Overview

This section describes the options under the Grade Estimation submenu. Vulcan caters for the following Grade Estimation packages:

  • Gslib - This is written by Andre G. Journel and Clayton V. Deutsch, Department of Applied Earth Sciences, Stanford University, California.
  • Geostokos - Access to the Geostokos toolkit (developed by Dr Isobel Clark) is handled through an interface. This interface allows you to generate files from Vulcan composite sample data so that they can be used by the toolkit.

The range of estimation techniques that is provided with the above packages include:

  • Inverse Distance - This technique uses the inverse of the sample distance to the estimation point to weigh the samples.
  • Simple kriging - This technique takes a standard mapfile and, using the information derived from a variography study and a geological interpretation of the ore body, uses classical simple kriging to interpolate grade values to each of the nominated cells of a block model.
  • Ordinary kriging - This technique takes a standard mapfile and, using the information derived from a variography study and a geological interpretation of the ore body, uses classical kriging to interpolate grade values to each of the nominated cells of a block model.
  • Indicator kriging - This technique splits the grades into a series of ranges of "cuts". Estimation is then performed for each of the cuts; the full grade picture is then built up from an average of these cuts.
  • Stochastic Simulation - This technique calculates block values in a way that provides information about the variability of the sample data and the uncertainty in block estimates. Stochastic simulation does not store the average grade in a block, instead it stores a random value from the distribution of grade values.

Factors Controlling Runtime

The main factors controlling the runtime of grade estimation using inverse are:

  • the number of sample points inside the search radius.
  • the number of sample points per estimate.
  • the number of discretisation points.

The main factors controlling the run time of grade estimation using Simple kriging, Ordinary kriging, Indicator kriging and Stochastic Simulation are:

  • the number of sample points inside the search radius.
  • the number of sample points per estimate. For ordinary kriging or indicator kriging the time spent solving the kriging equations is proportional to the cube of the number of points.
  • the number of discretisation points.
  • the number of variogram structures. Each structure must be evaluated between all pairs of sample points.
  • the number of indicator cutoffs (for indicator kriging). Each cutoff requires the solution of a system of kriging equations. However, if two cutoffs use the same variogram parameters, the weights for one cutoff are reused for the next cutoff.

At the time of the estimate other values such as the average sample distances and the number of samples used can be recorded.

Parallel Estimation

Parallel grade estimation utilises multiple processors or multiple cores to accelerate the run time of grade estimation. When the environment variable VULCAN_THREADING is defined, parallel grade estimation is turned on. The value of the variable should be set to '2' for two threads on a dual core processor, or to '4' on a four quad core processor.

Note:   Multithread estimation requires clean data. Duplicated samples are not allowed.

Tie Breaking

When parallel grade estimation is enabled, the tie breaking algorithm is changed. Occasionally, when estimating a block, two samples have the same distance from the block centre. This means that the estimation program must make an arbitrary choice of which sample to consider first. This choice can have a significant effect on the resulting grade estimation because the second sample may be excluded from the estimation. A sample may be excluded because the maximum number of samples has been reached, or maximum number of samples per octant or per hole have been reached.

When parallel grade estimation is turned off, the tie breaking algorithm uses a pseudo random order based on the sequence of blocks previously estimated. When parallel grade estimation is turned on, the blocks are estimated in a different order. This causes the pseudo random order of sample selection to be different in some cases.

To provide consistent tie breaking with parallel grade estimation, the grade estimation program implements a tie breaking approach based on the block centre coordinates. This produces a tie break independent of the order of the blocks. The details of selected samples can be viewed through the Explain option, or with the '-x' flag when using the djbmest external program.

Debugging Information

If you want to see more details on how a block was calculated, then you can turn on debug printing. Debug printing generates a file that lists the samples used for a block, the kriging matrix (if ordinary kriging) and the weights.

To use debug printing, you need to run grade estimation from the command line. Enter the following command:

source $VULCAN/.maptek_setup_login

followed by

$VULCAN/bin/exe/djbmest <proj><bfi>.bmf <proj><name>.bef <parset> -x <blockid>

where

<proj> Project
<bfi> Block file identifier
<name> Estimation parameter file name
<parset> Name of parameter set in the .bef file.
<blockid> Block identifier (block number)

Example:  $VULCAN/bin/exe/djbmest abcmain.bmf abczone1.bef PARAM1 -x 17552

where abcmain.bmf is the name of the block model; abczone1.bef is the name of the parameter file; PARAM1 is the name of the parameters set in abczone1.bef. We are asking to dump block 17552.

Use the Inquire option (under the Block > Viewing submenu) to find the block ID for a particular block, or the '-I' switch when using the BEXPORT external program. The resulting debug file is named <proj><name>.bef_<parset>_explain.