Supervisor Basic Tutorial

May 2020
 Supervisor Basic Tutorial

This document has been prepared by Snowden Mining Industry Consultants Pty Ltd (‘Snowden’) as a guide
to the basic functions of Supervisor software and not for any other purpose.

 2020

All rights in this document are reserved by Snowden.

Prepared by: Oscar Rondon

Principal Geostatistician

Reviewed by: Anthony Cook

Product Manager

Issued by: Perth Office

Doc ref: SPV Basic Tutorial v1
Last edited: 8/05/2020 10:58 AM

OFFICE LOCATIONS

PERTH JOHANNESBURG
Level 6, 130 Stirling Street Henley House, Greenacres
Perth WA 6000 Office Park, Cnr. Victory and
AUSTRALIA Rustenburg Roads, Victory Park
Johannesburg 2195
TEL +61 8 9213 9213 SOUTH AFRICA
ABN 99 085 319 562
PO Box 521278, Saxonwold 2132
[email protected] SOUTH AFRICA

www.snowdengroup.com TEL +27 11 782 2379

FAX +27 11 782 2396
REG 1998/023556/07

[email protected]
 Supervisor Basic Tutorial

Table of Contents

1 INTRODUCTION ................................................................................................................... 4

2 BASE FAMILIARISATION WITH THE SUPERVISOR ENVIRONMENT ............................... 5

2.1 Screen layout ..................................................................................................................... 5
2.2 Project window ................................................................................................................... 6

3 LOADING DRILL HOLE DATA INTO SUPERVISOR............................................................ 7

4 BASIC STATISTICS ............................................................................................................ 13

4.1 Histograms ....................................................................................................................... 13
4.2 Log Probability Plots ........................................................................................................ 15
4.3 Mean and Variance Plot .................................................................................................. 18

5 COMPARING DATA............................................................................................................ 21
5.1 Box-and-Whisker plots..................................................................................................... 21
5.2 QQ plots ........................................................................................................................... 23
5.3 Scatter plots ..................................................................................................................... 24

6 VARIOGRAPHY .................................................................................................................. 27
6.1 Modelling a variogram ..................................................................................................... 30
6.2 Indicator variography ....................................................................................................... 37

7 KRIGING NEIGHBOURHOOD ANALYSIS ......................................................................... 39

8 MODEL VALIDATION ......................................................................................................... 47

Figures

Figure 2.1 Supervisor software screen layout ............................................................................................ 5

Figure 2.2 Typical project hierarchy ........................................................................................................... 6
Figure 4.1 Histogram display example .....................................................................................................13
Figure 4.2 Log Probability plot display example ......................................................................................15
Figure 4.3 Mean and variance plot display example ................................................................................18
Figure 5.1 Box-and-Whisker plot display example ...................................................................................21
Figure 5.2 Box-and-Whisker plot display example – multiple domains ...................................................22
Figure 5.3 QQ plot display example .........................................................................................................23
Figure 5.4 Scatterplot display example ....................................................................................................25
Figure 6.1 Variogram ................................................................................................................................27
Figure 6.2 Variogram fan display example...............................................................................................28
Figure 6.3 Variogram display example .....................................................................................................29
Figure 7.1 KNA plot display example – KE/Slope ....................................................................................39
Figure 7.2 KNA plot display example – Negative Weights .......................................................................40

Draft 8 May 2020 PAGE 3

 Supervisor Basic Tutorial

1 INTRODUCTION
Supervisor software was developed by Snowden Technologies Pty. Ltd. (Snowden), a Perth-based
independent information systems solution provider for the mining industry. The platform on which is
based Ph-Xtend (X10) was developed by Phinar Software Pty Ltd.
Snowden Supervisor can be used to perform any part of the geostatistical analysis that you need to
perform on your data, reducing the requirement for individual software packages. At the same time,
however, Supervisor allows you to import models from other programs, maintaining compatibility with
existing software on the market, including other releases made by Snowden – Analysor, Visor and
Maximisor. Supervisor therefore is not only very powerful, but also extremely flexible – your existing
projects can simply be imported into a new Supervisor project.
The following features are common to all analysis tools available in Supervisor:
• The ability to import data from a wide variety of sources
• Multiple data files/variables per scenario
• Filtering data based on location or assay value
• Calculating and reporting of key statistics
• Customised grouping, allowing changes to be applied to all components in a group
• Automatic updating to reflect changes in key parameters
• Facilities to export in comma-separated values (CSV) or GeoEAS (GSLIB) format data
• Plotting on any device supported by the operating system
• Declustering assay types
Supervisor sets a new benchmark for high-quality resource estimation tools, by combining statistical
analysis, variography, estimation and conditional simulation into an integrated, high-speed, analysis
and visualisation platform.
This document is intended to show you the basic capabilities available within Supervisor, including data
importing, basic statistics, data comparisons, variography, Kriging Neighbourhood Analysis (KNA) and
model validation.
The document will demonstrate the ease of use and flexibility of the system for carrying out standard
geostatistical functions, with the main emphasis being on the practical application of the techniques
using Supervisor.
Prior experience with Supervisor is not essential for the use of this document, however it assumes you
are familiar with the basic principles of statistical analysis, variography and the modelling of variograms.
There is a specific set of data that accompanies this document and all examples are based on this data
set. This data is available upon request from Snowden.

DRAFT 8 May 2020 PAGE 4

 Supervisor Basic Tutorial

2 BASE FAMILIARISATION WITH THE SUPERVISOR

ENVIRONMENT
2.1 Screen layout
When you run the program, you will see the Snowden Supervisor Screen Layout, and a new Project
Window. You can then begin to import files and models to analyse in your project. Below is an
illustration of the Snowden Supervisor Screen Layout, and the various components you will see when
working on a Supervisor project (Figure 2.1).

Figure 2.1 Supervisor software screen layout

The control window consists of seven main areas. These are listed and explained below:
• Menu Bar: File, Edit, View, Update and Help menus are used to manage the Supervisor project.
• Toolbars: The Main Toolbar contains shortcuts to a number of commonly used functions. The
Display Toolbar allows the user to customise the display of data points or blocks in a model.
• Project window: Displays the components of the project organised into a tree structure.
• Display window: This is where the selected plot is displayed. In addition to the selected plot tab the
Display Window contains the 3D Viewer tab and the Results tab.
• Properties window: Displays the property options for the selected plot type.
• System window: Includes various tabs containing components that are relevant to any loaded
project.
• Status window: Displays errors, warnings and other update information.

DRAFT 8 May 2020 PAGE 5

 Supervisor Basic Tutorial

2.2 Project window

This section illustrates the main items you will see in the project window – namely, the loaded data and
the associated analysis components of your project. Snowden Supervisor displays the project in a tree-
like structure, where components at a lower level are dependent on those at a higher level of the
hierarchy. For example, an object placed directly under a domain is dependent on that domain and will
reflect the data contained in the domain. A typical project hierarchy is illustrated below in Figure 2.2.
Figure 2.2 Typical project hierarchy

DRAFT 8 May 2020 PAGE 6

 Supervisor Basic Tutorial

3 LOADING DRILL HOLE DATA INTO SUPERVISOR

Start the Supervisor software. Supervisor will automatically load the last project.
The procedure for starting a new project and adding data to the project is as follows:
1) Start a new project by left clicking on the New button and selecting Project.

2) Right-click on the Data tab at the top of the tree on the left-hand side of the screen. The top entry
"Data" is the Data Root Component. This tree will often contain analysis and visualisation
components for a specific data set. Select Add | Point Data | Text

An alternative way to add point data is to open the Add Data drop down on the main toolbar and
select Text. This menu will show the last used data loader commands.

3) After selecting the Text option, select a file with the File Open dialog.

DRAFT 8 May 2020 PAGE 7

 Supervisor Basic Tutorial

Note that you can select multiple file types. When prompted select the data_analysis.csv file and
click Open

DRAFT 8 May 2020 PAGE 8

 Supervisor Basic Tutorial

4) You will be presented with the Data Loader screen. The fields already ticked in the Import column
have been found by the data loading library and selected automatically. Make sure all fields
required for analysis are ticked in the Import column and to get the Mapping Type correct.

Make sure the grouping is set to Assay/Domain. Data grouping is a powerful feature that will
create a hierarchically ordered tree of data components. When set to Assay/Domain, Supervisor
creates a tree with all input assay names. The domain names appear underneath each assay
name and contain the value of the corresponding assay in the indicated domain. If set to
Domain/Assay, then a tree is created with all domain names and underneath are all the assay
names with the values of the corresponding assay in the domain.
5) Left click on the Load Data button on the top right-hand section of the screen.

DRAFT 8 May 2020 PAGE 9

 Supervisor Basic Tutorial

6) You will be presented with a Load Summary of your data which has a warning of 36 errors in the
input data (red box).

7) Click onto the red error count box and expand out the Load Error/Warning column to get details
of the errors.

DRAFT 8 May 2020 PAGE 10

 Supervisor Basic Tutorial

8) Open the input data in Microsoft Excel, filter out the #NUM values in the Au field and reset them to
absent values. Save the adjusted file as data_analysis_fixed.csv.

9) In Supervisor, return to the Data Loader tab and select your adjusted input file.

10) If any of the field names in your file have changed than left click Refresh Mapping to update the
Data Loader screen.

DRAFT 8 May 2020 PAGE 11

 Supervisor Basic Tutorial

11) Load the data and you should now be presented with the following Load Summary with a missing
count of 36 and no errors.

DRAFT 8 May 2020 PAGE 12

 Supervisor Basic Tutorial

4 BASIC STATISTICS
After loading the data, Supervisor automatically generates the histogram, log probability plots and mean
and variance plot for each domain.
A histogram is a plot of frequencies of values in a dataset. Histograms can be inserted under any set of
data which contains only one assay type. They can be applied to an assay type containing one or more
domains, a domain that is a subset of an assay type, or a domain which contains only one assay type.
Cumulative Distribution Functions, or CDF plots, are used to observe the distribution of the points in the
data set (data file or block model) and can be displayed in four different ways: Cumulative Distribution,
Log Cumulative Distribution, Probability and Log Probability. The type of plot is determined by the axis
scaling selected.
The mean and variance plot is used to assess the impact of applying a global assay top cut value. The
plot displays the mean and the variance of the drill hole data as function of different top cut values.

4.1 Histograms
The histogram will appear similar to the one shown below (Figure 4.1), with key statistics marked on the
graph and reported in a summary table. Note that the x-axis is labelled with the assay type, and the
units which you entered when the data was imported.
Since a histogram does not have an associated 3D representation, when viewing a histogram, the 3D
window will display the dataset used to create the histogram (assay type or domain).
Figure 4.1 Histogram display example

DRAFT 8 May 2020 PAGE 13

 Supervisor Basic Tutorial

1) Expand the tree and select the Histogram view for Domain 1. Note how the Assay/Domain setting
has organised the data in the tree.

2) Repeating this for the remaining domains allows to quickly have a look at the basic descriptive
statistics for AU in the different domains. Hold the Shift key down and click in any of the histogram
by domain to visually compare all histograms in a single view.

DRAFT 8 May 2020 PAGE 14

 Supervisor Basic Tutorial

3) A summary of the basic statistics can be exported to a csv file for all domains at once by right-
clicking on the AU tab at the top of the tree and selecting Export | Stats

4.2 Log Probability Plots

The Log Probability plot has Cumulative % on the y-axis, and Log (base 10) assay values on the x-axis.
The three coloured vertical lines are the Mean, Geometric Mean, and Log Estimated Mean values. The
three blue dashed lines represent the 25%, 50% (Median) and 75% thresholds. The standard log
probability plot is shown below in Figure 4.2. The Log (base 10) assay value transformation may be
deactivated in the CDF tab.

Figure 4.2 Log Probability plot display example

DRAFT 8 May 2020 PAGE 15

 Supervisor Basic Tutorial

1) Because of the positively skewed nature of the data set, a log histogram is more appropriate to
analyse the histograms. Select the Histogram view for any domain and tick the Log button in the
Histogram tab on the right-hand side of the screen to change the histogram to log scale. Select
the Log Probability Plot view for the Domain 6 data. Make sure to tick the Report stats option
under the Stats tab. Percentiles 25th, 50th (median) and 75th are reported and indicated in the log
probability plot along with the mean value.

2) Other percentiles can be computed by adding a threshold to the log probability plot. Right-click on
the probability plot for Domain 6 and add a threshold.

DRAFT 8 May 2020 PAGE 16

 Supervisor Basic Tutorial

Supervisor automatically adds the 10th percentile (yellow line). Note the threshold is reported as a
percentage and as an assay value (cut-off).

3) To automatically add more thresholds, right-click on the probability plot for Domain 6 and select
Add | Threshold Deciles

DRAFT 8 May 2020 PAGE 17

 Supervisor Basic Tutorial

4) The data deciles are automatically added to the probability plot and can be modified in the
Threshold tab. This is a useful feature when using Multiple Indicator Kriging (MIK) as will be
discussed in Section 6.2.

4.3 Mean and Variance Plot

The Mean and Variance graph plots linear assay values along the x-axis, mean values on the left y-axis
and variance values on the right y-axis. The coefficient of variation (CV) can be displayed on either y-
axis, depending on the scaling of these axes. The mean plots as red, variance as green, and coefficient
of variation as blue. An example mean and variance plot is shown below in Figure 4.3.

Figure 4.3 Mean and variance plot display example

DRAFT 8 May 2020 PAGE 18

 Supervisor Basic Tutorial

Note that the solid lines are the statistics below the cut-off, and the dashed lines are the statistics above
the cut-off. Since a mean and variance plot does not have an associated 3D representation, when
viewing a mean and variance plot the 3D window will display the dataset used to create the plot (assay
type or domain).
1) The plot displays the mean and the variance of the drill hole data as function of different top cut
values. It can be customised to display other information like the coefficient of variation (CV).

2) The selected top cut can be applied to the data by setting a filter. Click onto the domain level on
the tree (e.g. ‘1’) and apply a maximum value in the box on the right-hand side under the AU tab.
To apply a top cut put an @ symbol in front of the value, e.g. upper limit @30.

DRAFT 8 May 2020 PAGE 19

 Supervisor Basic Tutorial

All assays greater than 30 are reset to a value of 30. Therefore, the maximum assay value is now
30, as shown in the histogram below.

DRAFT 8 May 2020 PAGE 20

 Supervisor Basic Tutorial

5 COMPARING DATA
5.1 Box-and-Whisker plots
Box-and-Whisker plots provide a visual representation of the statistics and dispersion of a data set. An
example of a Box-and-Whisker plot is shown below in Figure 5.1.
The box shows the range between the two specified percentiles (in this case, 25% and 75%), with the
50-75% range shaded dark grey, and the 25-50% shaded light grey. The vertical lines span the total
range of the data. The horizontal blue (dashed) lines are the 25th, 50th and 75th percentiles, while the
red, green and blue solid lines are the mean, geometric mean and log-estimated mean respectively.

Figure 5.1 Box-and-Whisker plot display example

The previous illustration showed a box-and-whisker plot that was inserted under a single domain of an
assay. The plot shown below (Figure 5.2) was inserted at assay level, where the assay contained
multiple domains.
By inserting a box-and-whisker plot under a multi-domain assay (or similarly for a multi-element
domain), you can quickly compare the ranges and variability in each subset of the data.

DRAFT 8 May 2020 PAGE 21

 Supervisor Basic Tutorial

Figure 5.2 Box-and-Whisker plot display example – multiple domains

1) Compare the same assay across different domains in the data_analysis_fixed data set. Right-click
on the AU tab at the top of the tree and select Add | Box-and-Whisker-Plot. Make sure to change
the scale to log by ticking on the Show Log (base 10) box on the right-hand side.

DRAFT 8 May 2020 PAGE 22

 Supervisor Basic Tutorial

5.2 QQ plots
QQ plots are used to visually compare two distributions to determine their similarity. A straight line will
indicate identical distributions. An example of a QQ plot is shown below in Figure 5.3. Note that the two
data sets whose distributions are being compared are shown under the title to the plot. Dataset 1 is on
the x-axis, and Dataset 2 is on the y-axis.
Figure 5.3 QQ plot display example

1) Domains 1 and 3 in the data_analysis_fixed data set have the most data. Create a QQ plot to
compare these populations in more detail. Right-click on the Domain 1 tab in the tree and select
Add | QQ Plot

DRAFT 8 May 2020 PAGE 23

 Supervisor Basic Tutorial

2) Select Domain 3 as the second data set for comparison in the QQ plot.

3) Tick on the Log-scale box on the right-hand side to change the scale to log QQ plot for Au grades
in Domain 1 and 3.

5.3 Scatter plots

A scatter plot is a graph of the values of one variable against the values of a second variable. These
plots are used to show the relationship between two variables. When working with more than one grade
attribute it is useful to look at the scatterplot to determine if a relationship exists between the attributes.
A scatterplot directly compares paired data and the degree of scatter or spread of results on the graph
gives a visual indication of the correlation between the attributes. This information can be used to aid
domaining and to determine which attributes can be estimated inside the same domains and which will
require separate domains. It is also useful for variography as highly correlated attributes will have
similar variability and behave in a similar fashion.
An example of a scatterplot is shown below in Figure 5.4. The correlation coefficient and regression
equation are displayed under the title of the scatter plot. The regression line (line of least squares

DRAFT 8 May 2020 PAGE 24

 Supervisor Basic Tutorial

regression) is shown in green, and the dashed line represents the line y=x. By default, the mean is
indicated in red and the median (50% percentile) in blue.

Figure 5.4 Scatterplot display example

1) Load the data DHdata.csv into Supervisor. It contains FE2O3 and MGO grades from a nickel
laterite deposit.

DRAFT 8 May 2020 PAGE 25

 Supervisor Basic Tutorial

2) Right-click FE2O3 in the tree and select Add | Scatter Plot | MGO

3) The scatter plot should look like the figure below. The red lines on the graph correspond to the
mean grades of FE2O3 and MGO.

DRAFT 8 May 2020 PAGE 26

 Supervisor Basic Tutorial

6 VARIOGRAPHY
Spatial analysis or variography is carried out in order to understand how sample values relate to each
other in space, which can then help to infer possible similarities between known samples and points
that have not been sampled. A variogram is calculated by plotting the average variability for all sample
pairs at a certain distance apart against that separation distance.
Variograms can be calculated either purely by separation distance or by separation distance and
direction. If direction is not taken into account and all sample pairs at a given separation distance are
used in the calculation of the average variability, it is referred to as an isotropic or omni-directional
variogram. If direction and distance are used to select sample pairs, then it is referred to as a directional
variogram.
When calculating variograms, the separation distance is termed the lag or “h” (for example 10 m).
When calculating directional variograms, “h” refers to the distance and direction vector (for example 10
m north-south). The gamma symbol () is the standard symbol for variability in a variogram. On the
variogram we plot (h) being the average variability (or variogram value) of all sample pairs separated
by vector “h”. The average variability is calculated for a series of lags and plotted against lag distance to
create a variogram plot (Figure 6.1). The separation distance at which the sill is reached is called the
range or range of continuity and indicates the distance at which there is no longer correlation between
the samples.

Figure 6.1 Variogram

The normal scores variogram uses the traditional variogram equation but transforms the data to a
Gaussian or normal distribution before variogram calculation (also called a Gaussian variogram).
Transformation of data into normal scores or Gaussian data is used for variogram calculations of
positively skewed datasets (normal scores variograms).
For highly skewed populations, transforming the data prior to calculating the variogram can provide a
clearer view of ranges of mineralisation continuity as the transformation removes the influence of the
skewed tail of the population distribution. As a result of this process however, the variability is also
reduced and in order to obtain a final variogram model the normal scores model must be transformed
back into traditional space prior to use in estimation.

DRAFT 8 May 2020 PAGE 27

 Supervisor Basic Tutorial

Snowden Visor is the variogram module of Snowden Supervisor, which can be used to run variography
in various spatial orientations (2D and 3D). It provides interactive graphical tools for analysing,
interpreting and modelling the parameters applicable to kriging interpolation or conditional simulation.
In Snowden Visor, there are two major stages to analysing continuity. The first involves the analysis of
a series of fans, in the horizontal, across-strike vertical and dip planes. Each fan is made up of
variogram contours, allowing you to analyse and select the direction of maximum continuity in the fan.
Each direction is used to determine the placement of the next plane. At the second stage, the selected
strike, dip and plunge are used to locate the three directions for which variogram models will be
constructed.
To view a variogram fan in your project, click on the chosen plane (e.g. Horizontal) in the project tree.
The fan will look similar to the one shown in Figure 6.2.
The Fan window displays contours of standardised variance in the given plane. The purple arrow
indicates the current orientation of the protractor, which can be adjusted by clicking and dragging the
arrow to a new orientation. Note that the position of this protractor is used by Visor to calculate the fan
in the next plane of interest. The yellow bar underneath the fan allows you to modify the lag distance.
This is done by clicking on the bar and dragging it to the right to increase or the left to decrease the lag
distance. Unlike virtually all other software packages, changing the lag distance in Visor does not
require the continuity analysis to be recalculated. The scale bar displayed to the left of the variogram
fan shows the distance over which the fan is plotted. This distance can be changed by clicking and
dragging the scale bar. Labels are automatically inserted at twice the active radial spacing. This and
other settings that control the display of the Fan window can be modified in the Fan Tab of the
component properties window. For continuity analyses of a particular domain, the domain name is used
as a subheading of the variograms and any directional models that are produced.

Figure 6.2 Variogram fan display example

DRAFT 8 May 2020 PAGE 28

 Supervisor Basic Tutorial

Clicking on any ray in the fan will open the relevant variogram in the window. Viewing the variograms
associated with rays can help you decide on a preferred direction of continuity in the chosen plane.
When this option is activated, the Display window will change to show a variogram similar to the one
shown in Figure 6.3.
The variogram for each ray is made up of two graphs overlaid on the one plot. The red line graph is the
(standardised, if selected) variogram for the ray. The value to which the variance has been
standardised, if any, is printed in brackets against the label on the Gamma axis. The blue column graph
represents the number of pairs that have been used in the computation of the corresponding points in
the variogram. Either portion of the graph can be viewed in isolation.
When modelling a variogram, you need to first set the nugget, then apply a number of structures to the
variogram. Visor allows you to apply up to three nested structures of a range of types. The models used
are indicated and controlled by the use of a number of draggers (crosses) on the variogram.
Figure 6.3 Variogram display example

DRAFT 8 May 2020 PAGE 29

 Supervisor Basic Tutorial

6.1 Modelling a variogram

In this section a complete variography analysis (orientations of grade continuity and model variograms)
for data set variography data.csv is presented. The objective is to analyse grade continuity so the
variogram parameters can be used for grade estimation in any software that provides ordinary kriging
estimation.
Variography data.csv is a point file comprising drill hole samples for a single domain. The drill holes are
dominantly orientated -60º to the east and are located on a 5 m east-west by 10 m north-south grid.
There is around 1.5 m distance between samples downhole.
1) Load variography data.csv into Supervisor. Review the data in the 3D viewer and the various
statistics to determine the orientation of the domain, type of distribution and any potential issues
with outliers.

2) Right-click on the domain 1 tab on the tree and insert a continuity analysis for the horizontal plane
(strike) (Add | Horizontal Continuity Fan).

DRAFT 8 May 2020 PAGE 30

 Supervisor Basic Tutorial

3) Given that the population is positively skewed a normal scores transform is appropriate for the
variogram calculations. In the properties on the right-hand side set Transform | NormalScores
and click on the update button to start the variogram calculation.

4) Select the Fan + Variograms tab at the bottom of the screen to see the variograms associated
with each radial line on the fan.

It is worth noting that you can:

• Change the scale on the fan by dragging the scale bar on the left-hand side.
• Change the lag by dragging the yellow bar at the bottom.
• Change the angular tolerance and bandwidths in the properties on the right-hand side (under
the Update tab)
• Double click on a variogram to see one at a time
• Use the up and down arrows to change between variograms or click on a radial on the fan.

DRAFT 8 May 2020 PAGE 31

 Supervisor Basic Tutorial

5) Look at the Normal Scores Horizontal fan and variograms and determine the direction with the
longest continuity (longest range) ensuring to select an appropriate variogram type and tolerance
parameters. This is the strike of the mineralisation. On the fan, drag the pink arrow into the strike
direction to select it. Right-click on the Normal Scores Horizontal fan on the tree and insert a
continuity analysis for the across-strike vertical plane (dip) (Add | Across Strike Continuity Fan).

6) Look at the Across Strike Continuity fan and variograms and determine the direction with the
longest continuity (longest range) ensuring to select an appropriate variogram type and tolerance
parameters. This is the dip of the mineralisation. On the fan, drag the pink arrow into the dip
direction to select it. After doing so, right-click on the Normal Scores Across Strike fan on the tree
and insert a continuity analysis for the dip plane (plunge) (Add | Dip Plane Continuity Fan).

DRAFT 8 May 2020 PAGE 32

 Supervisor Basic Tutorial

7) Look at the Dip Plane Continuity fan and variograms and determine the direction with the longest
continuity (longest range) ensuring to select an appropriate variogram type and tolerance
parameters. This is the plunge of the mineralisation. On the fan, drag the pink arrow into the dip
direction to select it. If there is no plunge, then select the strike or dip as the direction of longest
continuity. The direction selected will become direction 1 when fitting the experimental variogram.

8) The experimental variogram is fitted by right-clicking on the Normal Scores Dip Plane fan on the
tree and inserting continuity models (Add | Continuity Models).

DRAFT 8 May 2020 PAGE 33

 Supervisor Basic Tutorial

9) This will give you a downhole variogram and 3 directional variograms in the major, semi-major and
minor directions as well as an axis rotation and back-transformed model.

10) The Downhole variogram can be set to any user-defined direction using the Dip -> Dip Direction
settings. Model the nugget using the Downhole variogram with the dominate drill hole orientation
as the direction of analysis. Ensure you select appropriate parameters and lag. The lag of the
Downhole variogram is generally the length of the composites.

If drill hole information has been loaded using a borehole or drill hole ID field then Supervisor
automatically computes the Downhole variogram along the drill holes.
Directional variograms can be modelled interactively with up to three different variogram models and a
nugget structure. The Auto Fit button, under the Model tab, applies the auto-fit algorithm to the
variogram. The auto-fit will honour the enabled structures and their structure types. It will determine the
best fitting sills (including nugget) and ranges. Although the auto-fit simply fits the current variogram it
may also change the sills (or ranges) of other directions if the sills (or ranges) are linked.

DRAFT 8 May 2020 PAGE 34

 Supervisor Basic Tutorial

11) You can see and model all variograms at once by clicking on Continuity Models in the data tree
and select the Variograms tab.

12) Once you are happy with the models, set the rotation axes to the appropriate rotation for your
software (Rotation Type).

DRAFT 8 May 2020 PAGE 35

 Supervisor Basic Tutorial

13) Right-click on Back Transform Model in the data tree to export the variogram parameters.

14) The variogram parameters can be exported in various formats for compatibility with your software.

DRAFT 8 May 2020 PAGE 36

 Supervisor Basic Tutorial

6.2 Indicator variography

To include a continuity analysis for an indicator in your project, you need to insert a CDF (cumulative
distribution function) plot under the data set. This plot is automatically inserted at domain level when the
assay data is imported, but not at assay type level. (This is because you will usually be primarily
interested in analysing the continuity within the identified domains, not across the domain boundaries,
which is what is done when the entire assay type is analysed.)
1) Insert the 50th percentile in the log probability plot of Au in the variography data.csv.

2) After setting the 50th percentile, insert the three continuity analysis fans as shown before to get the
indicator experimental variograms and model them.

DRAFT 8 May 2020 PAGE 37

 Supervisor Basic Tutorial

3) When using several indicators, insert the corresponding percentiles in the log probability plot, let’s
say 15th, 30th, 70th and 85th. Click on the 50% Horizontal continuity analysis and use the Ctrl+c
combination to copy the already computed 50th percentile experimental variogram and model.
Using the Ctrl+v combination, paste the continuity analysis in all other percentiles and update
Supervisor to get the experimental variograms for all of them. Model the directions and variogram
model parameter accordingly.

DRAFT 8 May 2020 PAGE 38

 Supervisor Basic Tutorial

7 KRIGING NEIGHBOURHOOD ANALYSIS

Kriging Neighbourhood Analysis (KNA) is a process for optimising estimation parameters, including
block size, number of informing samples, search range and the number of discretisation points.
Parameters are evaluated based on two conditional bias statistics:
• Kriging Efficiency (KE), which measures the effectiveness of the kriging estimate to reproduce the
local block grade accurately. A low kriging efficiency indicates a high degree of over smoothing.
Conversely, a low kriging efficiency indicates a low degree of over smoothing.
• Slope of regression (or conditional bias slope), which summarises the degree of over smoothing of
high and low grades. A slope close to 1 indicates that the regression between the estimated and
the actual grades is likely to be very good, meaning there is limited over smoothing. Conversely,
low slope values indicate that there is over smoothing and hence a poor relationship between the
estimated and the actual block grades.
A KNA is performed by calculating the KE and Slope of regression for varying combinations of
estimation parameters. The resulting graph is then used to help in selecting which parameters result in
the least over smoothing.
In Supervisor, a KNA component may be added to any Continuity Models component. By default, the
KNA will use the search ellipse defined by the Continuity Models. the Params tab can be used to view
and edit the block centroid, search angle and range, samples count and discretisation parameters. The
Blocks tab is used to define the different block sizes to evaluate, as well as to begin the KNA
refinement.
The KNA graph displays both the Kriging Efficiency (KE) as a percentage on the left-hand Y-axis, and
the Slope of regression as a value no greater than 1 on the right-hand Y-axis (Figure 7.1).

Figure 7.1 KNA plot display example – KE/Slope

DRAFT 8 May 2020 PAGE 39

 Supervisor Basic Tutorial

The negative weights associated with the KE and Slope can also be viewed, by selecting the Neg
Weights tab (Figure 7.2). The Sum of negative weights is shown on the left-hand Y-axis and the count
of negative weights on the right-hand Y-axis.

Figure 7.2 KNA plot display example – Negative Weights

To minimise the estimation error, it is important to ensure that you use an optimal block size and
optimal estimation parameters. While common sense and following a few sensible guidelines can aid
this, the conditional bias statistics can help quantify the effectiveness of the estimation process.
In what follows, a step by step KNA is carried out for illustration purposes only on how to do a KNA in
Supervisor.
1) Right-click on the Continuity Models for Domain 1 on the tree and add a KNA (Add | KNA).

DRAFT 8 May 2020 PAGE 40

 Supervisor Basic Tutorial

2) Select the block sizes to be tested. To remove any unnecessary block size, click on the blank
space of the corresponding row. When highlighted in blue, use the Delete key to remove it.

3) Click on the Params tab to indicate the location(s) to be used. A single location can be specified by
indicating the block centroid at the top. Alternatively, go into the 3D viewer and select the KNA tab.
Use the drag bars to adjust the location of the block centroid. Multiple locations can also be used
by enabling the multi-block KNA option. In this case, a tolerance distance away from the data (for
each direction) is used to restrict the blocks and avoid excessive extrapolation beyond the data.
Minimum and maximum coordinates specify the coverage of the blocks.

DRAFT 8 May 2020 PAGE 41

 Supervisor Basic Tutorial

4) After setting the parameters, hit update to run the KNA. A screen as shown in the figure below will
show the results for all the KNA scenarios for the different block sizes. Note that positioning the
cursor over any Box and Whisker plot provides a quick summary of the conditional bias statistic.
Furthermore, on the second tab at the bottom, the number and sum of any negative weights
generated during the estimates can also be reviewed. After analysing the results, the appropriate
block size can be selected by clicking in either the kriging efficiency or regression slope Box and
Whisker Plot. To deselect the scenario, click again over any of the Box and Whisker plot.

DRAFT 8 May 2020 PAGE 42

 Supervisor Basic Tutorial

5) Select the 15x15x2 block size and then right-click to proceed with the analysis of the maximum and
minimum number of samples (KNA | CreateKnaSamples). The KNA samples graph will
automatically be added.

6) Set the ‘Min’ sample, ‘Max’ sample and Step to calculate the regression slope and KE statistics.
The Samples group is used to control the variation in the minimum and maximum number of
samples for each scenario. The minimum defines the smallest number of samples to be evaluated.
The maximum defines the largest number of the samples to evaluate. The Step value determines
the increment between each successive KNA scenario. The number of cases to be evaluated is
given by the difference between the maximum and the minimum divided by the step value.

DRAFT 8 May 2020 PAGE 43

 Supervisor Basic Tutorial

7) Update to get the KNA results for the different number of samples. Select the minimum number
required and the maximum number required and select Refine Search to bring up the KNA search
parameters graph.

8) Select the scenarios with five samples (minimum) and 20 samples (maximum) by clicking in the
corresponding Box and Whisker plots or in the data tree. Proceed as before to continue with the
KNA of the search parameters (KNA | CreateKnaSearch).

9) Multiple search ellipsoids can be tested at the same time by entering multiple scenarios. Note the
number of informing samples will be based on your previous selection, however this can limit your
results as the results will level when you reach the maximum number of samples. It is possible to
adjust the maximum number of samples to review this impact.

DRAFT 8 May 2020 PAGE 44

 Supervisor Basic Tutorial

10) Update to get the KNA results for the different search ellipsoids.

11) Select the scenario corresponding to a search ellipsoid of dimensions 10x10x5 and proceed to the
final KNA stage which is analysing the impact of the block discretisation. Update to get the results.
To show a summary of the KNA results, tick the Show Summary box.

DRAFT 8 May 2020 PAGE 45

 Supervisor Basic Tutorial

In summary, the result of the KNA is:

• Block size of 15x15x2
• Search dimensions: 10x10x5
• Minimum and maximum samples in the search neighbourhood: Five and 20 respectively
• Block discretisation: 3x3x3

DRAFT 8 May 2020 PAGE 46

 Supervisor Basic Tutorial

8 MODEL VALIDATION
The validity of grade estimates is subject to every decision made and parameter used throughout the
resource estimation process. There are multiple points at which things can go wrong and hence it is
essential to validate the final model to ensure that the estimated grades reflect the input information.
Currently Supervisor is the only major software package that has a model validation component. It is
simple to set up and use even for multiple domains and/or multiple elements.
Several standard validation checks can be carried out in Supervisor which include:
• Global statistics validation (statistical comparison table)
• Population distribution validation (validation histogram, CDF plot and QQ plot)
• Global trend validation (trend/swath plot)
• Validation grade-tonnage (GT) Curve, which also supports a Global Change of Support (GCoS)
function.
In what follows, a step by step model validation in Supervisor is carried out for illustration purposes
only.

1) Load in the file trmod.csv. Make sure to select block data since it is a block model that is being
loaded and set the grouping to Assay/Domain. Important: The block model Assay/Domain
grouping must match the grouping on the point data, otherwise the model validation will not work.
With variable block sizes, load the block dimension fields and map them to XINC, YINC and ZINC.
Variable density can be mapped to DENSITY.

DRAFT 8 May 2020 PAGE 47

 Supervisor Basic Tutorial

2) Your Supervisor file should mirror the figure below. The block model and the assay data have been
loaded into one Supervisor window.

3) Right-click on Domain 1 and add a model validation component (Add | ModelValidation). This
must be done under the point data, since there is no menu for model validation under block data.

The model validation component includes:

• The trend/swath plots along the X, Y and Z as well as the strike and across strike directions
according to the variogram model fitted
• Histogram of block estimates and assay values
• Log probability plots for block estimates and assay values
• QQ plot for comparing the distribution of block estimates and assay values
• Global GT curves comparison to the expected theoretical results
• Summary of global statistics comparison

DRAFT 8 May 2020 PAGE 48

 Supervisor Basic Tutorial

4) Use the Data 1 tab to enter the details of the model and assay required for the model validation.
There are three Data source tabs. Multiple models can be validated (compared) at the same time
but there can only be one drill hole data source. If block dimensions have not been loaded, set the
default block size under XINC, YINC and ZINC to allow tonnage weighting. Likewise, DENSITY
field can also be set a default.

5) Update to get the Model Validation results and graphs.

DRAFT 8 May 2020 PAGE 49

 Supervisor Basic Tutorial

6) Click on the bottom right graph to enable the grade tonnage comparison with the parameters as
shown in the figure below. Under the GT curve tab, set the cut-offs to go from zero to two with
increments of 0.2.

7) Under the Draw tab untick the Point support to display only the comparison of the theoretical
(SMU) and block model GT curve at the same support.

DRAFT 8 May 2020 PAGE 50

 Supervisor Basic Tutorial

8) The Statistical Comparison Table contains a comparison of block model and drill hole data that
can be easily exported to a CSV file or copied into a document. To do this, right-click on the top left
part of the table.

9) Model Validation QQ plots are used to visually compare block model estimates with naive and
declustered data to determine their similarity. A straight line indicates identical distributions. Note:
Block model estimates are weighted by the configured volume and density values.

DRAFT 8 May 2020 PAGE 51

 Supervisor Basic Tutorial

10) A Model Validation Histogram plots the frequency of different assay values in the underlying
dataset(s). The histogram overlays the plot of the parent sample assay data against the block
estimate. Note: Block model estimates are weighted by the configured volume and density values.

Now that the model validation is set up, the model validation component can be copied and pasted to
all of the other domains then updated. This saves a lot of time when there are multiple domains and/or
multiple elements.
If data needs to be declustered this can be done by adding a Decluster under the AU tab (Add |
Decluster); this way the declustering will filter down to all of the domains. Alternatively, if separate
domains require different declustering grids, then the declustering can be inserted under each of the
domains.

DRAFT 8 May 2020 PAGE 52

 Supervisor Basic Tutorial

DRAFT 8 May 2020 PAGE 53