When more than one experimental variogram is
defined, lets you step through the individual variograms. The same as
pressing Page Up and Page Down on the keyboard.
 Show Together : When more than one experimental variogram is
defined, toggles the display between showing individual and all
variograms.
 Create Variogram Control File : For directional variograms created
from a variogram map, creates a variogram control file containing the
model parameters. Detailed in Lesson 4.

You’ll learn about model fitting in the next lesson.

Display Modes
You configure the appearance of each displayed variogram during the initial
setup by enabling or disabling the Display options on the Semi Vario-
grams dialog. Or, simply use the options in the variogram Properties win-
dow to interactively change the display. The following modes are available:

 Line: Draws a line connecting the individual markers, making it easier to

evaluate the shape of the curve.
 Scale: Scales the markers in proportion to the number of contributing
pairs, providing an indication of quality (lags with very few pairs indicate
a poorer estimate).
 Pairs: Labels each marker with the number of contributing pairs. Use as
an alternative to Scale.
 Marker: Defines the marker shape. Set it to None to disable markers.
 Colour: Defines the marker colour.

To hide a less likely variogram candidate (but keep it in the file), set the
Marker to None and disable all other display options.

20 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 2 Summary Notes:

This lesson introduced h-scattergrams and experimental variograms. The

topics covered so far are:

To manually produce an h-scattergram:

Visually pair samples at a set distance and direction, then

Plot the values using the first (tail) sample as the X-axis value and the
second (head) sample as the Y-axis value

To manually create an experimental variogram:

Visually pair samples at a set distance and direction, then

Calculate the difference between the two values, and
Square the difference, then
Sum the squares, and
Divide by twice the number of sample pairs, and
Repeat the above steps at each distance, then
Plot the results using lag as X-axis and gamma (semi-variance) as Y-axis

To identify a good experimental variogram:

Look for low values near the origin, and

An indication of a plateau, and
A smooth increase in value, and
A reasonable distance range

To display an experimental downhole variogram in Micromine:

Select Stats | Semi Variograms from the main menu, and

Set the Mode to Calculate and the Direction to Downhole, and the
Type to Variogram, and
Choose the Raw Data and optionally apply a Transformation, then
Click the Downhole Semi Variograms button, and
Define the interval and number of Lags along with their Display
properties, then
Enable Show semi variograms and Write semi variogram file, then
Click OK to calculate the variogram.

To choose a lag interval and number of intervals:

Use a lag interval at least equal to the sample spacing, and

Set the number of lags so that half the diagonal dimension of the study
area is covered.

© Copyright MICROMINE 2014 21

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 2 Summary

Good Practice

Because the modelling process requires you to repeatedly define slightly dif-
ferent experimental variograms a systematic workflow is vital; if you’re not
systematic you’ll accidentally overwrite files and form sets. A good way to
work is to:

1. Set the Mode to Calculate at the beginning of each new variogram, and
choose the appropriate Direction.
2. If needed, re-select the correct Raw Data fields.
3. Immediately enter the name of the new Semi variogram File.
4. Immediately Save the Semi Variograms form as a new form set.
5. Click the Semi Variograms button in the Search Definition and
define the desired lags.
6. Run then close the variogram.
7. Change the Mode to Display Existing.
8. Re-save the form set and re-run the semi variogram.

Help Topics

For information on: See:

Semi-variograms Statistics > Semi-variograms

Additional Resources and Further Reading

Coombes, J. (1997). Handy Hints for Variography. Snowden Associates: Perth

Cowan, E.J. (2014). X-Ray Plunge Projection – Understanding Structural

Geology from Grade Data. In Mineral Resource and Ore Reserve
Estimation: The AusIMM Guide to Good Practice . 2nd ed. The Australasian
Institute of Mining and Metallurgy: Melbourne.

Goovaerts, P. (1997). Geostatistics for Natural Resource Evaluation. (Applied

Geostatistics Series). Oxford University Press: New York

Isaaks, E.H. & Srivastava R.M. (1989). Applied Geostatistics . (Applied

Geostatistics Series). Oxford University Press: New York

Pyrcz, M.J. and Deutsch, C.V. (2014). Geostatistical Reservoir Modeling. 2nd
ed. Oxford University Press: New York.

22 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 3 – Variogram
Modelling
Duration: XXX minutes

This lesson introduces variogram terminology. You’ll learn to fit a variogram

model to an experimental variogram, which you can then use to measure
important parameters like nugget, range, and partial sill.

After this lesson you’ll be able to:

 Understand the terms nugget effect, range, sill, and partial sill
 Set nugget, range, and partial sill values in Micromine
 Fit a variogram model using more than one component.

Introduction
Once you’ve displayed an experimental variogram you model it by overlaying
a theoretical variogram model onto the experimental data. Fitting a theo-
retical variogram is similar to the statistical decomposition described in
MG 205 – Statistical Essentials , where you drew a histogram (the experi-
mental data) and overlaid a statistical model (the theoretical model).

You’ll incorporate the techniques from this and the preceding lesson in the
variography workflow introduced in Lesson 5.

Describing a Variogram Model

To define a variogram model you need to consider the following properties,
which are illustrated in Figure 3.1:

 Nugget (C0): In theory the semi-variance at the origin (lag = zero)

should be zero. If it’s different from zero at lags very close to zero, then
that difference is referred to as the nugget effect. It represents
variability at distances smaller than the typical sample spacing, including
measurement error.
The name comes from the influence a high-grade (e.g. gold) nugget
might have on a sample: changing the sample location by just a few
centimetres (zero at the scale of most mineral deposits) could have a
major effect on the sample value.
The nugget effect occurs near the Y-intercept of the variogram graph. It
is most easily seen on a downhole variogram because downhole data
is more closely spaced than between-hole data. To model the nugget
effect, extrapolate the first few experimental variogram values down to
the Y-axis. In Micromine this is an interactive process.

© Copyright MICROMINE 2014 23

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:

Figure 3.1: Features of a one-component variogram

 Number and type of structures/components: The number of

structures depends on the study area, but you should always identify
the fewest possible. Look for distinct shoulders or changes in the slope
of the experimental variogram, which may suggest the number of
structures in the data. You then model each structure using a single
component.
Components in a variogram model are usually labelled Cn, where n is the
component number (Figure 3.1). The nugget is Component zero ( C0).

 For each component, the range (a) and partial sill (Cn): This is also
an interactive process in Micromine.
Range: The distance at which the variogram reaches the sill value. Pre-
sumably, sample pairs are no longer correlated beyond this distance.
Sill: The semi-variance (height) at which the variogram levels off. The
total sill (height) of a variogram model is simply the sum of its
components’ partial sills ( C0 + C1 + … + Cn).
Some component types require different parameters such as slope or
power. Component types are discussed in the next section.

Component Types
An individual variogram component may be defined using one of several dif-
ferent types, some of which are shown on Figure 3.2. There is rarely any
physical ground for choosing a particular type (Goovaerts, 1997, p.101) , and
for the most part the best choice is the one that provides the closest fit to
the experimental data using the fewest components. The types are grouped
into two main categories: those that eventually level off (reach a sill) and
those that do not.

24 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

The properties and typical uses of frequently used component types are: Notes:

 Spherical (Sph in Figure 3.2): Linear from the origin and then quickly
changes slope as it reaches the sill. Used for grade/quality estimates.
 Exponential (Exp): More curved than the spherical model, showing a
much more gradual change in slope towards the sill. Used for
grade/quality estimates.
 Gaussian (Gaus): Parabolic near the origin and then gradually changes
slope as it reaches the sill. Use with caution; best for highly continuous
phenomena like air- or water-borne contamination.
 Linear ( Lin) and General Linear (Gen Lin): These models do not reach
a sill. General linear combines power and slope parameters. Used for
surfaces such as topographic elevation or coal seam thickness. May also
be used as a generic variogram model.

Figure 3.2: Theoretical variogram model types

Although there are no specific rules for choosing a model type, a spherical
or exponential model is typically best for most hard-rock grade or coal
quality modelling. In other modelling situations, such as coal seam modelling,
a linear or general linear model may be better.

Fitting a Variogram Model

To fit a variogram model, begin by displaying the experimental data. Next,
click the Chart Control Pane button on the Chart toolbar to enable the
Chart Controls. Micromine will display a default linear model using esti-
mated parameters. Change the component Type as needed and then adjust
its parameters by dragging the handles within the chart window. Refine an
individual parameter by clicking into its edit box in the Chart Controls and
then rolling the mouse wheel. Remove decimals for faster mouse-wheel
adjustments or add decimals for finer control.

© Copyright MICROMINE 2014 25

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Don’t over-fit a variogram model; you should aim to represent the major spa-
tial characteristics of the study area rather than obtain an exact fit to the
experimental data. The shape of the model is most important, not the nu-
meric values of the input parameters. Equally important is the number of
pairs contributing to each lag, which is shown symbolically or numerically,
depending on the chosen display options. Toggle the Scale and Pairs op-
tions as often as necessary while fitting a [Link] a Variogram with
More Than One Structure

An experimental variogram with distinct shoulders or changes in slope usually

suggests the existence of more than one nested structure. Nested struc-
tures frequently arise in geology because the continuity of ore grades varies
at different measurement scales, and a typical grade variogram might include
both short-scale and long-range trends.

You model nested structures by adding a component for each structure,

progressively adding, fitting and refining each one in order of increasing dis-
tance until the model fits the experimental data. Add as few components as
possible. Remembering to avoid over-fitting, it is more important to fit the
model to short-scale structures than long-range trends (e.g. Pyrcz &
Deutsch, 2014, pp. 100-101).

Although it is best to determine the number of structures by eye beforehand,

they can sometimes be hard to identify. An alternative is to see how well you
can fit just one component, and only add another if you cannot obtain a
good fit. Try changing the initial component’s type; for example a model
requiring two spherical components might be more simply constructed using
one exponential component instead. As you gain experience you will find it
easier to choose the best approach for your data.

The fastest way to add a component is by right-clicking the chart near the
new component’s estimated range and choosing Add Component here
from the pop-up menu. Set the new component’s type and then adjust the
model to suit the experimental data. Lastly, readjust any shorter-scale com-
ponents back onto the experimental data.

A long-range component still influences the variogram at short lag

distances. Because of this effect it is necessary to adjust any existing
short-scale components whenever you add another component.

The height of a theoretical model containing more than one component is

simply the sum of its components’ partial sills. Micromine automatically calcu-
lates the partial sill values whenever you adjust a model by dragging its han-
dles. Adjustments in the Chart Controls are applied directly to the partial sills.

With this understanding of how to set up a multi-component model you can

turn your attention to modelling the downhole variogram from Exercise 2.2.

26 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Exercise 3.1: Model a downhole variogram Notes:

The experimental variogram from the previous lesson has a non-zero Y-

intercept, meaning it exhibits a nugget effect, and two distinct slope regions,
suggesting it includes two nested structures:

 A short-scale structure gradually curving from the origin to a height of

about 0.7 at a lag distance of about 9 m.
 A long-range structure staying almost linear until it reaches a total sill
of about 0.77 at a lag distance of about 20 m.

Thus the experimental data will probably require a model with two compo-
nents: exponential (for the curved short-scale structure) and spherical
(for the more linear long-range structure). Optionally compare your exper-
imental variogram with the model shapes in Figure 3.2 to confirm that these
model types are most appropriate for the data.

Armed with this information you can now fit the model:

1. Click the Chart Control Pane button on the Chart toolbar to display
the Chart Controls.

Micromine will add a default linear model using estimated values.

2. In the Chart Controls, change Component 1’s Type from LINEAR to

EXPONENTIAL.
3. Drag the Nugget and Component 1 handles so that the model fits the
first (short-scale) structure. Ignore the long-range structure for now.
Your model should only reach a semi-variance (height) of about 0.70.
4. Right-click the chart at a lag distance of about 20 m and choose Add
Component here from the pop-up menu.
5. Set Component 2’s Type to SPHERICAL and then drag the
Component 2 handle onto the final sill of the experimental data. It
should be located close to (20, 0.77).

Move your mouse over a handle to read its coordinates in the status bar.

© Copyright MICROMINE 2014 27

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 6. Drag the Component 1 handle back onto the experimental data.
7. Adjust the Nugget , Component 1 and Component 2 handles until you
achieve a good fit with the experimental data.
8. For fine control, click into a Range or Partial Sill box in the Chart
Controls and adjust the value by rolling the mouse wheel.
9. Inspect the value of the Goodness of Fit (Noel Cressie Statistic) at
the bottom of the Properties window as you make your adjustments. A
lower value indicates a better fit between the model and data:

The Nugget should be about 0.22. Component 1 should have a range of

about 6.7 and partial sill of about 0.39, and Component 2’s range and par-
tial sill should be around 19.0 and 0.17.

10. Note down your nugget value here. You’ll need it later:

Nugget ( C0):

11. Click the Forms button in the Chart Controls and save your model as a
form set with the Title Au (cut) downhole.
12. Close the Semi Variograms window once you are done.

28 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Alternative Method for Identifying Structures Notes:

If you have trouble identifying the structures in advance you can try this
more experimental workflow:

1. Create and fit a model with one component that spans the entire
experimental variogram.
2. Try changing the Type to see which one provides the best overall fit.
3. If you cannot get a good fit, move Component 1 ’s handle down to the
first obvious shoulder or change of slope in the experimental data
(defining the short-scale structure) and fit the model to that structure.

This is now the first component. Ignore any other structures for now.

4. Right-click the chart near the final range of the experimental data and
choose Add Component here from the pop-up menu.
5. Fit Component 2 onto the experimental data.
6. If you still cannot get a good fit, repeat Steps 3 through 5, adding and
refining a single component each time.

At this point the only important value is the nugget, which is displayed near
the top of the Chart Control Pane. The ranges may also help you to under-
stand the continuity of your data, although they are not essential right now.
Don’t worry if your values don’t match those of other trainees – the visual fit
of the graph is all that matters.

Validating the Model

In the last exercise you used the Goodness of Fit (Noel Cressie Sta-
tistic) to evaluate the goodness of fit between your model and the raw data.
This statistic is a weighted least-squares measure that calculates the sum
of the squared differences between the experimental data and theoretical
model. It gives more weight to the first few lags in the variogram, which is
where the model will have the strongest influence on the kriging calculation.

Micromine always displays the Noel Cressie Statistic in the Properties win-
dow whenever a variogram model is visible. In theory a smaller value indi-
cates a better model. However, the best guide is the visual fit between the
model and the major characteristics of the experimental data.

Why Fit a Theoretical Variogram?

In Lesson 6 of MG 301 – Block Modelling you hand-calculated the value of an
unknown point using Inverse Distance Weighting (IDW). After measuring,
squaring, and taking the inverse of the distance to each input data point, you
used these values to determine the allocation of weights for the input data.
The bulk of the work was in allocating the weights, and the final estimate for
the unknown point became an ordinary weighted average once the weights
were known.

© Copyright MICROMINE 2014 29

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Kriging uses variogram models to allocate the sample weights. This offers a
major advantage over IDW: the weights are not derived from an arbitrary
inverse distance relationship but are based on your understanding of grade
continuity. In effect the variogram models act as lookup tables: the kriging
interpolator measures the distance to an input data point, compares the dis -
tance with the appropriate variogram model, and looks up the amount of
weight to assign that data point. Once the weights are assigned there is little
difference between IDW and kriging: ultimately they both calculate a
weighted average.

However, you can’t simply use the experimental variogram to calculate the
sample weights. The experimental data is mathematically too complex and
may be noisy or contain outlying values that would give an incorrect result if
used for kriging. Instead, the theoretical model smooths out any irregularities
and provides a well-defined mathematical function from which the kriging
interpolator can determine its weights.

Interpreting a Fitted Theoretical Model

A fitted theoretical variogram model may be interpreted in a variety of ways
based on the arrangement of the nugget, ranges and sills:

The relative nugget is the sill (height) of the nugget relative to the total sill
of the model, expressed as a percentage. It indicates the relative proportions
of spatial and random variation within the study area. For example, if the
height of the nugget was around one-third of the total sill you would say the
relative nugget is about 30%. You would then infer that around 70% of the
variation within the study area is spatial, with the remaining 30% of variation
being random (or at a scale smaller than you can measure).

You can use the range of a theoretical variogram to help determine the size
of the search neighbourhood. However, this must be determined in conjunc-
tion with the relative nugget, since the influence of samples beyond the
range increases in proportion to the relative nugget. When the relative nug-
get reaches 100% all samples within the neighbourhood contribute equally,
no matter what their distance, and under these circumstances it is important
to enlarge the search neighbourhood in order to achieve the most reliable
estimate (e.g. Vann et al, 2003).

Although the partial sills of the individual components have no geological

meaning, there may still be a geological explanation for the presence of any
nested structures.

30 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 3 Summary Notes:

This lesson introduced variogram modelling. The topics covered so far are:

To display a downhole experimental variogram:

Select Stats | Semi Variograms, and

Set the Mode to Calculate, the Direction to Downhole, and the Type
to Variogram, and
Set up the Raw Data and Data Values options, then
Click the Downhole Semi Variograms button, and
Set the Lag Interval to the dominant sample length, and
Set the number of Intervals to a value that will span one half-of the
study area at the chosen lag interval.

To create a variogram model:

Click the Chart Control Pane button on the Chart toolbar, and
Fit the model to the data by dragging the handles in the chart.

To fit a multi-component variogram model:

Visually identify the candidate structures in the experimental data, then

Create a model with a nugget and one component, and
Drag the Nugget and Component 1 handles so that the model fits the
first (short-scale) structure, ignoring any other structures, then
Right-click near the range of the next structure and choose Add
Component here from the pop-up menu, and
Drag the Component 2 handle so that the model fits the second
structure, and
Adjust the Component 1 handle back onto the first structure, then
Repeat for each additional structure.

To validate a variogram model while you’re fitting it:

Inspect the Goodness of Fit (Noel Cressie Statistic) in the

Properties window, then
Adjust the model, and
Continue until you reach the lowest Noel Cressie Statistic for your
chosen visual fit.

© Copyright MICROMINE 2014 31

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 3 Summary

Good Practice

 Try to avoid drilling on a regular square pattern. A regular pattern may

make the nugget effect hard to determine if the variogram range is
smaller than the drillhole spacing.
 Use downhole (or vertical) variograms to estimate the nugget effect.
 When fitting a variogram with nested structures, add one component at
a time, modelling the observed structures in increasing order of range.
 An alternative is to try modelling the entire experimental variogram
using a nugget and just one component, and only add components if
you cannot obtain a good fit.
 Even if you think IDW is adequate for grade interpolation the
orientations and ranges of the variogram axes may provide useful
information about the study area that you might otherwise have missed.

Help Topics

For information on: See:

Semi-variograms Statistics > Semi Variograms

Semi-variogram Statistics > Semi-variograms >

modelling New workflow (page topic)

Semi-variogram Statistics > Semi Variograms >

model types Semi Variogram Model Parameters >
type of model (link on page)

Additional Resources and Further Reading

Goovaerts, P. (1997). Geostatistics for Natural Resource Evaluation. (Applied

Geostatistics Series). Oxford University Press: New York

Pyrcz, M.J. and Deutsch, C.V. (2014). Geostatistical Reservoir Modeling. 2nd
ed. Oxford University Press: New York.

32 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 4 – The Variogram
Map
Duration: XXX minutes

This lesson introduces the variogram map along with the concept of anisot-
ropy. You’ll learn to set up a variogram map and use it to interactively iden-
tify and measure anisotropy.

After this lesson you’ll be able to:

 Understand the concept of anisotropy and the difference between

geometric and zonal anisotropy
 Set up a sequence of variogram maps that quantify the 3D anisotropy of
the study area
 Create a variogram control file for use elsewhere in Micromine.

Introduction
A mineral deposit with strongly directional grade continuity requires an un-
derstanding of the direction and strength of that continuity before it can be
adequately modelled. For example you would expect grades in a narrow vein
gold deposit to continue over a longer distance along the strike of the vein
than across its thickness. Such a deposit could be described as strongly ani-
sotropic and the modelling methodology would need to be adapted to ac-
commodate that anisotropy.

On the other hand, if grades within a massive porphyry copper deposit vary
equally in all directions, that deposit would be described as isotropic and
would require a different modelling approach.

You’ll incorporate the variogram map in the more general modelling workflow
introduced in Lesson 5.

Anisotropy
A definition of anisotropy is to say the properties of something vary with
direction, and in geostatistics something is anisotropic if its spatial conti-
nuity is the property that varies. One way to measure anisotropy is to dis-
play experimental variograms in three perpendicular directions, oriented to
suit the directions evident in the mineralisation. You then quantify the ani-
sotropy ratio of the study area by comparing the ranges (and possibly the
sills) of the fitted models. For instance, if the along-strike range of the nar-
row vein gold deposit was 30 m and the across-thickness range was 5 m you
would describe the anisotropy ratio as being six-to-one (30 ÷ 5).

© Copyright MICROMINE 2014 33

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: There are two types of anisotropy, illustrated using variograms on Figure 4.1.

 Geometric: The nugget and sills are the same but the ranges are
different for the three directions.
 Zonal: In pure zonal anisotropy the ranges are the same but the sills
are different in the three directions. In practice a degree of geometric
anisotropy is usually present and the ranges will also vary.

(a)

(b)

Figure 4.1: Geometric (a) and zonal (b) anisotropy

Micromine uses geometric anisotropy in its variogram models. If zonal ani-

sotropy is present an additional component must be used, with a very long
range (such as 10 000 metres) to bring the sills to the same level. However,
zonal anisotropy is usually a sign of mixed populations in the direction with
the highest sill and is best handled by re-domaining the study area.

When interpreting the variograms, the direction in which the grades are
most continuous is the one with a variogram that rises the most slowly,
maintaining the lowest values over the longest range as shown in red on
Figure 4.1(a). Similarly, the direction in which grades are least continuous is
the one that rises the most quickly, with the shortest range as shown in blue.

This lesson describes the steps for quantifying anisotropy in three dimen-
sions. To work with two-dimensional data (such as the elevation or thickness
of a coal seam) simply leave the Z field blank. Micromine will automatically
adjust the available options to suit.

34 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Measuring Anisotropy
Notes:

Measuring anisotropy requires you to know both the direction and distance of
grade continuity. Although these properties may already be known in a ma-
ture mineral deposit, they must be found experimentally when the geology is
not well understood. You do so by assessing the grade continuity in different
directions until you find the directions that represent the greatest, least, and
intermediate continuity distances within your study area.

Always rely on your knowledge of the geology when searching for anisot-
ropy. If you know the mineralisation has a preferred orientation, use this
knowledge to build an anisotropic model even if there is insufficient data to
show it in the variograms. The opposite is also true: what looks like strong
anisotropy in limited data, which can’t be proven by fieldwork, may be better
ignored. You will see an example of this in the iron data.

Although it is possible to find anisotropy using a variogram fan (a sequence

of directional variograms at regular directional increments spanning a 180°
arc), it is easier to visualise on a variogram map, which is a continuous 2D
surface that represents the semi-variance at different lag distances and di-
rections. Once the presence or absence of anisotropy is established on the
variogram map it can be analysed more rigorously on the corresponding
variograms. Appendix 3 describes the traditional variogram fan approach.

Each cell on a variogram map represents the semi-variance of all sample

pairs separated by the distances and directions of its boundaries, as defined
by their polar (angle and distance) coordinates (Figure 4.2). Only the input
pairs separated by those distances and directions (plus a tolerance, described
later in this topic) are used to calculate its value. The origin of the map is at
its centre and represents zero separation between sample pairs.

Figure 4.2: Values falling within each cell

Like any variogram, low values indicate stronger correlation and better con-
tinuity between sample pairs. Interpreting the map is simply a case of locat -
ing the zone of low values running through its centre, paying most atten-
tion to the region nearest the origin.

© Copyright MICROMINE 2014 35

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: If geometric anisotropy is present the zone will be roughly elliptical and its
long axis will represent the direction of greatest continuity. A narrower zone
indicates stronger anisotropy and a wider zone weaker anisotropy. Zonal
anisotropy is marked by a parallel-sided zone. If the zone is circular or hard
to see then anisotropy is not present. Figure 4.3 uses theoretical variogram
maps to illustrate the two forms of anisotropy and highlight the direction of
maximum continuity. The trends are much less distinct with real-world data.

Figure 4.3: Anisotropy and the direction of greatest continuity

Complex geology may produce multiple zones of low values with different
orientations, and in this situation you may need to test more than one can-
didate to determine the most representative direction.

Choosing Initial Parameters

You create a variogram map in Micromine via Stats | Semi Variogram
Map on the main menu, and the following topics introduce the parameters
and modelling steps needed to find and measure anisotropy.

The Input Data tab of the Semivariogram Map dialog is divided into two
parts. You supply the File and field names, along with the Transformation
and Display options in the upper part of the dialog. (Transformations are
described in Lesson 6.) The Display Options require you to choose a
palette whose colours will be assigned to the semi-variance values shown in
the map. Simply double-click in the Palette box and choose one with a
smooth gradation of colours. The Range and Percentile buttons control
how colours are allocated to the numerical values. In most cases Percentile
mode produces the best result.

You configure the initial appearance of the map via the Processing Options
in the lower part of the dialog. You only need to complete the first row of the
Processing Options grid; the other parameters will be created as you interact
with the map display.

36 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Directions and intervals Notes:

The number of directions controls the size and number of sectors around
the circumference of the map. Likewise, the lag interval controls the radial
size of the cells and the map’s overall radius. For example:

 72 directions produce a map with 72 cells around its circumference,

each of which is 5° wide (360 ÷ 72)
 A lag interval of 15 with 8 lags produces a map with a radius of 8½
cells, each of which is 15 units long, yielding a real-world radius of 127.5
units (15 × 8.5). The half interval is situated at the origin of the map.

The amount of available data determines the number of directions, with

large datasets generally able to support more directions than small ones.
Similarly the sample spacing determines the lag interval, which tends to be
some multiple of the dominant spacing. (You’ll learn another way to estimate
the lag interval in the next lesson.) Lastly, the physical size of your study
area determines the number of lags, remembering that a variogram needs
to span one-half of the study area at the chosen lag interval. All of these
numbers may be refined in the interactive map display.

Overlaps (tolerances)

Directional and lag tolerances are defined by the amount of overlap with the
surrounding cells, avoiding the ambiguity of the term ‘tolerance’ (which can
mean including or in addition to the current cell). Directional overlaps are
specified by the number of adjoining sectors and lag tolerances by the
number of adjoining lags used to estimate each cell (Figure 4.4).

Figure 4.4: One-cell direction and lag overlaps

include all cells in a 3 × 3 group

The overlaps can be any real number, not just integers. Each cell is then
estimated using data falling within the overlapping regions:

 An overlap of 1 includes the current cell plus one cell on each side, so
that data falling within all three cells contributes to the current cell.
If this overlap is applied both direction and lag, a total of nine cells
contribute to the current cell.

© Copyright MICROMINE 2014 37

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: The directional overlap may be further modified by a bandwidth, which

determines the maximum width of the search region in coordinate units and
prevents it from becoming too wide at long lag distances:

 A bandwidth of 25 limits the search window to a maximum width of 50

units (25 units on either side of the current direction).

You use the overlaps to adjust the quality of the map by smoothing out er-
ratic values or noise in the display. This noise may be considered in two
dimensions: around the circumference of the map and along its radius. In-
crease the direction overlap when you see noise around individual rings of
cells in the map, and increase the lag overlap when you see noise along
radial lines of cells.

Don’t over-smooth a noisy map. Your ability to identify anisotropy decreases

as the overlaps increase and you should try to strike a balance between de-
tail and noise. Your goal is to see a smooth zone of low values running
through the origin (centre) of the map.

You’ll create a variogram map in the next exercise.

Exercise 4.1: Display a variogram map

In this exercise you’ll display an initial strike map using cut gold grades,
starting by displaying the corresponding data in Vizex:

1. Switch back to Vizex, change to a Plan view, and use the Measure
Tool to determine the horizontal trend of the grades (if any).

The grades don’t really show an obvious trend so you’ll need to investigate
this map with an open mind. A good initial number of directions is 72,
producing 5° angular sectors. You’ll use an initial lag interval of 15 m,
matching the average line spacing. Lastly, the diagonal dimension of the
mineralised zone is a little over 500 m, suggesting that the number of lags
should be around 15 (roughly 500 ÷ 2 ÷ 15). You’ll refine these parameters
once the map is displayed.

To display the variogram map:

2. Select Stats | Semi Variogram Map from the main menu.

3. Click the browse button in the File response and choose
Gold_Assay from the list.
4. Click the list ( ) button in the Semi variogram field response and
choose Au1_ppm_CUT.
5. Enable the Ignore blanks Numeric Exception.
6. Set the Transformation to NATURAL LOG.

The variogram map has no equivalent of the pair-wise relative variogram you
used in Lesson 2. However, the differences between relative and natural log
variograms are minor and can be safely ignored. The visual appearance of
the map is what matters, not the values of its individual cells.

38 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

7. Double-click the Palette box, choose a palette for the map colours, set Notes:
the mode to Percentile, and enable Spectrum mode:

8. Enter the following parameters on the first row of the Processing

Options grid. Ignore the rest of the form:

Prompt Setting
Directions > Number 72
Directions > Overlap [blank]
Directions > Bandwidth [blank]
Lag > Interval 15
Lag > Number 15
Lag > Overlap [blank]
Title Strike [default]

9. Click OK to display the initial map, which should look like this:

The map is colourful but not at all informative, although there do seem to be
more low (greenish blue) cells near its centre. Clearly it needs more work.

Once an initial map is displayed you then refine its parameters via the Chart
Control Pane by adjusting the interactive sliders or typing a value and

© Copyright MICROMINE 2014 39

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: pressing Enter. Micromine caches the data so that it can instantly respond to
your adjustments; when you are finished simply click the Calculate True
button to draw the map directly from the raw data.

You’ll refine the display in the next exercise.

Exercise 4.2: Refine the variogram map

To refine the map:

1. Adjust the Lag Interval and the two Overlap sliders to refine the map
and make the dominant trend as clear as possible.

You should see a distinct zone of low values with a north-westerly strike of
about 320°, matching the geological description. Try a lag interval of 18, a
lag overlap of 2 and a direction overlap of 3 if you can’t see it.

Where possible a variogram map should highlight data near the origin and
show data equally in all directions. Although your map is factual there is a
little too much noise at its outer edge, suggesting that you should reduce the
number of intervals.

2. Reduce the Number of Intervals to 9 and press Enter to eliminate the

surrounding noise and emphasise the map origin.
3. Click the Calculate True button to finalise the map. Your display should
look like this screenshot:

4. Leave the map open for the next exercise.

40 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Measuring Directions Notes:

The variogram map uses a three-stage workflow to identify the strike and
dip of the plane containing the mineralisation, and then measure the pitch
of the dominant axis in that plane (the direction of greatest grade continui-
ty). Once that is done, all of the other orientations (rotations, strike/dip, azi-
muth/plunge, etc.) may be derived as needed.

You display each map (strike, dip and pitch) in turn, starting with the strike.
Use the Select Direction button to identify the direction of continuity,
paying most attention to the region near the centre of the map. Click the
Next button to proceed to the next stage. Refine the new map if neces-
sary, and then reuse the Select Direction and Next buttons.

Strike is always right-handed and Micromine will automatically reverse

the strike direction if you measure a left-handed dip.

The Align Vizex View button adjusts the first Vizex window to the same
orientation as the map, providing a visual validation and making the data
easier to interpret. Tic marks around the map provide a directional reference,
displaying strike, dip or pitch values for the relevant map. Dip and pitch
tics are only shown in the lower hemisphere, and these maps include labels
showing the left- and right-handed strikes. The status bar always displays
the lag distance and azimuth/plunge of the mouse pointer.

The workflow is finished once you measure a direction in the pitch map, and
the Next button changes to indicate that it is the last step. The line you
measure here represents the 3D direction of greatest continuity, known as
the main axis or Axis 1. From here Axis 2 is positioned at right-angles to
Axis 1 whilst staying within the orebody plane and Axis 3 is at right-angles
to Axes 1 and 2.

You’ll complete this workflow in the next exercise.

Exercise 4.3: Measure the 3D direction of greatest

continuity

Before measuring the direction of greatest continuity on the strike, dip and
pitch maps you need some idea its real-world orientation, which you can
obtain by inspecting the data in Vizex. To do so:

1. Switch to the Vizex tab and set the display to Plan View .
2. Zoom out until the high-grade points start merging together, which will
highlight the grade continuity.
3. Using the Measure Tool , measure the trend of the samples.

With a little imagination you should be able to see a weak north-westerly

trend in the most densely drilled portion of the project area.

4. Switch back to the Semivariogram Map tab.

© Copyright MICROMINE 2014 41

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 5. Click the Select Direction button and move your mouse until the
direction pointer is centred on the average orientation of the low values
running through the map origin.
6. Click to define the direction, and then inspect its values in the
Properties window.

You should obtain a Direction of around 325°, although the zone is broad
and you might pick a direction anywhere between 315° and 340°.

7. Click the Next button to display the dip variogram map.

The dip map

The dip map is aligned vertically at right-angles to the direction you meas-
ured in the strike map, and you are now looking along the direction of strike.
Aligning Vizex to this orientation will make the map easier to interpret:

1. Click the Align Vizex View button and inspect the data.

There is no obvious trend because the data includes both the supergene and
primary mineralisation. However, grades in the most densely-drilled region
may suggest a gentle dip to the south-west. You should also keep in mind
the dominant drilling orientation when you inspect the map:

2. Switch back to the Semivariogram Map and adjust the sliders to refine it.
3. Click the Calculate True button to finalise the map, and then inspect
the result. Try a lag interval of 12, a lag overlap of 3 and a direction
overlap of 2 if you can’t get a good map.

Your display should look like the screenshot on the facing page. The
strongest trend is a gentle dip towards the south-west (1), matching the
Vizex view and presumably dominated by the supergene zone. But the map
also reveals a weaker moderate dip to the south-west (2), possibly repre-
senting the quartz veins and shear zones; a weak sub-vertical dip (3); and a
weak moderate dip to the north-east (4), dominated by the drilling direction.

42 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:

Dealing with different but equally valid directions is often necessary when the
geology is complex, and in the workplace you would evaluate them all or
consider dividing the data into separate domains. For this course you’ll treat
the data as a single domain and use the gentle south-westerly dip (1).

4. Click the Select Direction button and move your mouse until the
direction pointer is centred on the south-west-dipping zone of low
values.
5. Click to define the direction, and then inspect its values in the
Properties window.

You should obtain a Direction of around 10° towards the south-west.

6. Click the Next button to display the pitch variogram map

The pitch map

The pitch map is aligned to the plane defined by your strike and dip meas-
urements. You are now looking at an oblique an inclined plane, and aligning
the Vizex view is vital for understanding the data in this orientation:

1. Click the Align Vizex View button to update Vizex.

You should see a trend in the high grades running roughly “north-west” with
respect to the screen.

2. Switch back to the Semivariogram Map and refine it by adjusting the

sliders. Click Calculate True once you are done. Your map should look
like the screenshot overleaf.

© Copyright MICROMINE 2014 43

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:

In this map a horizontal line drawn between the two 0° tics corresponds to
the strike of the orebody plane. (The left- and right-handed strikes are
shown in brackets.) Similarly, a vertical line drawn through the 90° tic cor-
responds to its dip and dip direction.

The continuity direction changes at very short lag distances, but short dis-
tances are dominated by the drilling direction and should be ignored in
favour of the more obvious trend.

3. Click the Select Direction button and move your mouse until the
direction pointer is centred on the zone of low values.
4. Click to define the direction, and then inspect its values in the
Properties window.

You should obtain a Direction of around 20° towards the south-east.

5. Lastly, click the Next button to finalise the variogram map.

The Next button only appears as a tick when all necessary directions have
been measured.

6. Leave the display open for the next exercise.

Passing Map Directions to Other Functions

Now that you have created all three variogram maps you have fully defined
the 3D orientation of the main axis or Axis 1. Along with the orebody plane
this is all that Micromine needs to calculate the direction of Axis 2 (the in-
termediate axis) and Axis 3 (the minor axis), which are always at right-
angles to each other. To help visualise this result, the perspective view in
Figure 4.5 shows the pitch map with the axes in their true 3D orientations.

44 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:

Figure 4.5: Pitch variogram map and resulting axes

in their 3D orientations (approx. axis plunges and azimuths shown)

This information may then be saved as a variogram control file to be used

in the search ellipsoid and variogram chart, assisting with the creation and
modelling of 3D anisotropy. You’ll create a control file in the next exercise.

Exercise 4.4: Create a variogram control file

To create a variogram control file:

1. Click the Create Variogram Control File button on the Chart

toolbar.
2. Set the Variogram control file to AuCut, and click OK to create it.
3. Click the Create Variogram Control File button a second time and
then right-click the Variogram control file.
4. Inspect the file contents, and then close the Variogram Control File
dialog and Semivariogram Map window once you are done.

This file lists the orientations of Axes 1 through 3 using different conventions
(rotation, azimuth/plunge, and strike/dip/pitch), making it easy to relate
them to your knowledge of the real-world geology. In the next lesson you’ll
use the variogram control file to set up a three-axis model for the gold data.

© Copyright MICROMINE 2014 45

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Interpreting the Variogram Map
Using the variogram map is conceptually easy: you configure the best display
from the source data and use that to identify the directions of greatest con-
tinuity in three dimensions. But interpreting the display isn’t always so
straightforward. Here are some tips for understanding you see.

What appears to be strong continuity sometimes isn’t

Have you ever driven past a vineyard and noticed how the posts seem to line
up at certain angles? The same thing can happen with drillholes situated on a
very regular drilling grid, and if these chance alignments coincide with a map
direction they may create a false indication of strong continuity.

Use your knowledge of the project area to validate any candidate directions,
and be prepared to ignore any directions that are the result of chance drill-
hole alignments.

Near-vertical continuity doesn’t imply a vertical orebody

Sometimes the dip or pitch map shows a very steep zone of strong continu-
ity. This doesn’t imply a vertical orebody: it’s another effect of the data.

The dip and pitch maps are usually drawn in a vertical or near-vertical orien-
tation, which means the downhole sample spacing typically dominates their
vertical axes, whereas the between-hole spacing dominates their horizontal
axes. In a typical mineral deposit the downhole spacing is much smaller than
the between-hole spacing, which means there is more downhole data than
between-hole data. This imbalance may produce a zone of false continuity
that runs roughly parallel to the dominant drilling direction.

Use your knowledge of the project area to validate the directions, and ignore
any that don’t make geological sense.

Align Vizex to the map orientation to understand the map

The best way to understand the map is by loading the source data in Vizex
and clicking the Align Vizex View button. Aligning Vizex to the map
provides a visual reality check that helps you to understand your data. It may
also reveal the cause of any false continuity from chance alignments or an
overabundance of data in any single direction.

Calculate True before measuring a direction

The interactive map simplifies its calculations to provide real-time updates as

you refine its processing parameters; clicking Calculate True forces it to
recalculate itself using the full dataset. The differences between the inter-
active and true maps are generally minor, but you must be looking at the
real data before measuring a direction. This is vital if you make any inter-
active bandwidth adjustments.

46 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 4 Summary Notes:

This lesson introduced the variogram map along with the concept of ani-
sotropy. The topics covered so far are:

To understand and search for anisotropy in your study area:

Something is anisotropic if its properties vary with direction, and

In geostatistics spatial continuity is the property that varies.
Always rely on your knowledge of the geology:
Build an anisotropic model for mineralisation that has a proven preferred
orientation, even if there is insufficient data, or
Consider ignoring apparent but unproven anisotropy in limited data.

To choose initial variogram map parameters:

Choose a number of directions based on the amount of available

data. Large datasets generally support more directions than small ones.
Optionally use a bandwidth to prevent inappropriate sample pairings at
long lag distances.
Choose a lag interval that matches a multiple of the sample spacing.
Choose enough lags to span one-half of the study area at the chosen
lag interval.

To display a variogram map:

Select Stats | Semi Variogram Map, then

Choose a file and semi variogram field, and
Choose a palette, then
Enter the parameters on the first row of the processing options grid.

To interactively adjust a variogram map:

Use the sliders to interactively adjust the map, or

Type a number into a box and press Enter to apply the change.
Use overlaps to smooth out erratic values or noise in the display:
Increase the direction overlap when rings of cells are noisy, or
Increase the lag overlap when radial lines of cells are noisy.
Click the Calculate True button to update the map display.

© Copyright MICROMINE 2014 47

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 4 Summary

To measure the 3D orientation of grade continuity:

Click the Align Vizex View button to align Vizex to the map plane,
and
Look for the map direction with values that rise the most slowly,
maintaining the lowest values over the longest distance, and
Ignore any directions dominated by drilling directions, then
Click the Select Direction button and move the direction cursor
onto that direction, and
Click to define the direction, then
Click the Next button to display the next variogram map.

To create a variogram control file:

Click the Create Variogram Control File button, and

Enter the name of a variogram control file.

Good Practice

 Try to avoid drilling on a regular square pattern. Along with potentially

making the nugget effect hard to determine it may also create artificial
geometric alignments between holes, which may influence the perceived
anisotropy in the data.
 Always relate what you see in the map back to the real world by aligning
the Vizex display to the map orientation.

Help Topics

For information on: See:

Semi-variogram Map Statistics > Semi Variogram Map

Finding directions Statistics > Semi Variogram Map >

Viewing the Semi Variogram Map

48 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 5 – Modelling 3D
Variograms
Duration: XXX minutes

This lesson introduces three-directional variograms and the variogram model-

ling workflow. Armed with your knowledge from the preceding lessons you'll
use the variogram map and variogram chart together to model a set of
three-directional variogram axes. You’ll use them for cross-validation and
kriging later in this course.

After this lesson you’ll be able to:

 Combine the variogram map and variogram chart into one workflow
 Use that workflow to model the variograms needed to describe the
continuity and orientation of mineralisation within the study area
 Refine your initial estimates as you learn more about the data.

The Workflow
The preceding lessons introduced two essential tools for variogram model-
ling: the variogram chart and the variogram map. Together these tools
provide a framework for finding and quantifying the 3D directions and dis -
tances of grade continuity. Although they were examined in separate lessons,
in the workplace they are used together in the variogram modelling
workflow, which is illustrated in Figure 5.1 and is discussed in this lesson.

You must model variograms separately for each element and domain. For
each element/domain combination you may need up to six experimental
variograms, although only three are used for variogram modelling and sub-
sequent grade interpolation.

Clearly this represents a great deal of work in a complex study area. For ex-
ample, assuming a mineral deposit has three elements and five domains, you
may need to run 90 separate experimental variograms (3 × 5 × 6), of which
45 (3 × 5 × 3) will be modelled and used for grade interpolation. If multiple
indicator kriging (MIK) with ten grade bins is applied the number of final
variograms blows out to 450. As a result, variography can take anywhere
from a few hours to several weeks depending on the number of elements
and domains, and the chosen modelling method.

Finding Workflow Shortcuts

The main directions of anisotropy may be obvious if the geology and mineral-
isation are well studied and interpreted, and you may be able to simplify or
skip some steps. Similarly, if you are updating an existing resource report
you may be able to reapply many of the original parameters. However, you
must explain all assumptions and justify all decisions.

© Copyright MICROMINE 2014 49

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:

Figure 5.1: The variogram modelling workflow

The variograms illustrated in Figure 5.1 must be run for each element/ do-
main combination. In more detail they are:

 Use a downhole variogram to model the nugget effect, which is as-

sumed to be isotropic (Goovaerts, 1996, p.102) and is modelled once
per element/domain.
 Optionally, use a sequence of omnidirectional variograms (omni-
variograms) with different lag intervals to evaluate the general ranges
and sills of the grade populations.
Good omni-variograms may predict a successful outcome and suggest
an initial lag interval for the variogram maps. Conversely, variography
may be difficult if you cannot obtain at least one good omni-variogram.

 Use a sequence of variogram maps to identify anisotropy and

measure the 3D orientation of the direction of greatest continuity.
Create a variogram control file for later use.
 Use the variogram control file together with individually adjusted lag
intervals to create a set of three directional variograms, and fit them
with theoretical models.

The following topics discuss the workflow steps and their outcomes.

50 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Downhole: The Nugget Effect

Notes:

You model the nugget effect by carrying out the steps in Lessons 2 and 3. In
most cases only one experimental downhole variogram is needed because
the downhole interval is typically constant or dominated by a preferred sam-
ple length. You then fit a theoretical model to the experimental data.

The outcome of this step is to quantify the nugget effect.

Omnidirectional: Evaluating Data

In this step you create a sequence of experimental omni-variograms using a
number of different lag intervals. (Because an omnidirectional variogram
searches equally in all directions the lag interval is the only relevant parame-
ter. The number of intervals should, of course, be enough to span half of
the study area at the shortest lag interval.)

The primary outcome of this step is to understand the between-hole

grade continuity of your data, which will become evident when you obtain
good omni-variograms. When the variograms are displayed together they
should all form about the same shape with varying levels of noise or detail. If
you cannot obtain good omni-variograms then the range of grade continuity
may be smaller than your drillhole spacing or your sampling may suffer from
assay quality control issues.

The choice of lag intervals depends on your data and should always reflect
the between-hole sample spacing. A good starting point is to use a min-
imum value that is one or two times the minimum between-hole spacing and
a maximum that is around twice the maximum spacing. To create the
variograms simply divide this range into as many increments as necessary.
You do this interactively on the Omnidirectional Semi Variograms dialog.

These variograms are used for evaluating your data and do not contribute to
the modelling process. Along with helping you understand the between-
hole grade continuity they may also suggest an initial lag interval to use
on the variogram map. You may choose to omit this step if your study area is
mature and its geostatistical properties are well understood.

You’ll create omni-variograms in the next exercise.

Optional Exercise 5.1: Create omni-variograms

From the data description you know that the drill line spacing ranges from
5 m to 100 m, averaging around 15 m, and the hole spacing varies from 4 m
to over 50 m, averaging about 25 m. Thus the range of lag intervals might
be from 5 m to around 100 m. Incrementing the intervals in steps of 5 will
ensure that variograms with lag intervals matching the average line and hole
spacing are created, while still producing enough experimental variograms
for a good evaluation.

© Copyright MICROMINE 2014 51

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Based on an average lag interval of 20 m, around 12 intervals (500 ÷ 2 ÷ 20)

are needed to span half of the study area. It's safe to use this number on the
longer intervals too; you can manually reduce the number of intervals for
these variograms later.

To create the variograms:

1. Select Stats | Semi Variograms from the main menu and switch to
the Input Data tab.
2. Set the Mode to Calculate and the Direction to Omnidirectional.
3. Change the Raw Data fields to EAST, NORTH, and RL.
4. For safety, skip the Search Definition and enter a new Semi Variogram
File name of SVG_AuCut_Omni.
5. Save a new form set with the Title AuCut Omni.

Now that you’ve entered a new filename and created the new form set you
are ready to set up the individual variograms. If you were to skip these steps
you might overwrite the previous file or form set.

6. Click the Omnidirectional Semi Variograms button in the Search

Definition.
7. Click the Clear button near the top right of the form.
8. Right-click in the empty Lag Interval cell and choose Calculate from
the pop-up menu. Fill out the Calculate dialog as shown here:

Prompt Setting
First value 5
Last value 100
Create using INTERVAL SIZE
Value 5

9. Click OK to apply the calculation, and inspect the values in the form.

Micromine added lag intervals ranging from 5 m to 100 m in steps of 5.

10. Enter 12 anywhere in the #Intervals column.

11. Right-click the newly entered number and choose Replicate from the
pop-up menu.
12. Enable any Line checkbox and then right-click | Replicate it.
13. Enable any Scale checkbox and right-click | Replicate it.
14. Right-click anywhere in the Colour column, avoiding the colour
swatches, and choose Fill With Random Colours.
15. Back on the main dialog, switch to the Chart tab and set the Title to
Omnidirectional Au (cut) .
16. Click OK to calculate and display the experimental omni-variograms.
Your display should resemble the screenshot on the facing page.

52 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:

17. Keep the chart open for the next exercise.

Quickly Filling Out the Semi Variograms Form

The last exercise introduced a number of shortcuts for quickly filling out the
Semi Variograms form. In summary they are:

 For numbers that increment, right-click in any cell and choose

Calculate from the pop-up menu. You can define the increments by
interval size or number of intervals.
 For numbers that stay constant, enter the value in any cell and then
right-click it and choose Replicate from the pop-up menu.
 For the Display options: to apply the same option to all variograms,
right-click that cell and choose Replicate.
 For variogram Colours: right-click any cell in the Colour column and
choose Fill With Random Colours.

Next, you’ll evaluate the variograms.

© Copyright MICROMINE 2014 53

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Optional Exercise 5.2: Evaluate the variograms and

find an initial lag interval

Displaying the variograms together allows you to see the overall shape of the
graph. Does the graph have low values near the origin, a smooth increase in
value, and an indication of a sill)? Can you estimate a range?

Rough estimate of range ( a):

The next step is to find the variogram that best represents the lag interval:

1. Click the Show Together button on the Chart toolbar to display the
first (5 m) variogram.

This variogram does not have enough intervals to reach the range, which
you can fix by adding intervals:

2. Change #Intervals in the Properties window to 40 and press Enter.

3. Click the Next button or press the Page Down key to step to the
10 m variogram.
4. Change its #Intervals to 30 and press Enter.
5. Click the Previous or Next buttons or press the Page Up and
Page Down keys to step through the variograms until you find the one
you consider to be best. Toggle the Pairs option as you assess each
variogram.
Hint: Ignore everything beyond your estimated range (around 130 m).
Refer to Lesson 2 if you have trouble identifying the best variogram.

Lag interval estimate (Δh):

As you page through the graphs you’ll see reasonable variograms starting at
lag intervals of around 20 m, which become progressively less detailed as the
interval increases. Note how lag intervals beyond 50 m do not show an indi-
cation of a sill, suggesting that they are too long to be useful.

At this point you might overwrite these settings with the next variogram. For
safety, switch the mode to Display Existing and re-save the form set:

6. Click the Form button at left of the Chart toolbar, change the Mode
to Display Existing, and re-save the form set.
7. Click OK to apply the change, overwriting the variogram file if prompted
to do so.
8. Keep the chart open for the next exercise.

54 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Variogram Map: Anisotropy

Notes:

This modelling step has two outcomes: to detect the presence of anisot-
ropy and, if present, determine the direction of maximum grade con-
tinuity. You do so by carrying out the steps in Lesson 4.

Anisotropy should be evident in the strike map. If so, use the dip and pitch
maps to find the 3D orientations of the plane containing the orebody and
the 3D orientation of Axis 1. Micromine will then calculate the direction of
Axes 2 and 3, which are always at right-angles to Axis 1 and each other.

You typically save this information as a variogram control file to be used

in the directional variograms.

If you cannot detect anisotropy on the strike map then your study area is
probably isotropic, at least in plan view. It may still be vertically anisotropic
so you should evaluate a number of dip maps aligned at 90° or 45° strike
increments before considering a fully isotropic model.

Directional: Modelling Three Axes

The last group of variograms is dedicated to modelling the directions of
grade continuity identified on the variogram map. Alternatively, you may
use the more traditional variogram fan workflow described in Appendix 3.

The outcome of this step is to obtain fitted variogram models for the axes
that define the directions of grade continuity. Three-directional variogram
models are necessary for 3D data such as a structurally controlled hard-rock
orebody. Only two-directional models are needed for 2D data such as the
elevation or thickness of a coal seam, and a single omnidirectional model is
enough for a purely isotropic study area. To work with two-dimensional data
just leave the Z field blank. As with the variogram map Micromine will adjust
the available options to suit.

The created models are then used as input for cross-validation and kriging,
described later in this course.

Kriging requires either an omnidirectional or a full three-axis set of

variograms. To convert a two-axis model to three axes, create a third
axis with a very long range at right-angles to the two existing axes.

Although the variogram map and directional variograms share many input
parameters they use them very differently. A variogram map is a two-
dimensional plane and its input parameters are a compromise that must
accommodate all possible directions falling within that plane. In comparison a
directional variogram is a one-dimensional graph, and because each vari-
ogram represents a single direction its lag interval, tolerance and bandwidth
can be individually tailored. Although map parameters may provide useful
starting values for a variogram graph, and vice versa, you should not force-fit
the parameters from one to the other.

© Copyright MICROMINE 2014 55

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Setting Directions

The recommended way to set the directions of the variogram axes is to ob-
tain them from a variogram control file created in the variogram map.
This method has several advantages over the manual approach:

 It automatically passes in the directions without you having to manually

re-enter them
 It ensures the three axes are at right-angles to each other
 It synchronises the nugget and the number of components along with
their types and partial sills across the three axes.

Using a variogram control file enforces geometric anisotropy so you can

focus on fitting the theoretical models.

In comparison, the manual method requires you to know the azimuth and
plunge of Axis 1, along with the plunge of Axis 2. You typically obtain this
information from a series of variogram fans as described in Appendix 3. To
ensure the axes are at right-angles to one another, enable Automatically
calculate angles for second and third directions, set the Direction to
SECOND or THIRD as needed, and enter the azimuth and plunge(s).

Choosing Tolerances
Variogram directions incorporate a tolerance, which represents the angular
size of the search; Micromine includes points within that amount on either
side of the azimuth (Figure 5.2). Increasing the tolerance widens the search,
like defocusing a flashlight, and reducing the tolerance focuses it. A Conical
search uses a circular cone shape, whereas a Pyramid-like search uses a
four-cornered pyramid shape.

Figure 5.2: Angular tolerance and bandwidth

The angular tolerance is 22.5° and the bandwidth is 45 m. Any points that
fall within the shaded area are paired with the tail.

56 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Increasing the tolerance may help to obtain more point pairs and smooth out Notes:
a noisy variogram, but at the expense of reducing the visible anisotropy. (As
an extreme example, increasing the tolerance to 90° will reproduce an omni-
directional variogram.) A good initial tolerance is 22.5°, which produces a
45° search that is neither too selective nor too broad.

This tolerance is applied differently compared to the variogram map, and is

measured in degrees instead of sectors. Although you should not force-fit
map values into the graph, the following equation may be helpful for com-
paring a variogram map with a directional variogram:

Directional tolerance = overlap × (360 ÷ number of directions)

The tolerance may be further modified by a bandwidth, which determines

the maximum width of the search in coordinate units and prevents the
search region from becoming too large at long lag distances. Unlike the map,
this bandwidth represents the total width of the search corridor. Consider
applying a bandwidth to limit the overall width of the search when the tol-
erance or study area is large.

You’ll set up the directional variograms using a variogram control file in the
following two-part exercise.

Exercise 5.3 (Part 1): Set up variogram directions

using a variogram control file

In this part of exercise you’ll define the directions and their tolerances:

1. Click the Form button at left of the Chart toolbar, change the Mode
to Calculate and the Direction to Directional.
2. For safety, skip the Search Definition and enter a new Semi Variogram
File name of SVG_AuCut_Three_Axes.
3. Save a new form set with the Title AuCut All Three Axes .

You’re ready to set up the individual variograms without accidentally over-

writing the previous file or form set.

4. Click the Semi Variogram Directions button in the Search Definition.

5. Click the Clear button near the top right of the form.
6. Click the browse button in the Variogram control file response
and choose the AuCut file you created in Lesson 4. Inspect the dialog.

The dialog now lists the azimuths and plunges you defined on the map,
which are not editable, in Axis 1 to 3 order. Now you must supply the tol-
erances. For this exercise you’ll use 15°, which is equivalent to the overlap
used in the variogram map. You can always adjust it later.

7. Enter 15 in any Azimuth Tolerance cell and then right-click |

Replicate it. Repeat in any Plunge Tolerance cell.
8. Keep the dialog open for the second part of the exercise.

© Copyright MICROMINE 2014 57

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Adjusting the Lag Intervals

The three axes you loaded in Exercise 5.3 are one-dimensional variograms,
and because of this there is no need to compromise on their parameters.
Because the sample spacing is likely to be different for each axis it is vital to
adjust the lag interval to suit each one. You may also wish to experiment
with the tolerance or bandwidth. Unlike the variogram map, there are no lag
overlaps in the variogram chart; the lag tolerance is always defined as plus
and minus one half of the lag interval, as illustrated on Figure 2.5.

You adjust the lag intervals using the corresponding Interval slider in the
Properties window. Aim for a smooth graph that shows the properties of a
good variogram. Lags supported by many pairs are more reliable than those
with few pairs, and the first third of the variogram should ideally have the
best support. Longer lags are not as important. Toggle the Scale option for
a visual check or the Pairs option for a numerical check.

Your aim is to produce three variograms that represent the spatial charac-
teristics of the data while maximising detail and minimising noise. This aim
may be hard to achieve, and in practice the final variograms are usually a
compromise between quality and detail. For example it may be hard to
gather enough pairs for the first (and theoretically the most important) lag
interval without sacrificing detail at the beginning of the graph.

The Axis 3 lag interval

In a well-designed drilling program the variogram for Axis 3 sometimes re-

sembles a downhole variogram. This occurs when Axis 3 is sub-parallel to the
drilling and is informed by closely-spaced downhole data, whereas Axes 1
and 2 are informed by more widely spaced between-hole data (Figure 5.3).

Figure 5.3: Axis 3 is sometimes sub-parallel to the drilling

Consider using a short lag interval for Axis 3 (approaching the downhole
interval length) whenever its direction is close to the drilling direction. How-
ever, it should not be treated as a copy of the downhole variogram because
it has a tolerance and will most likely capture data from nearby holes at
longer lag distances.

58 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Exercise 5.3 (Part 2): Set up lag intervals and display Notes:
the variograms

In this part you’ll use the information gained from the omni-variograms to
provide initial lag interval estimates:

1. Enter a lag Interval of 20 and number of intervals of 12 in the first

row (Axis 1).

Good omni-variograms began to appear at a 20 m lag interval, and 12 inter-

vals will span half of the study area. For now you’ll use these values for all
three axes.

2. Right-click each value and Replicate it to the second and third rows
(Axes 2 and 3).
3. Enable any Line and Scale checkbox and right-click | Replicate them
to the other rows.
4. Set the first row (Axis 1) colour to red, the second (Axis 2) to green and
the third (Axis 3) to blue.

Although you can use any colours, the red-green-blue sequence is often used
to indicate a first-second-third hierarchy.

5. Back on the main dialog, switch to the Chart tab and set the Title to
Three-Axis Au (cut) .
6. Click OK to calculate and display the three-directional variograms. Your
screen should resemble this screenshot:

The initial variograms show reasonably coherent shapes, and you can now
adjust them to reveal as much information as possible.

© Copyright MICROMINE 2014 59

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 7. Click the Show Together button to show the Axis 1 variogram.
8. Click into the Interval cell in the Properties window and adjust the
slider bar until you produce the best possible variogram.
9. Click the Next button to move onto the next variogram.
10. Keep adjusting the intervals, using Previous or Next to step
between axes, until you produce suitable variograms.
11. Click the Show Together button to show the variograms together.

Your final variograms should look like this screenshot (lag intervals are 30,
18 and 23; number of intervals changed to 10, 10 and 13):

The axes generally display the properties of a good variogram, reaching a

total sill around 0.87. Although the Axis 3 graph (blue) appears to be under-
supported at short lags, the first lag incorporates over 13,000 pairs, which is
more than adequate.

12. Click the Form button at left of the Chart toolbar, change the Mode
to Display Existing, and re-save the form set.
13. Click OK to apply the change.
14. Keep the chart open for the next exercise.

Fitting the Theoretical Models

You have finally determined the three experimental variograms that define
the continuity axes of mineralisation within the study area. The last step is to
fit theoretical models; once these are fitted you will have the last piece of
data needed for grade interpolation using a kriging interpolator.

60 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Earlier in this course you learned the difference between geometric and zonal Notes:
anisotropy. Geometric anisotropy is much easier to model than zonal anisot -
ropy and is the model that you should attempt unless there are clear geo-
logical reasons for choosing otherwise.

For the models to maintain geometric anisotropy they must have:

 Identical nugget values

 The same number and type of components
 Identical partial sills, but
 Different ranges

Micromine takes care of these requirements when you use a variogram con-
trol file; you must manage them yourself if you choose the manual workflow.

Using a Variogram Control File

A variogram control file does more than just store axis directions defined in a
variogram map. In a 3D setting it also synchronises the nugget and the
number of components along with their types and partial sills across the
three axes. Only the ranges can differ between axes. In other words Micro-
mine automatically enforces geometric anisotropy so you can concentrate
on fitting the theoretical models.

Mouse actions can be imprecise, and you should enable Lock Axes on the
Chart Controls once you are satisfied with the partial sill of each component.
This option locks the sills whilst still allowing adjustments to the ranges.

Once you have created a three-axis model you then save it to a new vari-
ogram control file, which includes the original orientations along with the
properties of the model axes. Although it is possible to overwrite the original
file we recommend always saving a new file; working this way protects your
original orientations and provides a simple way to assess alternative
scenarios by saving a different file each time.

The most effective way to fit models using a variogram control file is to work
in four steps, repeating steps 2 through 4 until the models are finalised:

1. All axes visible: Enter the nugget from the downhole variogram
2. All visible: Fit components to any nested structures. Give each an
approximate partial sill that is suitable for all three axes
3. Per axis: Refine the models and then Lock Axes for each component,
remembering that sill adjustments also affect the invisible axes.
4. Per axis: Refine the ranges.

You’ll fit theoretical models using this workflow in the next exercise.

© Copyright MICROMINE 2014 61

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Exercise 5.4: Fit theoretical models to the three axes

To fit models to the variograms:

1. Ensure Show Together is enabled.

2. Click the Chart Control Pane button to show the Chart Controls.

Micromine fits linear variograms to the three axes using estimated parame-
ters, but because they are identical they appear as one. However, the visible
handles only apply to the selected variogram direction, which is usually the
first one in the list.

3. Enter the Nugget value you determined in Exercise 3.1 into the Nugget
response in the Chart Controls.

This value, which you measured on the downhole variogram, will be applied
to all three axes. Don’t change it.

4. In the Chart Controls, change Component 1’s Type from LINEAR to

EXPONENTIAL.
5. On the graph, drag Component 1’s handle until the model roughly fits all
three axes.

Partial sill (i.e. vertical) adjustments apply to all three axes, but changes
to the range only apply to the selected axis.

6. If necessary change the component type or add extra components to

get a reasonable fit with all three axes.
7. Disable Show Together to show only the first variogram, and use
the handles to make small adjustments to the fitted curve.

Remember: when you use a variogram control file any vertical (partial sill)
adjustments also affect the directions you can’t see.

8. Click the Next or Previous buttons (or press Page Down or

Page Up) to switch between variograms, adjusting each one in turn.
9. Optionally, give each axis a new Colour via the Chart Controls.
10. For fine control, click into a Range or Partial Sill response and adjust
the value by rolling the mouse wheel. Inspect the Goodness of Fit in
the Properties window as you work.

As you refine the axes you should be making smaller adjustments.

11. Enable Lock Axes to lock each component’s partial sill.

12. Make small changes to the Ranges for each component/axis
combination as needed.
13. Periodically toggle Show Together to visually validate the three
variograms.

Your result should look like the screenshot on the facing page.

62 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:

No single sill value suits every axis and component, and some compromises
were necessary to fit the models. In this exercise you avoided zonal anisot-
ropy (on the assumption that the data was properly domained) by selecting
an average sill that matched the three axes as closely as possible.

That said, only the shapes of the graphs matter and the overall sill is largely
irrelevant. The fitted axes do still represent the main spatial characteristics of
the study area: they are aligned to the directions of continuity, Axis 1 has the
longest range, and Axes 2 and 3 have progressively shorter ranges. Cross-
validation of the variograms will help to determine their suitability for kriging.

The last step in the variography workflow is to create a new variogram con-
trol file. For convenience you’ll also add that file to the original three-axis
variogram form set so that it will always be available.

Exercise 5.5: Save and embed a variogram control file

To save the new variogram models into a new control file:

1. Click the Create Variogram Control File button at right of the

Chart toolbar and enter a File name of AuCut_Directions_(Scenario_1) .
2. Click Save to create the file.

To embed the new control file into the original form set:

1. Click the Form button at left of the Chart toolbar, followed by the
Semi Variogram Directions button on the Input Data tab.
2. At the bottom of the form, replace the AuCut Variogram control file
with AuCut_Directions_(Scenario_1) .

© Copyright MICROMINE 2014 63

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 3. Right-click the new control file and inspect its contents.

Along with the original orientation information the new control file also con-
tains all of the processing parameters associated with the experimental and
fitted theoretical variograms. The file is fully self-contained, making it easy to
share your modelling parameters with co-workers.

4. Close the Directions dialog and then re-Save your AuCut All Three Axes
form set.

Now the variogram models will appear whenever you re-open the form set.

5. Close the Semi Variograms window once you’re done.

64 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 5 Summary Notes:

This lesson introduced the variogram modelling workflow and 3D variogram

modelling. The topics covered so far are:

The variogram modelling workflow comprises following variograms:

1. Downhole (estimate the nugget effect)

2. Omnidirectional (optional – assess the data)
3. Variogram maps (detect anisotropy and determine continuity directions)
4. Directional (fit theoretical models)

To display an omnidirectional experimental variogram:

Select Stats | Semi Variograms, then

Set the Mode to Calculate and the Direction to Omnidirectional, and
Click the Omnidirectional Semi Variograms, then
Fill out the Omnidirectional Semi Variograms form.

To quickly fill out the semi variograms form:

For incrementing numbers, right-click a cell and choose Calculate.

For constant numbers, right-click a value and choose Replicate.
For Display options: right-click a cell and choose Replicate.
For Colours: right-click a Colour cell and Fill With Random Colours.

To obtain directions from a variogram control file:

Select Stats | Semi Variograms, then

Change the Mode to Calculate and the Direction to Directional, and
Click the Semi Variogram Directions button, then
Click the browse button in the Variogram control file response
and choose the control file.

To apply a directional tolerance:

Use a value of 22.5° to produce a 45° search, or

Reduce the tolerance to better represent a highly anisotropic study area,
or increase it to capture more point pairs.

To adjust the lag interval:

Adjust the slider bar in the Properties window until you produce the
best possible variogram, remembering that
Axis 3 may be sub-parallel to the drilling and may support a smaller lag
interval than Axes 1 and 2.

© Copyright MICROMINE 2014 65

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 5 Summary

To assess the quality (level of support) for each lag interval:

Toggle the Scale option in the Properties window, or

Toggle the Pairs option, and
Aim for the first third of the variogram to have the most pairs.

To comply with geometric anisotropy the models must have:

Identical nugget values

The same number and type of components
Identical partial sills, but
Different ranges

To fit a set of 3D variogram models when a control file is used:

Click the Chart Control Pane button, and

Enter the Nugget value you measured on the downhole variogram, then
Fit components to any nested structures, while
Giving each component a partial sill that suits all three axes, then
Refine the models and, when suitable sills are found,
Lock Axes for each component, then
Refine the ranges.

To save a variogram control file containing fitted models:

Click Create Variogram Control File on the Chart toolbar, and

Enter a File name, then
Click Save to create the file.

To embed a variogram control file in the main Semi Variograms form:

Click the Form button on the Chart toolbar, followed by the Semi
Variogram Directions button, and
Select the Variogram control file, then
Re-Save the main form set.

66 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 5 Summary Notes:

Good Practice

 Use downhole (or vertical) variograms to estimate the nugget effect.

 Handle zonal anisotropy by re-domaining the study area in the
direction with the highest sill. Otherwise, add another component with a
very long range to the model.
 Distinct shoulders or changes in slope will help you determine the
number of structures in the experimental variograms.
 Concentrate on Axis 1 (the direction in which grades are most
continuous) and Axis 3 (the direction in which you have the most data).
 The three axes will rarely share the same partial sills, but as long as they
are close you can find average sills (by eye) that provide a reasonable
fit for all three axes.

Help Topics

For information on: See:

Semi-variograms Statistics > Semi-variograms

Semi-variogram Statistics > Semi-variograms >

modelling New workflow (page topic)

Semi-variogram Statistics > Semi Variograms >

model types Semi Variogram Model Parameters >
type of model (link on page)

Additional Resources and Further Reading

Goovaerts, P. (1997). Geostatistics for Natural Resource Evaluation. (Applied

Geostatistics Series). Oxford University Press: New York

Vann, J., Jackson, S., and Bertoli, O. (2003). Quantitative Kriging

Neighbourhood Analysis for the Mining Geologist – A Description of the
Method with Worked Case Examples, in Proceedings 5th International
Mining Geology Conference 2003. The Australasian Institute of Mining and
Metallurgy: Melbourne.

© Copyright MICROMINE 2014 67

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Lesson 6 – Other Variogram
Types
Duration: 30 minutes

This lesson introduces some alternative variogram types that may be helpful
for modelling mineral deposits that do not have a normal grade distribution.

After this lesson you’ll be able to:

 Use relative variograms when a study area exhibits proportional effect

 Understand the use and associated risks of log-normal variograms
 Use indicator variograms for highly irregular or mixed populations
 Use uniform/rank variograms when all else fails.

Introduction
A commodity with a positively skewed population may contain very large
values that adversely affect the variograms; in other cases a domain may
contain multiple populations that cannot be spatially separated from one
another. Micromine provides a variety of variogram types that may prove
helpful in these situations.

Variogram Types
Pair-wise Relative (Proportional Effect Present)
A variogram may become difficult to interpret when proportional effect is
present. The geologists’ preference for sampling high grades more densely
than low grades (i.e. clustering the data) usually results in high-grade sam-
ples contributing heavily at short lags. As the lag increases, the contributing
data becomes more representative, and the lag mean and variance decrease
as a result. This adversely affects the experimental variogram by overesti-
mating the relative nugget. In some datasets the values at small lags may
appear larger than those at larger lag distances, giving a false impression
that the data is spatially unstructured (Goovaerts, 1997, p.84).

You can compensate for proportional effect by generating a pairwise rel-

ative variogram. The pairwise transformation scales the values by dividing
them by the squared average of the head and tail values contributing to each
lag. To display a pairwise relative variogram in Micromine, set the Type to
Pair-Wise Relative Variogram.

Micromine works with relative values whenever you model a relative vario-
gram. However, it back-transforms the values during kriging.

68 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Natural Log (Positively-skewed Population) Notes:

When a dataset contains very large values it may be necessary to transform

the data in order to reduce the influence of these values on the summary
statistics. One method is to use logarithms in place of the original values
(provided they are always greater than zero) by setting the Transformation
to NATURAL LOG. The Additive constant provides a way to improve the fit
of the log-transformed data with a true natural log distribution. It is calcu-
lated via Stats | Distribution > Model | 3 Parameter, and because it is
added to all values it has the effect of increasing the natural logarithms of
small values more than those of large values.

Although it is easy to apply this transformation to positively-skewed data, it

may cause problems when the kriged values are back-transformed (e.g.
Goovaerts, 1997, pp.16-17; Clark & Harper, 2000, Section 12.2.2). This back-
transformation may exaggerate any errors associated with interpolation,
which tend to be worst for extreme values. You should always validate the
result carefully if you do use this method. However, to avoid this uncertainty
we recommend using pair-wise relative variograms instead, which are
generally analogous to natural log variograms (Clark & Harper, 2000).

Indicator (Mixed Populations Present)

The variograms studied so far have measured spatial continuity over the full
range of grades within the study area. However, it is possible for grade conti-
nuity to be different in high grade areas versus medium or low grade areas.
Consider a typical precious metal deposit: random and discontinuous high-
grade zones may exist within a much more continuous low-grade back-
ground. Modelling variograms for the high grades would return a very differ-
ent result than modelling the entire dataset.

An indicator variogram models the proportion of points that falls above or

below a specified indicator value (referred to as the Cutoff value in
Micromine). To create an indicator variogram in Micromine, simply set the
Transformation to INDICATOR and enter a Cutoff value. Micromine will
assign a value of 0 to values below the indicator and a value of 1 to values
above it, and use the 0’s and 1’s to construct the variogram instead of t he
raw data. Indicator variograms are essential for study areas that contain
mixed populations and can’t be statistically or spatially separated into do-
mains. Only use them with non-linear kriging interpolators such as median
indicator kriging or multiple indicator kriging.

You can define the Cutoff value in several different ways, but some com-
mon techniques are:

 Determine the median of the data and use this as the only Cutoff value.
 Use a histogram to find the actual transitions between populations
and use these as Cutoff values.
 Arbitrarily divide the data into quantiles (such as quartiles or deciles)
and use these as Cutoff values.

Using multiple indicators can easily blow out the amount of time needed to
model variograms, especially if your study area requires you to model multi-
ple domains or multiple elements.

© Copyright MICROMINE 2014 69

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: In the workplace you should run the full suite of indicator-transformed ex-
perimental variograms described in Lesson 3, including an indicator trans-
formed downhole variogram. The indicator-transformed continuity and
anisotropy will most probably be very different from the untransformed data.
You should also perform these steps on a per-domain basis.

You’ll model median indicator-transformed variograms in the next exercise.

Optional Exercise 6.1: Create median indicator-

transformed variograms

You’ll begin this exercise by finding the median value:

1. Select Stats | Quick Summary from the main menu.

2. Use Gold_Assay as the Input File and Au1_ppm as the first Field (do
not use Au1_ppm_CUT).
3. Enable the Detailed Report.
4. Enable the Ignore Blanks numeric exception and then Run the
function.
5. Leave the Report Viewer open with the Median value visible.

For brevity you’ll assume the median indicator-transformed variograms have

identical orientations to the ordinary variograms you created in Lesson 3.

To model the median indicator variograms, starting with the nugget:

1. Select Stats | Semi Variograms and Open your previously created

AuCut Downhole form set.
2. Change the Mode to Calculate and the Semi variogram field to
Au1_ppm. Do not use Au1_ppm_CUT.
3. Change the Transformation to INDICATOR and enter your median
value as the Cutoff value.
4. Change the Semi Variogram File name to SVG_M_AuCut_Downhole.
5. Save the new settings as a form set with the Title AuCut Downhole
(Median=x), where x is your median value.
6. Click OK to run the function, then click the Form button, change the
Mode to Display Existing, and re-Save.
7. Fit a theoretical model using the procedure from Lesson 3. Enter your
nugget value here:

Nugget

8. Save the variogram model as a form set with the Title Au (cut)
downhole (Median=x) , where x is your median value.

70 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Next, the directional variograms, which you’ll assume have the same orienta- Notes:
tions as the originals (in reality this would be very unlikely):

1. Open your AuCut All Three Axes form set.

2. Change the Mode to Calculate and the Semi variogram field to
Au1_ppm. Do not use Au1_ppm_CUT.
3. Change the Transformation to INDICATOR and enter your median
value as the Cutoff value.
4. Click the Semi Variogram Directions button and change the
Variogram control file back to AuCut.
5. Change the Semi Variogram File name to SVG_M_AuCut_Three_Axes.
6. Save the new settings as a form set with the Title AuCut All Three Axes
(Median=x) , where x is your median value.
7. Click OK to run the function, then click the Form button, change the
Mode to Display Existing, and re-Save.
8. Fit theoretical models to the three variograms.
9. Save the models as a variogram control file entitled
AuCut_Directions_(Scenario_2_Median=x) , where x is your median.
10. Close the Semi Variograms window once you are done.

Uniform/Rank (Strongly Skewed Population)

The UNIFORM/RANK transformation alters your data so that it has a uniform
distribution, in which the probability of each value occurring is equal. Values
are sorted in increasing order and are assigned a relative rank ranging from
zero (for the minimum value) to 100 (for the maximum), in effect converting
them to percentiles.

This transformation makes no assumptions about the data distribution and

may be the best option for strongly skewed or otherwise intractable datasets.

Use relative semi-variograms if...

• Proportional effect is present

Use indicator semi-variograms if

• You cannot separate mixed populations, or

• Grade continuity and orientation is different at different
grade ranges

Figure 6.1: Decision tree for choosing variogram types

© Copyright MICROMINE 2014 71

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 6 Summary

This lesson introduced some more advanced variogram types. The topics
covered so far are:

To create a pairwise relative variogram:

Change the Type to Pair-Wise Relative Variogram, and

Calculate the variogram.

To create a log-normal variogram:

Set the Transformation to NATURAL LOG, and

Optionally enter an Additive constant.

To create an indicator variogram:

Set the Transformation to INDICATOR , and

Enter a Cutoff value.

To create a uniform/rank variogram:

Set the Transformation to UIFORM/RANK.

Help Topics

For information on: See:

Semi-variogram Statistics > Semi Variograms >

model types Semi Variogram Model Parameters >
Type of model (page topic)

Additional Resources and Further Reading

Clark, I. and Harper W.V. (2000). Practical Geostatistics . Geostokos (Ecosse)

Ltd.

Goovaerts, P. (1997). Geostatistics for Natural Resource Evaluation. (Applied

Geostatistics Series). Oxford University Press: New York

72 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Notes:
Lesson 7 Variogram Cross-
Validation
Duration: 45 minutes

This lesson will teach you to use cross-validation to validate your variogram
models to ensure they’ll give you the best possible interpolation result.

After this lesson you’ll be able to:

 Understand and use cross-validation to confirm the quality of the

variogram models
 Plot and analyse the cross-validation results to identify conditional bias.

Introduction
Creating a variogram model is an inexact and somewhat artistic process, and
many different combinations of values produce what look like reasonable
models. Regardless of this subjective nature you still require a statistical
measure of the model’s goodness of fit. Cross-validation is one method for
quantifying this fit.

Cross-validation is performed by temporarily removing a raw data value and

using the variogram models to estimate the missing value from the sur-
rounding data. The difference between the removed value and the estimate
(the residual) is calculated, and the value is replaced. The process is re-
peated for each record in the dataset and cumulative statistics on the differ-
ences are generated.

Micromine also calculates the kriging standard error (the standard devi-
ation of the kriging error), which it uses to normalise the residual, producing
the error statistic. This is the variable of interest. The average error sta-
tistic should be close to zero and its standard deviation should be close to
one; values between 0.9 and 1.2 are generally acceptable.

An average error statistic of zero indicates that your estimates are globally
unbiased, in other words the global mean of the raw data is close to the
global mean of your estimates. However, conditional bias may still be pre-
sent as discussed later in this lesson. A standard deviation greater than
one suggests that the raw data is more variable than your estimates, and
a value less than one suggests it is less variable.

A biased result may indicate that the variograms do not represent the data
and may need to be remodelled. Conversely, an unbiased result with a very
high standard deviation may indicate that the raw data is too variable to be
fully represented with the available variogram models. Given the smoothing
effect of linear weighted average interpolators like IDW and kriging, this may
reveal limitations of the data rather than signify an error.

© Copyright MICROMINE 2014 73

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Analysing the Result

Simply inspecting the error statistic is not enough to properly validate your
variograms; you should also assess the cross-validation result in a number of
other ways by displaying:

 Basic summary statistics and histograms of the estimated and actual

values, along with:
 A scattergram of estimated versus actual grades to see how well
kriging reproduces the sample data. This graph will reveal any
conditional bias (scatter cloud should lie close to 45°)
 A Q-Q plot of estimated versus actual grades, which will also reveal
conditional bias (quantiles should lie close to the 45° line)
 A histogram of error statistic or residual to ensure they are normally
distributed, even if the original data is log-normal
 A scattergram of error statistic versus the actual grades, which may
also reveal any conditional bias (cloud should be close to horizontal)
 A scattergram of error statistic versus estimated grades to reveal if
the error is dependent on grade (cloud should be close to horizontal).
 The 3D points in Vizex, coloured by error statistic or residual to
reveal any spatial bias (colours should be randomly scattered)
 A variogram of error statistic or residual to reveal any spatial bias
(entire variogram should be close to horizontal).

Conditional Bias
Conditional bias occurs whenever an error is dependent on the class of
values under consideration. Conditional grade bias often shows itself as an
over-estimation of low values and an under-estimation of high values, usually
as a result of the smoothing effect of weighted linear interpolators.

Values may also be spatially biased by being systematically overestimated

in some regions and underestimated in others, which is usually a sign of
incorrectly oriented variograms. These potential issues are quickly revealed
using the charts and Vizex display described above.

Limitations
Cross-validation does have limitations and should be used with care. If the
original data is spatially clustered then the cross-validation result is also
clustered. This may mean that the global result does not represent the entire
study area, although it may do better at representing certain regions within it
(Isaacs & Srivastava, 1989, pp.357-358). In this situation you may wish to
calculate a declustered global mean of the residuals using the methods de-
scribed in MG 301 – Block Modelling.

Declustering only generates values at existing sample locations and can-

not represent the estimates at all unsampled locations. Because a block
model includes many more locations than the raw data, with different sample
distances, the standard deviation of the block model may be different from
the both the cross-validation and raw data.

74 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

The Process
Notes:

You carry out cross-validation by selecting Stats | Cross Validation and

using the cut and composited assay file as the Input data. The Model pa-
rameters option allows you to choose a variogram control file or a set of
individual variograms based on the anisotropic properties of the study area.
If you use individual variograms Micromine will validate their parameters to
ensure their nuggets, components, and partial sills are identical and their
axes are at right angles. Any errors at this stage will prevent you from con-
tinuing. This extra validation is not necessary when you use a control file.

You also need to specify a Data search neighbourhood. A simple technique

is to create an ellipsoid that matches the orientation and ranges of your
variograms. Alternatively, you can use the same incremental search
neighbourhoods you would use for multiple interpolation runs, which may
reveal sensitivities in goodness of fit to distance.

You’ll carry out a basic cross-validation using a single search neighbourhood

in the next exercise.

Exercise 7.1: Cross-validate the variogram model

To cross-validate your variograms:

1. Select Stats | Cross Validation from the main menu.

2. Set the Mode to Calculate and choose Gold_Assay as the input File.
3. Select the input field names, setting the Analysis field to
Au1_ppm_CUT.
4. Enter an Output Stats file name of Rep_Gold_CV_Stats and a File
name of Gold_CV_Output.
5. You won’t display the results graphically in this function so you can
ignore the Display Limits and Graphic Display options.
6. Enable the Ignore blanks Numeric Exception.
7. Right-click the Model parameters response and create a new form set
with the Title Au Directional.
8. On the Semi Variogram Parameters dialog, select Variogram
control file and choose AuCut_Directions_(Scenario_1) .
9. Right-click the control file and note down the ranges of the axes. You’ll
need them later.
10. Close the control file and then click OK to accept the variograms.
11. Right-click the Data search response and create a new form set with
the Title Cross Validation.
12. On the Search Definition tab, set the Style to Ellipsoidal, the Radius
to 1, and Sectors to EIGHT.
13. Click the Import button at right and import the
AuCut_Directions_(Scenario_1) file.

© Copyright MICROMINE 2014 75

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Importing a control file automatically supplies the orientations. However, be-
cause there are many ways to define the radii of an ellipsoid you must enter
them yourself.

14. Switch to the Ellipsoid Properties tab and set the factors to the
Component 1 ranges from the variogram control file.
15. Run the function and inspect the data in the information bar at right,
particularly the Error Statistic row. Are the Mean and Std Dev around
zero and one, respectively?

Depending on your variograms, the mean error statistic will be very close to
zero and the standard deviation will be around 1.2. This indicates a good
global result but does not provide any information about conditional bias .

16. Close the Cross Validation display, then right-click | View the
Gold_CV_Output file and inspect its contents.

Each record in this file corresponds to an original data record. The

Au1_ppm_CUT field contains the original assays and the ESTIMATE field
contains the estimate. The RESIDUAL is simply the difference between the
two. ERROR_STAT is the residual divided by the STD_ERROR .

A perfectly normalised error statistic has a mean of zero and a standard

deviation of one.

Having generated the statistics and inspected the overall (global) result, you
can now turn your attention to some of the other methods listed at the be-
ginning of this lesson. You’ll display a scattergram of estimated versus actual
grades in the next exercise.

Exercise 7.2: Create a scatter-plot of the cross-

validation results

To create a scatter-plot of estimated versus actual grades:

1. Select Stats | Scattergrams from the main menu and ensure the
Input Data tab is active.
2. Set the File to Gold_CV_Output, the Y axis field to ESTIMATE and the
Y mode to Natural Log.
3. Set the first X Axis field to Au1_ppm_CUT and its Mode to Natural Log.
4. Enable the Ignore blanks Numeric Exception.
5. Enable Display Regression Line and set the Regression type to
Linear.
6. Ensure Force through origin is disabled.
7. Enable Display Confidence Bands and set the Confidence to 95%.
8. Switch to the Display Options tab and ensure Show Regression
Equation is enabled.
9. Change the Colour Coding to Density and set the % of range to 5.

76 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

10. Choose a Palette and ensure the Calculation mode is set to Notes:
PERCENTILE .
11. Switch to the Chart tab and set the Title to Cross-validation (Estimate
vs. Actual).
12. Click OK to display the graph, which should resemble this screenshot:

13. Inspect the data in the Properties window, in particular the means,
variances, slope of regression (b), Correlation coefficient, and
Precision.
14. Ensure Sync Selection on the Chart toolbar is set to Both.
15. Save a form set with the Title Cross-validation (Estimate vs. Actual).
16. Keep the chart open for the next exercise.

The regression line shows a slight overestimation of low grades along with an
underestimation of high grades, as shown by the slope of about 0.7. Such
conditional grade bias is a typical smoothing effect of linear interpolators
like kriging. At first glance the scatter cloud appears to spread widely from
the regression line. However most of the points lie relatively close to the line
as shown by the density colouring. This low precision is typical for precious
commodities where individual samples may be significantly different from
their neighbours. These samples are often so different that they cannot be
accurately estimated from the surrounding data. You can prove this by se-
lecting outliers in the top-left or bottom-right corners of the chart and com-
paring the estimates with the actual grades.

A scatter-plot (and the other graphs listed at the beginning of this lesson)
provides an overall indication of the estimated grades but cannot tell you
how those grades are distributed spatially. Nor can it highlight regions where
there is a systematic over- or underestimation of grade. You’ll assess the re-
sult spatially in the next exercise.

© Copyright MICROMINE 2014 77

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Exercise 7.3: Display the validation results in Vizex

In this exercise you’ll inspect the cross-validation results spatially in Vizex,

and investigate any locations where large errors occur. To do so:

1. Ensure the Wireframes and Drillholes Vizex saved view is loaded, and
turn off the Gold Samples [MIP] layer.
2. Create a new Vizex Point layer using Gold_CV_Output as the Input File.
3. On the Points tab, enable Show Points and Use Symbols and choose
a solid dot () scaled 3 grid units for the Default symbol.
4. Use ERROR_STAT as the Colour field.
5. Create a Colour set using 64 percentiles along with the Blue White
Red (64) palette. Save it as a form set Titled Cross validation dichroic .

The error statistic is a negative number (blue) when you overestimate the
grade and a positive number (red) you underestimate it. You can avoid this
by reversing the values before applying the palette.

The SIZE field is currently filled with zeroes so next you’ll fill it with the ab-
solute value of ERROR_STAT (there is no such thing as a negative size).

6. Enable the Scale option and set the Scaling field to SIZE.
7. Right-click | Edit the Scaling field response, and then use the
calculator to place the absolute value of ERROR_STAT into SIZE.
8. Enable the Natural log transform, set the Method to Factor, and the
Scaling factor to 2.
9. Click OK to create the display layer, which should resemble this:

10. Inspect the points and note the location of any large red or blue dots.
11. Right-click the Untitled (Gold_CV_Output.DAT) layer and choose Sync
Selection | Both from the pop-up menu.
12. Right-click the Scattergrams tab near the top of your screen and
choose New Vertical Tab Group from the pop-up menu.

78 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

13. Click a large red or blue dot in Vizex and then inspect the chart. The Notes:
point will be highlighted.
14. Select an extreme outlier in the chart and then inspect the highlighted
symbol in Vizex.
15. Close the chart once you are done.

Overall, small symbols (indicating good grade estimates) dominate the dis-
play. Where present, large symbols (poor estimates) are randomly scattered
throughout the data, indicating that there is no spatial bias (systematic
under- or overestimation). Furthermore, the large symbols usually occur in
red/blue pairs, suggesting that they represent grade changes at distances
too small to estimate using the available data and interpolation method.
Given that the smoothing effect of linear interpolators makes it impossible to
obtain a perfect match, you will treat this result as valid.

Advanced Exercise 7.4: Complete the validation

We leave the remaining validations as an advanced exercise that you can

complete in your own time. Your results should resemble these charts:

Scatter-plot (Estimate vs. Actual) Q-Q Plot (Estimate vs. Actual)

Histogram (Residual) Scatter-plot (Error Stat vs. Actual)

Scatter-plot (Error Stat vs. Estimate) Omni-variogram (Residual)

© Copyright MICROMINE 2014 79

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: The Q-Q plot illustrates the normal overestimation of low grades and under-
estimation of high grades associated with linear interpolators. It is a good
result, and is easier to see here than on the scatter-plot. However, a Q-Q
plot cannot show the spread of values like a scatter-plot.

The residual histogram reveals that nearly three-quarters of the residuals

fall between -0.1 and 0.2 ppm. This is also a good result.

The two error statistic scatter-plots display regression lines with slopes
close to zero, another indication of a good result. (10 ppm was added to the
error statistic to make all values positive and compatible with natural log
axes.) The error statistic has a slight tendency to increase at high grades,
which occurs because interpolated estimates typically underestimate high
values, resulting in a larger error. Random negative errors are associated
with overestimations that occur wherever a low grade is sandwiched be-
tween two high grades.

The residual omni-variogram (created by stacking different lag intervals

using the steps in Exercise 5.1) represents the residual spatial structure not
accounted for by the model being validated. In a perfect result this graph
would be a horizontal line, indicating no left-over spatial structure, but this
example shows a distinct structure with a range of around 40 m. The most
obvious explanation is the mixture of supergene and primary domains, alt-
hough there could be many other geological causes.

Based on the relative nugget (the total sill divided by the nugget) this struc-
ture represents at worst around 20% of the total variation in the residual,
with the remaining 80% being spatially unstructured. Whilst not ideal it is
acceptable given the limitations of the data.

The semi-variograms are valid if...

• The mean ERROR_STAT is close to zero, with a standard

between 0.9 and 1.2
• The scatter plots show no conditional bias
• A Vizex plot shows no spatial bias
• A variogram of residual is horizontal, showing no spatial
structure

Figure 7.1: Decision tree for variogram validation

80 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 7 Summary Notes:

This lesson introduced variogram validation. The topics covered so far are:

To cross-validate a variogram model:

Select Stats | Cross Validation from the menu, and

Set the Mode to Calculate and choose an input File, then
Enter Stats file and Output File names, and
Create or load a Model parameters form set containing the
variogram(s) to be validated, and
Create or load a Data search form set that matches your variograms or
the expected interpolation neighbourhood, then
Run the function and inspect the data, and
Check that the average error statistic is close to zero and its standard
deviation is close to one.

Generate the following output to assess the results for conditional bias:

Summary statistics and histograms of the estimated and actual values

A scattergram of estimated versus actual grades
(scatter cloud should lie close to 45°)
A Q-Q plot of estimated versus actual grades
(quantiles should lie close to the 45° line)
A histogram of error statistic or residual
(should be normally distributed and close to zero)
A scattergram of error statistic versus actual grades
(cloud should be close to horizontal)
A scattergram of error statistic versus estimated grades
(cloud should be close to horizontal).
The 3D points in Vizex, coloured by error statistic or residual
(colours should be randomly scattered)
A variogram of error statistic or residual
(entire variogram should be close to horizontal).

Help Topics

For information on: See:

Cross-validation Statistics > Cross Validation

Additional Resources and Further Reading

Isaaks, E.H. & Srivastava R.M. (1989). Applied Geostatistics . (Applied

Geostatistics Series). Oxford University Press: New York

© Copyright MICROMINE 2014 81

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Lesson 8 – Ordinary Kriging
Duration: 90 minutes

This lesson will teach you the basics of ordinary kriging (OK). After this
lesson you’ll be able to:

 Understand the differences between OK and other methods like IDW

 Set up an OK interpolation in Micromine
 Know when to use an alternative interpolation method like relative
variograms or indicator kriging.

Introduction
In MG 301, Lesson 6 you hand-calculated the value of an unknown point
using inverse distance weighting (IDW). You calculated the sample weights
by squaring the distance between each input data point and the unknown
point, taking the inverse of the square, and then rescaling the values so they
summed to one. The estimated value was then a simple linear combination
of the data values and their weights.

However, although IDW is conceptually simple it does not account for the
true spatial continuity and anisotropy of the data, is severely affected by
clustering, and does not minimise the variance of the estimation errors.

Kriging Advantages
Ordinary kriging is an interpolation method that overcomes the limitations of
IDW. It still uses simple linear combinations of weighted sample values to
calculate the estimate, but it calculates the weights in a way that accounts
for spatial continuity and anisotropy. Another feature that differentiates OK
from IDW is its attempt to minimise the error variance.

OK sample weights are calculated from the variogram(s), which use statisti-
cal distances to the input data instead of geometric distances. This is
achieved by constructing a matrix that records the distances between each
sample and every other sample, providing the kriging interpolator with infor-
mation on the clustering of the input sample data (Isaaks & Srivastava, 1989,
p.300). Because the between-sample distances are known, the distance
weights can be automatically adjusted to account for possible redundancies
between closely-spaced samples.

Figure 8.1: IDW and OK grade estimates from redundant data

82 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Consider the data in Figure 8.1, showing three input points at equal dis- Notes:
tances from the block under investigation. IDW produces a result of 8,
whereas the OK result is around 7. IDW ignores the obvious closeness (and
redundancy) of the points on the left, giving all three equal weighting.
Kriging detects the closeness of the left-hand points and reduces their
weights accordingly. In a more extreme situation a point may be given a
negative weight if it is screened (obstructed) by a nearer one.

If the distribution of grades within a deposit is found to be isotropic, kriging

is carried out using a single omnidirectional variogram model. However, three
directional variogram models must be used if the study area is anisotropic.
The orientation axes of the models should be aligned to the first, second,
and third directions of continuity (usually down-plunge, ≈down-dip, and
across-thickness), and must be at right angles to each other. The three
ranges should be modelled to reflect the continuity of grade in each
direction. Micromine only supports geometric anisotropy, which is most easily
handled by creating a variogram control file like you did in Lesson 5.

Ordinary kriging requires the input data to resemble a normal population. If

mixed populations are present (this is apparent on the histogram and prob-
ability plot) then an alternative method such as multiple indicator kriging
(MIK) must be used.

Ordinary Kriging with Micromine

You perform ordinary kriging via Modelling | 3D Block Estimate | Ordi-
nary/Universal Kriging. This dialog is split into three tabs: Input Data,
Modelling Parameters, and Output. The Input file typically consists of the
cut and composited assays, although some deposits may justify a different
approach. You must supply a Sample length field if you did not composite
the assays. You may also wish to enable Accumulate sample weights,
which writes the cumulative total weight contributed by each sample back
into the input file. This option is essential for learning how much each sample
contributes to the block model.

Select Define blocks from file to use a blank block model as a prototype or
to perform multiple interpolation runs. If the model is sub-blocked, enable
Interpolate parent blocks only to force the grade for all sub-blocks to be
calculated as if the parent block were still in place.

The differences between OK and IDW are found on the Modelling Param-
eters tab: Mode optionally enables BLOCK or POLYGON kriging, Trend op-
tionally performs kriging with a trend, and Transformation optionally
enables log-normal kriging (refer to Lesson 9 – Kriging Variations).

Discretisation is enabled whenever you set the Mode to BLOCKS. Specify

the number of divisions in each axis direction to calculate and then average
the grade for multiple points within each block. You may optionally Set
negative kriging weights to zero to handle screened or highly clustered
data by setting their weights to zero, effectively removing them from the
grade estimate. You typically enable this option for grade estimates and dis -
able it for modelling tangible phenomena such as elevations.

© Copyright MICROMINE 2014 83

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Each Attribute Field requires a Semi Variogram form set containing an
omnidirectional variogram model (for an isotropic study area) or a three-
directional variogram model (for an anisotropic study area). The same vario-
gram form set may be used for multiple attribute fields.

Specify the search ellipsoid in the Data search, and set the Count field,
Count reference field and Min Count as needed. You normally count the
number of hole IDs and limit the minimum number of holes for each run.
Counting drillhole IDs and discretisation are described in MG 301, Lesson 5.

Use the Write… options on the Output tab to assess the quality of each
block estimate, which may prove useful for classifying the resource. Enter the
name of the Output File and add any extra Fields. You may optionally
create a verbose weighting report, which contains the coordinates,
distance, grade, and calculated weight for every point that contributes to
every block, or an audit file, which summarises the parameters used in that
kriging run. The verbose weighting report file can become huge so only use it
to debug a small portion of your model.

The key stage in this process happens between the first and all successive
runs. At this point you must change the name of the Model file to that of
the output model. Using the Add Fields option to add the run number is
vital if you intend to perform multiple interpolation runs.

Interpolating Multiple Elements or Domains

To interpolate grades in a project with one primary element and many sec-
ondary elements, simply add each one to the Attribute Field list. The first
attribute field is treated as the primary element and controls the search
strategy for all secondary fields. Only samples with primary element values
are used to estimate the secondary elements. If a sample is missing its pri-
mary value then that sample is also excluded from all secondary element
estimations. Micromine will warn you if it detects this situation.

A project in which each element is equally important requires separate

per-element interpolations, with individual search strategies and output
models for each element. The most effective way to work with this kind of
deposit is to automate the process using a macro or Python script that
creates the models and appends them into a final master model.

In a multi-domain project you may have to perform separate interpolations

for each domain, depending on the boundary relationships between them.

When you consider these scenarios with multiple interpolations it is clear that
many separate runs may be needed. A systematic approach is essential.

Choosing Data for OK Interpolation

The right data for OK interpolation depends on your preparation. If you
composited your assays then you should use the composite file; if not then
use the assay file with a Sample length field. Similarly, if you applied a
top cut then you should use the cut field. Optionally, add the uncut field so
you can assess the effect of the cut value. Refer to Lesson 3 of MG 301 –
Block Modelling for more information on cutting and compositing.

84 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Using Variograms to Determine Initial Search Radii Notes:

In Lesson 2 you learned that a variogram is a graph of spatial correlation

versus distance (and direction), and in Lesson 3 you learned that the vario-
gram range is the distance beyond which sample pairs are no longer corre-
lated. Knowing this, you might use the variogram ranges to determine the
search neighbourhood radii. For variograms with more than one component
it may be appropriate to use a proportion of the short ranges for Run 1 and
multiples of the long ranges for Runs 2 and onward, for example:

 Run 1 radii: Equal to two-thirds of the short ranges

 Run 2 radii: Equal to the long ranges
 Run 3 radii: A multiple of the long ranges.

Use caution when applying this method, especially if your variograms include
a component that has a long range and small partial sill, as this may produce
a false indication of continuity. A safer alternative is to set the radii to the
distances at which the variograms reach a certain proportion of the total sill:

 Run 1 radii: Distances at which models reach 80% of total sill

 Run 2 radii: A multiple of the Run 1 radii; alternatively, the distances at
which models reach 95% of total sill
 Run 3 radii: A multiple of the Run 2 radii or long variogram ranges.

This method, shown graphically in Figure 8.2, offers some protection from
false continuity by decoupling the radii from the numeric values of the rang-
es, whilst still honouring the overall shape of the variogram models.

Figure 8.2: Using proportions of the sill to determine radii

The radii and proportions above are examples; like most aspects of the re-
source estimation workflow there are no rules for determining search ellip-
soid radii. Resource estimation is a geological process supported by numbers,
not the other way around, and your decision to use a certain percentage,

© Copyright MICROMINE 2014 85

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: proportion, or multiple must be based on your understanding of the geology

and grade continuity. You must be consistent in your choice of parameters.

The Iron Data

Although the massive iron deposit used for this course does not exhibit all of
the features necessary to learn about variography you do still need vario-
grams to krige the data. Appendix 2 explains how the supplied variograms
were created using the combined high- and low-grade domains.

The following exercises introduce the planning needed for multiple interpola-
tion runs using Micromine’s kriging tools, starting with the initial search radii.

Exercise 8.1: Determine initial search radii using

distances to reach proportions of the total sill

The variogram ranges in this project are large relative to the drillhole spac-
ing, so you’ll compensate by determining the initial search radii from the
distances at which the variograms reach 70% and 85% of the total sill.

1. Select Stats | Semi Variograms from the main menu and open the
supplied Iron (TFe cut) All Three Axes form set.
2. Determine the total sill by summing the nugget and partial sills of all
components, and then multiply this by 0.7 to determine the 70% value,
and by 0.85 to determine the 85% value. Note down your values here:

70% of total sill:

85% of total sill:

3. Move the mouse along the overlapping theoretical models for Axes 1 and
2 until the Y-coordinate displayed in the status bar is close to the values
you calculated above.
4. Make note of the X-coordinates in the table below, and then repeat the
above step for Axis 3. (d70 = distance to reach 70%.)

d70 d85

Axis 1 radii:

Axis 2 radii:

Axis 3 radii:

With the radii determined you are now ready to set up OK runs 1 and 2.
You’ll use a multiple of the Run 2 radii for Run 3.

86 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Exercise 8.2: Complete the first OK interpolation run Notes:

From your previous investigation of the training data ( MG 301, Lesson 2) you
know that the boundary between the IRON_LG and IRON_HG_BX domains
can be treated as a soft boundary, with grades from either side contributing
equally to the interpolated result. Thus you can plan the filter logic for each
domain in advance, as listed in Table 8.1. (In this exercise the All Ore filter is
not strictly needed as it returns all of the composites, but you’ll use it to
highlight the use of filters in a multi-domain workflow.)

Table 8.1: Filter logic for the IRON_HG_BX and IRON_LG domains
Input Assay Model File
Domain
Filter Filter
IRON_HG_BX Domain = All Ore Domain = IRON_HG_BX
IRON_LG Domain = All Ore Domain = IRON_LG

Because you are performing multiple interpolation runs you’ll only use the
blank model for the first run. All other runs (for both domains) will use the
output model as input, as shown in Table 8.2.

Table 8.2: Model and output files for the first and subsequent runs

Run Model File Output File

1 OBM_IRON_BLANK OBM_IRON_OK
2+ OBM_IRON_OK OBM_IRON_OK

The final planning stage is to determine the ellipsoid radii and minimum point
requirements for each interpolation run. You established the initial d70 and
d85 radii in the previous exercise, and they will suffice given the lack of
detailed project geology. For simplicity you’ll also assume the radii and point
requirements are the same for both domains. Table 8.3 overleaf lists the
parameters for Runs 1 to 3. Note how the minimum point restrictions are
relaxed as the overall radius increases for each run.

In the workplace, downgrade the resource category of areas where you

lack a detailed understanding of the geology, even if the samples are
closely spaced.

To set up the OK interpolation, beginning with the IRON_HG_BX domain:

Input Data

1. From the main menu, select Modelling | 3D Block Estimate |

Ordinary/Universal Kriging.
2. On the Input Data tab set the Input File to Iron_Assay_2m_Comp,
ensuring the coordinate fields are correct.
3. Enable the Filter option and choose your DOMAIN = All Ore filter.

© Copyright MICROMINE 2014 87

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: The input file contains the (cut and) composited assay intervals from within
the entire orebody, and the filter restricts the data to the specific intervals
needed for the current domain.

4. Enable the Ignore blanks numeric exception.

5. Select Blocks from file and set the File to OBM_Iron_Blank, ensuring
all fields are correct.
6. Enable the Filter option and choose your previously created DOMAIN =
High-grade breccia filter. Right-click | View the OBM_Iron_Blank file
to preview the filter.

This restricts the interpolation to the blocks within the IRON_HG_BX domain.

7. Enable Interpolate parent blocks only.

Modelling Parameters

1. Switch to the Modelling Parameters tab and change the Mode to

BLOCKS to enable block kriging. This is the same as using discretisation
in IDW interpolation.
2. Click the Discretisation button and set the divisions to 4, 4, and 4 for
East, North, and Z, respectively.
3. Enable Set negative kriging weights to zero.
4. Set the first Attribute Field to TFe_CUT and its Semi Variogram to
the supplied Iron 2D Anisotropy form set.
5. Press Enter to create a new row and set the second Attribute Field to
T_Fe. Press Ctrl+” (ditto) to duplicate the semi-variogram form set.
6. Right-click the Data search response and create a new form set with
the Title OK Run 1.
7. Click the Import button at right of the dialog and import the supplied
TFe_Cut_2D_Anisotropy... variogram control file.
8. Set the point Parameters and Radius to the Run 1 values listed in
Table 8.3 and the Factors to your results from Exercise 8.1.

Table 8.3: Search radii and minimum points requirements

Data Search Add
Min
Run Min Form Set Fields
Radius Factor count
Points Title Value
1 1 d70 4 OK Run 1 4 1
2 1 d85 3 OK Run 2 2 2
3+ 4 d85 Blank OK Run 3 1 3

9. Back on the main dialog, enter a Count field name of HOLES, then click
any other active response on the form to activate the Count reference
field and Min count responses. Set these to HOLE_ID and 4.

This instructs Micromine to count the number of hole IDs within the search
neighbourhood and only write a block value when the count is four or more.

88 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Output Notes:

1. Switch to the Output tab and enable the first five Write… options,
which will help to indicate the quality of each estimate.
2. Enter an Output File name of OBM_Iron_OK.
3. Lastly, click the Add Fields button and define the following fields, which
you’ll use to store the run number and, ultimately, the resource
classification. Because this is the first run, set the Value to 1.

Field Name Type Width Decimals Value

RUN SHORT 1
CLASS CHARACTER 10 [blank]

Adding fields now is more efficient than modifying the block model later.

4. Save the main Ordinary/Universal Kriging dialog as a form set with the
Title HG_BX Run 1.
5. Run the function to perform the first OK run, and right-click | View
the output file once completed.
6. Load the blocks into Vizex, filtered to the IRON_HG_BX domain and
coloured by the RUN field.
7. Inspect the distribution of blocks interpolated by Run 1 and compare
those with the blocks not yet interpolated.

The blocks in the IRON_HG_BX domain that contain interpolated grades

could possibly classified as measured resources provided you had a detailed
knowledge of the project. You must now complete the remaining runs for
both domains before the model is complete.

In the workplace, review your parameters now if you’re not happy with the
distribution of the Run 1 blocks.

Exercise 8.3: Complete the subsequent OK runs

To populate the remaining blocks using multiple interpolation runs , begin

with Run 2 for the IRON_HG_BX domain:

1. On the Input Data tab, change the (Blocks from file) File to
OBM_Iron_OK.

This allows you to add subsequent runs to the output block model without
affecting the first run results.

2. Switch to the Modelling Parameters tab and repeat Steps 6, 8 and 9

from that part of Exercise 8.2 using the Run 2 settings from Table 8.3.
Save the Data Search to a new form set with the Title OK Run 2.
3. Switch to the Output tab and change the Add Fields value to 2.

© Copyright MICROMINE 2014 89

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 4. Save the Ordinary/Universal Kriging dialog as a form set with the
Title HG_BX Run 2.
5. Re-Run the function. You are prompted to update the output model.

This method leaves the data from previous runs intact and only populates
the values of empty blocks.

6. Visually scan through the OBM_Iron_OK file and ensure that all
IRON_HG_BX blocks are populated with grade values.

With the IRON_HG_BX domain completed you can turn your attention to the
IRON_LG domain:

7. Switch to the Input Data tab and change the (Blocks from file) Filter
to your Low-grade halo form set. Right-click | View the
OBM_Iron_OK file to test the filter.
8. Switch to the Modelling Parameters tab and repeat Steps 6, 8 and 9
from that part of Exercise 8.2 using the Run 1 through Run 3 settings
from Table 8.3. Re-use the Run 1 and Run 2 Data Search form sets, and
save the Run 3 form set with the Title OK Run 3.
9. Switch to the Output tab and change the Add Fields value 1, 2 or 3,
respectively.
10. Save your changes to the Ordinary/Universal Kriging dialog as form
sets with Titles LG Run 1 through LG Run 3, respectively.
11. Run the function after setting up each form.
12. Visually scan the OBM_Iron_OK file and ensure that all blocks are
populated with grade values.

A quick way to count the number of populated blocks is to view the

model via Blocks from file, which automatically includes the filter, and
then show the statistics of the RUN field.

The individual runs are summarised below:

Domain Block Definition

Run Output File
Filter Model File Filter
HG_BX 1 ...BLANK HG_BX ...OK
HG_BX 2 ...OK HG_BX ...OK
LG 1 ...OK LG ...OK
LG 2 ...OK LG ...OK
LG 3+ ...OK LG ...OK

90 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Preparing for Resource Classification Notes:

Blocks interpolated in the first run could possibly be classified as measured

resources provided you had detailed knowledge of the project. Similarly,
blocks interpolated by the second run could be classified as indicated re-
sources, and those interpolated by the third (and subsequent) runs could be
classified as inferred resources. If you do use run number as a classification
tool, remember to choose radii that match the sample spacing and continuity
of geology and grades for each resource category. Resource classification,
covered in MG 301, Lesson 8, is a geological process, not a numerical one.

When not to Use Ordinary Kriging

An assumption of ordinary kriging (and inverse distance weighting) is that
the input data is reasonably close to being normally distributed. In a well-
behaved deposit you can usually assure a reasonable degree of normality by
splitting different ore zones into domains. There are two common situations
where this is not the case:

 Proportional effect is present: This often happens with strongly

skewed data such as a precious metals deposit. In this case it might be
possible to still use ordinary kriging, but with relative variograms,
which will minimise the impact of proportional effect.
 Mixed populations are present: Despite your best efforts it may be
impossible to physically divide a study area into well-behaved normally
distributed domains. In this case an indicator kriging method is needed.
Depending on the complexity of grade distribution, and the time avail-
able to model it, either median indicator kriging (the simpler of the
two) or multiple indicator kriging may be used. Both methods are
non-linear, that is they make no assumptions about the underlying sta-
tistical distribution, and can therefore handle mixed grade populations.

The next lesson introduces these and other kriging variations.

© Copyright MICROMINE 2014 91

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 8 Summary

This lesson has introduced the basics of ordinary kriging. The topics that
you’ve covered so far are:

To set up for ordinary kriging:

From the main menu, select Modelling | 3D Block Estimate |

Ordinary/Universal Kriging, then
Enter the Input File name, select Blocks from file, and choose the
blank block Model File, and
Optionally, if the model is sub-blocked enable Interpolate parent
blocks only, then
Optionally, set the Mode to BLOCKS for block kriging, and
Set the desired number of Discretisation divisions in each axis, then
Add the required Attribute Field(s) along with their corresponding
Semi Variograms, and
Load or create a Data search form set containing the search ellipsoid
parameters, then
Set the Count field, Count reference field, and Min Count as
desired, and
Enable the Write… options, then
Enter the name of an Output File and add any extra Fields, then
Run the function.

To determine search ellipsoid radii from the semi-variograms:

Choose percentages of the total sill to represent Runs 1 and 2, then

Find the distances at which the models reach those percentages, and
Use a multiple of the Run 2 radii or long variogram ranges for Run 3.

Good Practice

 Always use the data that matches your preparation (whether flagged,
cut, or composited).
 Be thoughtful in your application of search ellipsoid parameters, espe-
cially the radii. Grade interpolation proceeds relatively quickly on modern
computers so be prepared to adjust the radii after grades have been
interpolated (and re-run the interpolation). The final radii may not be the
ones you started with.
 For all but the last run set the minimum number of holes to 2 or more to
avoid ‘spotted dog’ grade interpolations and classifications.
 Consider using block kriging to better represent the sample support of
the material you may eventually mine.
 Consider interpolating parent blocks only to minimise computer overhead
and be compatible with regularised models.

92 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Help Topics Notes:

For information on: See:

Semi-variograms Statistics > Semi-variograms

The search ellipsoid Block Modelling > 3D Block Estimate >

Data Search Ellipsoids

Declustering Block Modelling > 3D Block Estimate >

Data Search Ellipsoids > Data Search Definition

Count reference field Block Modelling > 3D Block Estimate >

3D Block Estimate: Kriging >
Ordinary/Universal Kriging >
Modelling Parameters (link on page) >
Count Reference field

Discretisation Block Modelling > 3D Block Estimate >

3D Block Estimate: Kriging >
Ordinary/Universal Kriging >
Modelling Parameters (link on page) >
Discretisation (page topic)

Block kriging Block Modelling > 3D Block Estimate >

3D Block Estimate: Kriging >
Ordinary/Universal Kriging >
Modelling Parameters (link on page) >
Mode (page topic)

Additional Resources and Further Reading

Goovaerts, P. (1997). Geostatistics for Natural Resource Evaluation. (Applied

Geostatistics Series). Oxford University Press: New York

Isaaks, E.H. & Srivastava R.M. (1989). Applied Geostatistics . (Applied

Geostatistics Series). Oxford University Press: New York

Stephenson, P. R., Allman, A., Carville, D. P., Stoker, P. T., Mokos, P., Tyrrell,
J., Burrows, T. (2006). Mineral Resource Classification – it’s Time to Shoot
the “Spotted Dog”! In Proceedings of Sixth International Mining Geology
Conference, AusIMM: Melbourne.

© Copyright MICROMINE 2014 93

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Lesson 9 – Kriging Variations
(Optional)
Duration: 30 minutes

This lesson will teach you the common alternatives to ordinary kriging, with a
focus on median indicator kriging and multiple indicator kriging.

After this lesson you’ll be able to:

 Understand the differences between simple kriging, universal kriging,

and ordinary kriging
 Understand the need for kriging with relative variograms
 Understand the difference between median indicator kriging and multiple
indicator kriging
 Create median indicator-transformed variograms
 Use metal percentiles as indicator cutoffs.

Introduction
There are many published kriging variations and this diversity can seem
overwhelming. However, some of it arises through different authors using
different names for the same kriging method, and some from historic meth-
ods that are no longer in common use within the resource community.

For the most part, once you are familiar with the basic capabilities of each
method their selection becomes relatively straightforward. Variations sup-
ported by Micromine include ordinary kriging, universal kriging, indicator
kriging, rank kriging and multiple indicator kriging. However, ordinary kriging
and multiple indicator kriging are the methods in most common use.

Universal Kriging (Kriging with a Trend)

Universal kriging is also known as kriging with a trend. It differs from or-
dinary kriging in that the local mean within each search neighbourhood is
modelled as a smoothly varying function of coordinates, rather than being
held constant. The trend is usually modelled as a low-order polynomial func-
tion such as linear (power 1), quadratic (power 2), or cubic (power 3).

In an interpolation situation (estimating values at locations surrounded by

input data) universal kriging offers no advantages over ordinary kriging, and
is disadvantaged through having to perform additional calculations. In an
extrapolation situation (estimating values at locations beyond the input data)
universal kriging may offer some advantages, but only if the physics of the
phenomenon suggest the most appropriate trend model. Care is needed as
the estimated values will depend very heavily on the chosen model
(Goovaerts, 1997, p.152).

94 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

To perform universal kriging in Micromine, use Ordinary/Universal Notes:

Kriging and set the Trend to LINEAR , QUADRATIC, or CUBIC depending on
the desired tend model. It is most often used when modelling a physical
surface such as a coal seam, which shows a clear trend in orientation.

Log-normal Kriging
If your project data is positively-skewed and you modelled natural-log vario-
grams, you may wish to carry out Log-normal kriging. To do so in Micromine,
use Ordinary/Universal Kriging, set the Transformation to NATURAL
LOG and optionally enter an Additive constant.

Use this method with care as problems may occur when the kriged values
are back-transformed (Goovaerts, 1997, pp.16-17). This may exaggerate any
errors associated with interpolation, especially for extreme values. Although
Micromine uses the presumably more reliable back-transformation of Clark
and Harper (2000), you must still validate your result carefully. To avoid this
potential problem we recommend using relative variograms instead. Refer
to Natural Log (positively-skewed population) in Lesson 3 for more
information.

Ordinary Kriging (Relative Variograms)

If proportional effect is present and you have modelled relative variograms,
simply use these instead of ordinary variograms in Ordinary/Universal
Kriging. Micromine will automatically back-calculate the grade estimate from
the relative variograms. However, kriging variance and kriging standard error
are output in relative units and should not be used.

Rank Kriging
Rank kriging, also known as rank order kriging or rank uniform kriging,
is a non-parametric or distribution-free method for modelling very strongly
skewed or irregularly distributed data (Clark & Harper, 2000). Used together
with the UNIFORM/RANK variogram transformation, it works by sorting the
values in increasing order, ranking them from zero to 100, and kriging the
ranks instead of the grades. The kriged rank order values are then back-
transformed to the scale of the original data.

Although rank kriging is not well represented in the literature it may be the
only alternative for situations when all other methods fail.

Indicator Kriging
Indicator kriging (IK) is a very popular non-parametric method that does not
require normally distributed input data, and is typically used when a study
area exhibits a complex grade distribution. Setting up an IK interpolation via
Modelling | 3D Block Estimate | Multiple Indicator Kriging differs
from ordinary kriging in only one way: a form set containing the required
indicator Cutoffs and their matching Variogram Formsets must be de-
fined, as shown overleaf. The remaining prompts are identical to those found
in ordinary kriging.

© Copyright MICROMINE 2014 95

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Non-linear Transformation

IK requires a series of indicator cutoff values. Grades falling below a given
cutoff value are assigned a value of zero and those falling above it are
assigned a value of one, in what is known as a non-linear transformation.
The 0/1 values are interpolated to produce a probability of the grade being
greater than that value, and this is used to calculate weighted average of the
grades that fall within that range (also known as a bin). Thus it does not
matter if a grade value is slightly above the cutoff value or a long way above
it; it is always assigned the same value.

Indicator cutoffs may be set to actual cutoff values between mixed popula-
tions or to arbitrary cutoffs such as quantiles. A histogram is usually used to
locate actual cutoff values by identifying valleys between histogram peaks,
and a cumulative frequency curve or statistical summary is used to identify
percentile values. Deciles provide a convenient number of bins, however it is
not compulsory for them to be applied to all indicator cutoffs. For example,
the last cutoff may be defined at 95% or some other relevant figure.

Alternatively, quantiles may be calculated using contained metal, in which

case each bin contains the same metal content instead of the same number
of samples, biasing the cutoffs towards the grades that contribute most of
the metal. Even without running IK, simply inspecting the corresponding
grade values reveals the amount of metal that occurs at each cutoff. In
extreme cases the highest value may contribute more metal than the last
decile, so some manual adjustment may be necessary.

Grade Estimation
The grade estimate for each bin is usually calculated using the mean of the
samples falling within it (Glacken & Blackney, 1998). However, because the
last bin is likely to contain the very high values, the estimate for this bin is
calculated using the median, which gives a more conservative value for pos-
itively skewed data than the mean. In the screen image below, note how the
grade estimate for the last bin is calculated using the median.

96 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Example 9.1: IK in a gold deposit using five cutoffs Notes:

Consider a gold deposit in which the following indicator cutoffs were used:

Cutoff grade
(g/t Au)
1.0 2.0 3.0 5.0 10.0

A series of indicator variograms was modelled at each cutoff and then

matched to the appropriate cutoffs in Micromine. A block model was then
created; the results for one block in the model are shown here:

Cutoff Probability of Average

Value Exceeding Cutoff Grade in Bin
1.0 1.00 1.45
2.0 0.82 2.47
3.0 0.61 3.81
5.0 0.46 7.10
10.0 0.12 17.31

These results can be interpreted as:

Proportion of
Grade Range Values Within
Range
< 1 g/t 1.00 – 1.00 = 0%
1 – 2 g/t 1.00 – 0.82 = 18%
2 – 3 g/t 0.82 – 0.61 = 21%
3 – 5 g/t 0.61 – 0.46 = 15%
5 – 10 g/t 0.46 – 0.12 = 34%
> 10 g/t 0.12 – 0.00 = 12%

During an IK interpolation Micromine calculates the mean of the input grades

for all but the last bin, for which it usually calculates the median. It also cal-
culates the proportion of the input values falling within each bin. For exam-
ple, the data falling between 1.0 and 2.0 g/t has a mean of 1.45 g/t and
comprises 18% of the grades. Micromine then uses these figures to calculate
a simple weighted average for the block:

This calculation is called an e-type or expected value estimate, and is

repeated for all of the blocks in the model.

© Copyright MICROMINE 2014 97

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Indicator Kriging Approaches

IK requires indicator variograms for every cutoff value. Depending on the
complexity of the grade distribution and the time available to model it, two
different approaches to variography may be taken: multiple indicator
kriging (MIK) and median indicator kriging (mIK). Multiple indicator
kriging uses different variograms for each cutoff value, whereas median in-
dicator kriging uses the same set of variograms (modelled at the median
grade) for all cutoffs.

Multiple Indicator Kriging

MIK requires a different set of variograms for every cutoff value, each of
which is matched to the corresponding cutoff. It is more accurate than
median indicator kriging but the number of variograms needed can make it
very time-consuming. A lack of data at high or low cutoffs can also make
variograms difficult to model.

As well as handling mixed-population grade data, MIK also handles varying

anisotropy and directions of continuity at different grade levels. A simple
example of this grade-orientation-anisotropy relationship is a shear-hosted
orebody that contains high-grade tension gashes. Because of the stresses
responsible for their creation the gashes will be skewed relative to the sur-
rounding shear zone. Thus, the variograms for the high grade gashes might
have different orientations and their anisotropy ratios might differ from those
of the lower grade shear mineralisation.

MIK easily handles higher grade values and eliminates the need to apply a
top cut. It is also used for mining grade control, although that is beyond the
scope of this manual.

Median Indicator Kriging

Median indicator kriging uses one set of indicator variograms modelled at

the median grade and applied to all indicator cutoffs. It is not as accurate
as MIK for highly mixed populations, but it is simpler and can still handle high
grades and dual populations. The indicator cutoff for median indicator kriging
need not be the median value; it may be at a real cutoff value between two
populations.

Setting up mIK in Micromine also takes place on the Multiple Indicator

Kriging dialog, the only difference being the use of one Variogram Form-
set for all indicator Cutoffs.

Choosing Data for IK Interpolation

Always use the flagged (and optionally composited) file for interpolation. For
an IK interpolation, use the uncut grade field as input: the indicator values
will perform what amounts to an automatic top-cut.

You’ll carry out median indicator kriging in the next exercise.

98 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Optional Exercise 9.1: Calculate mIK grade deciles Notes:

In addition to the median indicator-transformed variograms you created in

Optional Exercise 6.1, you must also define a series of indicator cutoff values
before carrying out mIK. For simplicity you’ll use the entire orebody and set
the indicators to simple grade deciles. In the workplace you should perform
these steps on a per-domain basis.

A quick way to find the deciles is to use the quick statistical summary:

1. Select Stats | Quick Summary from the main menu.

2. Set the File to Iron_Assay_2m_Comp and apply your All Ore Filter.
3. Set the first Field to T_Fe. Do not use TFe_CUT.
4. Disable Detailed Report.
5. Enable Quantile Report and Deciles, and disable any other options.
6. Enter a Report file of Iron_Assay_Deciles and then Run the
calculation.
7. Ignoring the 25th and 75th percentiles, note down the decile values in
this table (the first decile has been entered for you):

Decile 1 2 3 4 5 6 7 8 9
Value 27.70

The median indicator-transformed variograms and decile cutoffs (for this

exercise) are the only extra information needed for median indicator kriging.
Now that you have this information you can set up the initial mIK
parameters.

Grades corresponding to metal cutoffs are far more important for precious
commodities than iron ore and are not relevant here. However, for com-
pleteness the following advanced exercise teaches you to calculate metal
deciles.

Advanced Exercise 9.2: Calculate mIK metal deciles

Calculating metal deciles requires you to sort the assays, so you’ll begin by
copying the assay file to avoid corrupting the drillhole database. For brevity
you’ll use the entire orebody.

1. Open Iron_Assay_2m_Comp and then save it as Iron_Assay_Metal.

2. Modify the file and remove all fields except INT and T_Fe.
3. Sort the file on ascending grade ( T_Fe). Delete any blank rows from the
top of the file.
4. Using the Field Calculator, perform the following calculation to
accumulate the metal (the result fields will be created for you):

© Copyright MICROMINE 2014 99

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Input Function Input = Result

INT Multiply by T_Fe METAL
1 Accumulate METAL CUM_METAL

5. Scroll to the end of the file and copy the last value in CUM_METAL. This
is the total contained metal in the assay file. You’ll need it later.
6. Right-click anywhere in the CUM_METAL field and choose Colour Sets
| Edit from the pop-up menu.
7. Click the Clear button at top right of the colour grid.
8. Click the Calculate button at right and apply these settings to the
colour set:

Prompt Setting
First value: 0

Last value: [Paste the copied value]

Create using: NUMBER OF INTERVALS
Value: 10 (for deciles)

9. Select or create a colour palette with good contrast between adjacent

ranges and then save the colour set with the Title T_Fe Metal Deciles .
Save and Close the Edit Colour Sets dialog once you are done.
10. Scroll though the file and note the T_Fe value (not CUM_METAL) at the
first row of each new colour. This represents the cutoff value for each
decile. Write each value below (the first one has been entered for you):

Decile 1 2 3 4 5 6 7 8 9
Value 29.48

11. Compare the deciles to those of the previous exercise. Are there any
significant differences?

You can now use either group of deciles in the following exercises to set up
the mIK interpolation. They assume you have already carried out an IDW or
OK interpolation and the filters and data search parameters have been set
up. Refer to Lesson 8 if you have not already done so.

100 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Optional Exercise 9.3: Complete the first mIK Notes:

interpolation run

To set up the mIK interpolation:

Input Data

1. From the main menu, select Modelling | 3D Block Estimate |

Multiple Indicator Kriging to display the Multiple Indicator
Kriging dialog.
2. On the Input Data tab, set the Input File to Iron_Assay_2m_Comp.
3. Enable the Filter option and choose your DOMAIN = All Ore filter.

The input file contains the composited assay intervals from within the entire
orebody, and the filter restricts the data to the specific intervals needed for
the current domain.

4. Enable the Ignore blanks Numeric Exception.

5. Select Blocks from file and set the File to OBM_Iron_Blank, ensuring
all fields are correct.
6. Enable the Filter option and choose your DOMAIN = High-grade breccia
filter. Right-click | View the OBM_Iron_Blank file to preview it.
7. Enable Interpolate parent blocks only.

Modelling Parameters

1. Switch to the Modelling Parameters tab and change the Mode to

BLOCKS to enable block kriging.
2. Set Analysis field to T_Fe.
3. Right-click the Cutoff response and create a form set with the Title TFe
Grade Deciles or TFe Metal Deciles .
4. Starting from an initial Cutoff of 0, enter the decile indicator Cutoff
values you calculated earlier. Don’t enter a 10th indicator; Micromine will
simply use everything above the 9th value for the 10th range.

If preferred you can set the 10th indicator to the maximum value of the
assay data.

5. Set all Grade Estimate options to MEAN except for the 9th to 10th
Estimate, which should be set to MEDIAN.
6. Right-click the first Variogram Formset and create a new form set
with the Title mIK.
7. Set the Main direction to Axis 1 (Median =x) , the 2nd to Axis 2
(Median =x) , and the 3rd to Axis 3 (Median =x) , where x is the value
you determined earlier.
8. Use the same mIK form set for the remaining ranges, and close the
Model Values dialog once you are done.

© Copyright MICROMINE 2014 101

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Because you’re performing median indicator kriging you will use the same
indicator variogram form set for each cutoff. For multiple indicator kriging
you would define a different set of indicator variograms for each cutoff.

9. Click the Discretisation button and set the divisions to 4, 4, and 4 for
East, North, and Z, respectively.
10. Enable Set negative kriging weights to zero.
11. Set the Data search to the OK Run 1 form set you created earlier.
You’ll reuse this form set for the mIK interpolation.

In the workplace you should determine the Data Search radii and
orientations from a full suite of indicator variograms.

12. Enter a Count field name of HOLES, then click any other active
response on the form to activate the Count reference field and Min
count responses. Set these to HOLE_ID and 4, respectively.

Output

1. Switch to the Output tab and enable the first four Write… options,
which will help to indicate the quality of each estimate.
2. Enter an Output File name of OBM_Iron_MedIK.
3. Lastly, click the Add Fields button and define the following fields, which
you’ll use to store the run number and, ultimately, the resource
classification. Because this is the first run, set the Value to 1:

Field Name Type Width Decimals Value

RUN SHORT 1
CLASS CHARACTER 10 [blank]

Adding fields now is more efficient than modifying the block model later.

4. Save the main Multiple Indicator Kriging dialog as a form set with the
Title HG_BX Run 1.
5. Run the function to perform the first mIK run, and right-click | View
the output file once completed.

The blocks in the IRON_HG_BX domain that contain interpolated grades

could possibly classified as measured resources provided you had a detailed
knowledge of the project. You must now complete the remaining runs for
both domains before the model is complete.

102 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Optional Exercise 9.4: Complete the subsequent mIK Notes:

runs

To populate the remaining blocks using multiple interpolation runs, begin

with Run 2 for the IRON_HG_BX domain:

1. On the Input Data tab, change the (Blocks from file) File to
OBM_Iron_MedIK .
2. Switch to the Modelling Parameters tab and repeat Steps 11 and 12
from that part of Optional Exercise 9.3 using the Run 2 settings from
Table 9.1 (Re-use the matching OK Data Search form set):

Table 9.1: MIK search radii and minimum points requirements

Data Search Add
Min
Run Min Form Set Fields
Radius Factor count
Points Title Value
1 1 d70 4 OK Run 1 4 1
2 1 d85 3 OK Run 2 2 2
3+ 4 d85 Blank OK Run 3 1 3

3. Switch to the Output tab and change the Add Fields value to 2.
4. Save the Multiple Indicator Kriging dialog as a form set with the
Title HG_BX Run 2.
5. Re-Run the function.
6. Scan through the OBM_Iron_MedIK file and ensure that all IRON_HG_BX
blocks are populated with grade values.

With the IRON_HG_BX domain completed you can turn your attention to the
IRON_LG domain:

7. Switch to the Input Data tab and change the (Blocks from File)
Filter to your Low-grade halo form set. Right-click | View the
OBM_Iron_MedIK file to test the filter.
8. Switch to the Modelling Parameters tab and repeat Steps 11 and 12
from that part of Optional Exercise 9.3 using the Run 1 through Run 3
settings from Table 9.1 (re-use the OK form sets).
9. Switch to the Output tab and change the Add Fields value 1, 2 or 3,
respectively.
10. Save your changes to the Multiple Indicator Kriging dialog as form
sets with Titles LG Run 1 through LG Run 3.
11. Run the function after setting up each form.
12. Visually scan the OBM_Iron_MedIK file and ensure that all blocks are
populated with grade values.

Blocks interpolated in the first run could possibly be classified as measured

resources provided you had a detailed knowledge of the project. Similarly,
blocks interpolated by the second run could be classified as indicated re-
sources, and those interpolated by the third (and subsequent) runs could be
classified as inferred resources.

© Copyright MICROMINE 2014 103

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Use ordinary kriging if...

• Your data for each domain consists of one normally

distributed population with a low COV, and
• Proportional effect is not present, and
• Grade continuity and anisotropy can be assumed equal at
all grade ranges

Use ordinary kriging with relative semi-

variograms if...

• Proportional effect is present

Use median indicator kriging if...

• Low grades have different distribution and anisotropy to

high grades, or
• You do not have enough data for multiple indicator
kriging, or
• You do not have time to set up multiple indicator kriging

Use multiple indicator kriging if...

• Your data consists of mixed populations, or

• Grade distribution and anisotropy are different at
different grade ranges
• You need to handle high grade values without using a top
cut

Figure 9.1: Decision tree for choosing a kriging method

Unsupported Kriging Methods

Simple Kriging
Simple kriging differs from ordinary kriging in that it interpolates from a
known mean, which is assumed to be constant throughout the study area. In
comparison, ordinary kriging calculates the local mean of the samples that
fall within each search neighbourhood, and as a result this mean will vary
across the study area.

The known-mean requirement makes simple kriging unsuitable for most re-
source estimation situations and it is not supported by Micromine.

104 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Co-kriging Notes:

Co-kriging requires an inverse relationship of one element to another in the

model area such as nickel and magnesium.

Disjunctive Kriging
Disjunctive kriging is another name for co-indicator kriging.

© Copyright MICROMINE 2014 105

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: Lesson 9 Summary

This lesson has introduced the basics of median indicator kriging and multiple
indicator kriging. The topics covered are:

To perform universal kriging:

Use Ordinary/Universal Kriging, and

Set the Trend to LINEAR, QUADRATIC, or CUBIC as desired.

To perform log-normal kriging:

Use Ordinary/Universal Kriging, and

Set the Transformation to NATURAL LOG.

To perform ordinary kriging with relative variograms:

Use Ordinary/Universal Kriging, and

Use relative variograms.
DO NOT use the kriging variance and kriging standard error.

To perform median indicator kriging:

Use Stats | Quick Summary to calculate the median grade, and

Use Stats | Histogram to identify actual grade cutoff values, or Stats
| Quick Summary to calculate grade percentiles, then
Select Stats | Semi Variograms and model one group of indicator-
transformed variograms with the indicator set to the median value, and
Use Multiple Indicator Kriging, and
Create or load a Cutoff form set containing the required indicator
Cutoffs and the previously created Variogram Formset, using
The cutoffs defined above, then
Set up the remaining prompts as for ordinary kriging.

To perform multiple indicator kriging:

Use Stats | Histogram to identify actual grade cutoff values, or Stats

| Quick Summary to calculate grade percentiles, then
Select Stats | Semi Variograms and model a different group of
indicator-transformed variograms with the indicator for each group set
to one of the cutoff values defined above, and
Use Multiple Indicator Kriging, and
Create or load a Cutoff form set containing the required indicator
Cutoffs and the previously created Variogram Formsets, using
The cutoffs defined above, then
Set up the remaining prompts as for ordinary kriging.

106 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Lesson 9 Summary Notes:

Good Practice

Use multiple/median indicator kriging if:

 You can’t domain your study area into zones with normal grade
distributions.
 The degree of anisotropy or directions of continuity are different at
different grade levels.
 You need to handle high grade values without applying a top cut.

Use median indicator kriging if your data is sparse or if you don’t have
enough time to model separate indicator variograms at each cutoff value.
Use multiple indicator kriging if time and data permit.

Always use the flagged and optionally composited file for interpolation. Use
the uncut grade field as input: the indicator values will perform what
amounts to an automatic upper-cut.

Help Topics

For information on: See:

Median/mode Statistics > Quick Summary

The histogram Statistics > Histogram

Decomposing data Statistics > Histogram >

Calculations in the chart display >
Decomposing data sets

Multiple indicator Block Modelling > 3D Block Estimate >

kriging 3D Block Estimate: Kriging >
Multiple Indicator Kriging

Additional Resources and Further Reading

Clark, I. and Harper W.V. (2000). Practical Geostatistics . Geostokos (Ecosse)

Ltd.

Glacken, I.M. & Blackney P.C.J. (1998). A Practitioners Implementation of

Indicator Kriging. In Beyond Ordinary Kriging. Geostatistical Association of
Australia: Perth

© Copyright MICROMINE 2014 107

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Appendix 1: Answers to Selected

Notes:

Exercises
Exercise 2.1: Build an E-W experimental variogram

At 60 m E-W spacing:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 - 40 16 36 - 34 4 34 - 32 4
42 - 43 1 35 - 30 25 30 - 29 1
37 - 37 0 42 - 39 9 39 - 36 9
36 - 36 0 42 - 39 9 39 - 40 1
38 - 35 9 35 - 37 4 37 - 33 16
37 - 35 4 35 - 38 9 38 - 35 9
38 - 35 9 35 - 33 4 33 - 29 16
35 - 35 0 40 - 37 9 41 - 38 9
40 - 40 0 39 - 41 4 37 - 34 9
43 - 39 16 38 - 37 1 32 - 28 16
37 - 38 1 37 - 36 1 29 - 32 9

Sum of squares: 234

Number of pairs × 2: 66
Semi-variance: 3.5455

At 90 m E-W spacing:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 - 42 4 37 - 37 0 34 - 29 25
42 - 42 0 36 - 33 9 30 - 30 0
37 - 35 4 42 - 37 25 39 - 38 1
35 - 35 0 42 - 41 1 37 - 34 9
36 - 35 1 35 - 37 4 33 - 28 25
37 - 38 1 35 - 36 1 - -
38 - 37 1 35 - 32 9 - -
35 - 34 1 40 - 36 16 - -
37 - 30 49 39 - 40 1 - -
40 - 39 1 38 - 33 25 - -
43 - 39 16 37 - 35 4 - -

Sum of squares: 233

Number of pairs × 2: 54
Semi-variance: 4.3148

108 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

At 120 m E-W spacing: Notes:

Pair Diff-sq Pair Diff-sq Pair Diff-sq

44 - 40 16 37 - 37 0 30 - 32 4
42 - 39 9 36 - 32 16 - -
37 - 38 1 35 - 29 36 - -
35 - 37 4 42 - 36 36 - -
36 - 34 4 42 - 40 4 - -
38 - 30 64 35 - 33 4 - -
37 - 37 0 35 - 35 0 - -
38 - 38 0 35 - 29 36 - -
35 - 33 4 39 - 38 1 - -
40 - 37 9 38 - 34 16 - -
43 - 41 4 34 - 28 36 - -

Sum of squares: 304

Number of pairs × 2: 46
Semi-variance: 6.6087

Plotting lag vs. gamma all the way to the maximum possible lag of
240 m produces this result:

Exercise 8.1: Determine initial search radii using distances to reach

proportions of the total sill

Total sill: 148.2

70% of total sill (rounded): 105

85% of total sill (rounded): 125

d70 d85

© Copyright MICROMINE 2014 109

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes:
Axis 1 radii (rounded): 120 165

Axis 2 radii (rounded): 105 150

Axis 3 radii (rounded): 40 60

Optional Exercise 9.1:

Here are the iron grade deciles. Note that the 5th decile is the same as
the 50th percentile or median value:

Decile 1 2 3 4 5 6 7 8 9

Value 27.70 30.55 33.58 37.39 41.10 45.79 50.32 55.82 61.03

Calculating the metal deciles is slightly more complex but is made simpler by
being able to visually match them to the corresponding grades. The high-
lighted cell in this snippet from the Iron_Assay_Metal file shows the T_Fe
value of the 9th decile, corresponding to the first blue CUM_METAL record:

The resulting metal deciles are:

Decile 1 2 3 4 5 6 7 8 9

Value 29.48 33.50 38.08 42.02 46.42 50.74 55.24 59.15 62.64

In this project there are only minor differences between the grade and metal
deciles, but a precious metal deposit will reveal a very different result. Try it
with the Gold_Assay file. Around 90% of that data falls within the first metal
decile.

110 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

Appendix 2: The Iron Variograms

Notes:

This appendix explains how the supplied iron variograms were created using
2-metre composites within the combined high- and low-grade domains:

Nugget

1. A downhole variogram produced a good result with a nugget of 3.38.

Anisotropy

2. Omni-variograms suggested the best lag interval would be 50 m.

© Copyright MICROMINE 2014 111

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Notes: 3. A strike variogram map with a 50 m lag interval produced a reasonable

display, with a small number of weakly developed directions.

Continuity directions towards 65° and 335° correspond to drillhole align-

ments and do not represent genuine anisotropy. Other possible directions are
towards 55° and 115°, although they are only weakly developed.

4. Based on this finding a horizontal variogram fan was created as a

cross-check, with azimuths ranging from 0 to 180° in 5° increments.
5. The initial 50 m lag interval was found to be too small to produce good
variograms and was experimentally increased to 100 m.
6. Fan directions from 45° to 65° produced reasonable variograms (not
necessarily the best continuity), and a direction of 55° was chosen,
corresponding to one of the directions seen on the map.
7. The measured strike on the variogram map was set to 55°. Dip and
pitch maps were created with dip and pitch both set to around 10°.
The map parameters were saved to a variogram control file.

Three-axis Variograms and Model Fitting

8. Three-axis experimental variograms were displayed using the control

file. Axis 3’s lag interval was experimentally reduced to 12 m to account
for the downhole data.
9. Variogram models were fitted, with Axis 1 and Axis 3 (near-vertical)
given the highest priority.

Axis 1 was prioritised because it represents the direction of greatest

continuity. Axis 3 was prioritised for two reasons. Firstly, it is closely aligned
to the downhole direction and is supported by the most data. Secondly, it
pierces the sub-horizontal layering within the orebody, and must be well-
fitted to represent those layers in the final block model. Axis 2 was simply
fitted to the sills determined on Axes 1 and 3.

The large variation in range between the Axes 1 and 2 and Axis 3, along with
the zonal anisotropy and substantially lower starting sill of Axis 3 required
models to be fitted with four components. Component 1 (spherical) was

112 © Copyright MICROMINE 2014

Micromine 2014 (Version 15.0) Training MG 311 – Geostatistics and Kriging

used to reach the beginning of the Axis 1 and Axis 3 experimental vario- Notes:
grams. It has no real effect on Axis 2.

Component 2 (exponential) was a compromise to match the slightly curved

shape of Axis 1. Ideally Axis 3 should use a spherical model, but this is at
odds with the shape of Axis 1.

Component 3 (spherical) represents the final sills of Axes 1 and 2. It has no

effect on Axis 3.

Component 4 (spherical) was used to handle the zonal anisotropy

between Axes 1 and 2 and Axis 3. (Axis 3’s sill is higher because of mixing
between the high- and low-grade domains, and other horizontal layering not
represented by wireframe boundaries.) For Axes 1 and 2 the range of this
component was arbitrarily set to 100 000 m.

10. Variogram cross-validation was carried out using a search ellipsoid

aligned to the variogram axes with radii set to the long ranges. It
produced a good result with a mean error statistic of 0.001 and a
standard deviation of 1.22.

© Copyright MICROMINE 2014 113

MG 311 – Geostatistics and Kriging Micromine 2014 (Version 15.0) Training

Appendix 3 – Variogram Fan

Notes:

Workflow

114 © Copyright MICROMINE 2014