You are on page 1of 41

Multiple Indicator Kriging (MIK)

Example Using Studio RM


Introduction
• MIK was introduced by Journel in 1983 and is now a widely used grade estimation
method.
• A significant feature of the method is that it is non-parametric; it does not depend
on the distribution of the samples and can therefore be used with mixed sample
populations. It also helps it to reduce the over-estimation effect of very high values.
• The sample data is transformed into sets of 0 (below cutoff) / 1 (above cutoff)
indicators for a series of cutoff grades. The indicators are interpolated into the block
model to estimate the grade distribution for each model block from which the MIK
estimate can be made.
• The theoretical basis of MIK is both elegant and simple. The same processes are
used as for a standard Ordinary Kriged (OK) estimation study.
• The most popular method for interpolating the indicators is OK. However Studio
also allows you to estimate indicators by Simple Kriging, Nearest Neighbour and
Inverse Power of Distance.
• These slides provide an example of an MIK study. Data and a macro are available.
MIK Pros and Cons
The advantages of Multiple Indicator Kriging are:
• non parametric ie it does not depend on the histogram of the samples
• transforming the data to indicators is an effective way of limiting the effect of
outliers. This is particularly useful in gold deposits where a single high grade can
result in many high grade block estimates if OK is used.
• in general the variograms are better behaved ie smoother and therefore easier to
model
• the method provides an estimate of the recovered grade and tonnes for each cutoff
• The method is easy to understand!

The disadvantages of Multiple Indicator Kriging are:


• an MIK study will take longer than an OK study because you need to calculate
variograms and fit models for each cutoff, although there are methods to reduce this
• the recovered grades and tonnes cannot be related to a specific size of SMU
• order relation problems need to be addressed
• there is no theoretical kriged variance for the MIK estimate
The MIK Method
• MIK requires a set of between about 8 and 15 cutoff grades which are applied to the samples
• It is beneficial to have more cutoffs at the high grade end of the distribution as this has the
most significant effect on the resource estimate
• A set of 0/1 values are calculated for each cutoff depending on whether each sample is below
(0) or above (1) the cutoff
• Variograms are calculated from the indicator values of each cutoff and models are fitted
• The models are reviewed; sometimes an acceptable approximation can be achieved by using
the model corresponding to the sample median for all cutoffs – median indicator kriging
• The indicators are then estimated into the block model using ordinary kriging so that each
block has an indicator estimate of between 0 and 1 for each cutoff
• The estimated block indicators can be interpreted in two ways:
• The probability that the grade of the block is above the specified cutoff
• The proportion of the block above the specified cutoff
• The average grade between adjacent pairs of cutoffs is estimated from the samples
• The MIK estimate can then be calculated from these average grades and the block indicators
• The proportion and grade above cutoff can also be calculated for each block
Calculating the MIK Estimate for a Single Cell
Cut Proportion > Cut Proportion Within Range Av. Grade in Range Grade > Cut
0 1.0
0.15 1.0 5.45 (e)
2 0.85
0.25 3.5 6.24 (d)
5 0.60
0.30 5.75 7.38 (c)
6.5 0.30
0.20 8.0 9.0 (b)
9.5 0.10
0.10 11.0 11.0 (a)
Cut Grade Above Cutoff Calculation g/t
a 9.5 Options include median, mean, etc 11.0
b 6.5 (0.10 * 11.0 + 0.20 * 8.0) / (0.10 + 0.20) 9.00
c 5.0 (0.10 * 11.0 + 0.20 * 8.0 + 0.30 * 5.75) / (0.10 + 0.20 + 0.30) 7.38
d 2.0 (0.10 * 11.0 + 0.20 * 8.0 + 0.30 * 5.75 + 0.25 * 3.5) / (0.10 + 0.20 + 0.30 + 0.25) 6.24
e 0.0 (0.10 * 11.0 + 0.20 * 8.0 + 0.30 * 5.75 + 0.25 * 3.5 + 0.15 * 1.0) / (0.10 + 0.20 + 0.30 + 0.25 + 0.15) 5.45
MIK Estimation Example

The main steps in this example are:


• Select 11 cutoff grades (QUANTILE)
• Calculate experimental variograms for each cutoff (VGRAM)
• Compare variograms for the cutoffs (CHARTING | VARIOGRAMS)
• Fit a model variogram to the median cutoff (CHARTING | VARIOGRAMS)
• Set up parameter files (ESTIMATE)
• Estimate indicators and calculate MIK estimate (ESTIMATE)
• Validate the results eg compare with OK estimates
Example Data and Macro
The following files are available in zip file MIK_EG.ZIP:
• DM Files
o COMPS_5.DM – drillhole composites, approximately 5m
o OREMOD4.DM – geological model

• Macros Files
• MIK_EG.MAC – 4 macros to pre-process DM files and run MIK
• OK_EG.MAC – 3 macros to pre-process DM files and run OK for comparison with MIK
• MIK_RESOURCE.MAC – calculate grade and tonnes above cutoff, method 1
• MIK_RESOURCE_2.MAC – calculate grade and tonnes above cutoff, method 2

The slides show how to set up the parameter files and krige the indicators using the
ESTIMATE Wizard. Macros M3 and M4 in MIK_EG.MAC cover the same functionality as
ESTIMATE but use database processes rather than the customised wizard interface.
Calculate Cutoffs
• An MIK study requires between about 8 and 15 cutoffs
• The cutoffs need to cover the full range of sample grades
• It is particularly important to have good coverage at the high grade end of the distribution as this has the most
significant effect on the resource estimate
• A common method of selecting cutoffs is to use quantiles of the sample distribution
• In this example the QUANTILE command has been used to select deciles (every 10%) as the primary division and
then to split the top bin (>90%) into 4 sub-bins
• Then 11 cutoffs corresponding to quantiles for 10%, 20%, … ,90%, 95%, 97.5% have been selected for the cutoffs

• Quantiles for 10%, 50% (median) and 97.5% are highlighted

As an example the 10%, 50% (median) and 90% quantiles are highlighted in the above tables.
The “Maximum Grade” values are then the 1st, 5th and 9th cutoffs.
A total of 11 cutoffs are used in this example.
Indicator Variograms for 5m Composites
Directional Variograms for Median Cutoff – 4.5 g/t

Omni-Directional Variograms for All Cutoffs

Unadjusted Normalized
Charting | Variograms
ESTIMATE – Files | Input
ESTIMATE – Files | Output
ESTIMATE – Search Volumes | Shape
ESTIMATE – Search Volumes | Category
ESTIMATE – Search Volumes | Decluster
ESTIMATE – Variogram Models | Rotation
ESTIMATE - Variogram Models | Structures
ESTIMATE – Estimation Types | Attributes

1st Cutoff
ESTIMATE – Estimation Types | Ind. Est.

1st Cutoff
ESTIMATE – Estimation Types | Attributes

2nd Cutoff
ESTIMATE – Estimation Types | Ind. Est.

2nd Cutoff
ESTIMATE – Estimation Types | Attributes

11th Cutoff
ESTIMATE – Estimation Types | Ind. Est.

11th Cutoff
ESTIMATE – Controls | Parameters
ESTIMATE – Preview | Settings
ESTIMATE – Progress Messages (1/2)
ESTIMATE – Progress Messages (2/2)
Output Model – Data Definition
Output Model – Records 1 - 17

Subset of fields
Model Section
• The graphic shows a plan through the model OREMOD5 at an elevation of 460m
• The model is coloured by PRAB3 – the proportion of each cell above a cutoff of 2.9g/t
Output Average Grades File
File: AVG_GRADES

The “Estimation Types | Indicator Estimation” tab on the ESTIMATE dialog included the selection:

Therefore the grade of each bin (BINGRADE) has been set equal to the mean of the samples within the
bin (SAMPMEAN) except for the top bin (BIN 12) where the grade of the bin is set equal to the median
of the samples within the bin.
Output Indicators File
File: INDICATORS

• The file contains all the fields from the samples file plus a 0 /1 indicator field (PRABi) for
each cutoff i
• The indicator is 1 if the grade (AU) is greater than the ith cutoff or zero if it is below
• Only a subset of the fields is shown in the table above
• The file could be used to display the samples in the 3D window, with a legend based a
PRABi value
Parameter Files
The standard Search Volume and Estimation parameter files are output by the ESTIMATE dialog.
They can be used for rerunning ESTIMATE or for running the MIK process INDEST.
Search Volume Parameter File

Estimation Parameter File

Variogram Model Parameter File


Command INDEST for MIK
• The easiest way to set up the parameter files for MIK is to use the ESTIMATE dialog as
shown in the previous slides

• Once the parameter files have been set up the estimation can be done using command
INDEST and run from a macro as shown in M4 in file MIK_EG.MAC

• The files, fields and parameters for INDEST are shown below
Resource Estimation
1. Full Cell Estimation
• The output model includes field AU_IK_5. This is the MIK estimate of each cell.
Calculating the resource for a set of cutoffs based on this value assumes that the parent
cell size is the Selective Mining Unit (SMU). The whole cell is either above or below cutoff.

2. Partial Cell Evaluation


• The output model also includes the proportion (PRABi) of each cell above the ith cutoff and
the grade (GRABi) of the material above cutoff. Calculating the resource based on these
values assumes that the cell can be selectively mined, separating the material into above
and below cutoff. This implies a very small SMU.

Run macro MIK_RESOURCE.MAC to calculate and compare the resource based on


both these methods.

Calculating the resource for specific sizes of SMU can be done using Uniform
Conditioning and Conditional Simulation. Both are available in Studio RM.
MIK Resource Comparison Table
The results from macro MIK_RESOURCE.MAC are in file RES_COMPARE.

METHOD 1 – Full Cell Evaluation


METHOD 2 – Partial Cell Evaluation
MIK Resource Comparison Charts
• The results from macro MIK_RESOURCE.MAC are in file RES_COMPARE.
• The results show the tonnes and grade above cutoff for a range of cutoffs.
• The file should be loaded into memory and input to Charting | Point/Line Plot
Ore Tonnes v Cutoff
Legend:
• Method 1 – Full Cell SMU – Blue
• Method 2 – Partial Cell SMU – Red
• Ore Tonnes x 1000

Method 2 is more selective so gives less


tonnes for cutoffs below the mean grade
(5.2 g/t) and more tonnes for cutoffs
above the mean grade.
Grade Above Cutoff v Cutoff
Legend:
• Method 1 – Full Cell SMU – Blue
• Method 2 – Partial Cell SMU – Red
• Grade Au g/t

Method 2 is more selective so a higher


grade for all cutoffs
Statistics for MIK and OK
Macro file OK_EG.MAC contains macros for estimating AU using Ordinary Kriging.
An analysis of the MIK and OK results is shown below

Compare OK and MIK estimates


• Scatter Plot
• Statistics weight by cell volume

As expected the scatterplot shows a high


correlation between the MIK and OK estimates.
However it can be seen that the OK estimates
are lower than MIK for low grades and higher
for high grades. This is also shown in the
statistics below. The reduction in high grade
estimates is a feature of MIK which is one of the
reasons it is particularly appropriate for sample
distributions with high anomalous values.
MIK Papers
Numerous papers have been written on MIK. A few of these are listed below.
• Journel , A G, 1983. Nonparametric estimation of spatial distributions, Math. Geol Vol
15, No. 3, p. 445-468.
• Carvalho, D & Deutsch, C, 2017. An Overview of Multiple Indicator Kriging,
GeostatisticsLessons.com
• Glacken, I M & Blackney, P C J, 1998. A practitioners implementation of indicator
kriging, The Geostatistical Association of Australasia, “Beyond Ordinary Kriging”
Seminar
• Al-Hassan, S & Boamah, E, 2015. Comparison of Ordinary Kriging and Multiple
Indicator Kriging Estimates of Asuadai Deposit at Adansi Gold Ghana Limited, Ghana
Mining Journal, Vol. 15, No. 2, pp. 42-49.
• Keogh, A & Moulton, C, 1998. Median Indicator Kriging – A Case Study in Iron Ore,
The Geostatistical Association of Australasia, “Beyond Ordinary Kriging” Seminar.
• Jones, I, 1998. A Case Study Using Indicator Kriging – the Mount Morgan Gold-Copper
Deposit, Queensland, The Geostatistical Association of Australasia, “Beyond Ordinary
Kriging” Seminar

You might also like