|
|
Uniform Conditioning - Decluster Creating more evenly spaced sample data |
Uniform Conditioning - Decluster
To access this dialog:
The Decluster screen is part of the Uniform Conditioning wizard and is used to uniformly distribute sample points as an aide to the Uniform Conditioning process.
Declustering is the process of weighting data points in densely sampled areas, to obtain a more representative distribution of grades. This phase of the process will create a temporary points file (as indicated) containing declustered sample information.
There are three primary user options:
- Sample declustering has already been performed and declustered weighting values have already been calculated and included in the input sample data file (e.g. as an output from Studio's DECLUST process) - in this situation, the input samples file (as specified on the Input Data panel) will already contain a numeric field containing declustered values for each data row. Selecting this option, and the predefined declustered data field, will ensure these values are fed into the Uniform Conditioning process that follows. In effect, these values will be used verbatim.
- Sample selection from a 3D grid - samples are already dispersed across a regular grid and you wish to select a single sample from each grid cell. An equal weighting (in effect, a weighting of '1') will be applied to all samples.
- Decluster samples to calculate a weight - where you wish assign every sample a target weight based on the number of samples in the grid cell, which is defined manually in terms of size and origin. This option makes use of the Studio DECLUST process (find out more about this process...).
If the Declustered Weight method is chosen then the weight, DCWEIGHT, for a sample is calculated as:
DCWEIGHT = NDATA / NCELLS / NPERCELL
where:
-
NDATA is the total number of samples
-
NCELLS is the number of grid cells containing one or more samples
-
NPERCELL is the number of samples in the grid cell
The sum of the weights over all samples equals the total number of samples (NDATA). Therefore if a sample lies in a high density area it will have a weight of less than 1, and if it is in a low density area it will have a weight of more than 1. The output file from the Declustered Weight method can be used to transform data into a normal distribution, for input to the NSCORE or SGSIM processes.
This stage of the process will model the declustered multivariate sample histogram in the form of a declustered samples data file. This is used in the next stage, creating variograms and global grade/tonnage curves.
How to Calculate Dispersion Variance
Dispersion variance (Variance of Z*) is an approximation of the variance of estimated block grades.
Using COKRIG or advanced estimation, Dispersion variance can be calculated directly by setting the field “Variance of Z*. This is calculated when field VARZSTR is set in the COKRIG field parameter file.
Dispersion variance can also be calculated from the Kriging Variance, Block Variance (BLOCKVAR) and Lagrange parameter(LAGRANGE); where DISPVAR= BLOCKVAR – VAR + (2*LAGRANGE).
Field Details:
The Decluster panel contains the following fields..
Samples are already declustered: select this option where your input Samples data file (as specified on the Input Data panel) already contains a data column including declustered weighting values for each record in the table. You will need to select the weighting field using the drop-down list below if you intend to use this option - values will be copied to the output declustered samples file.
Samples are on a regular grid...: samples are already dispersed across a regular grid and you wish to select a single sample from each grid cell. An equal weighting (in effect, a weighting of '1') will be applied to all samples.
Decluster samples to produce declustered weight: with this option, you will need to define a regular 3D grid that will be used to calculate the declustered weight values for each data row, based on the number of samples contained within each grid cell. One of the problems with the declustering method is that different grid sizes will generate different statistics. However, in general a regular grid about the size of the average sample spacing is suggested.
Run: create the output declustered samples file based on either the values in the input samples file, a static weighting value or the calculation of weighting as performed by the DECLUST process.
Declustered samples are stored in: the name of the temporary file created that contains declustered sample information, ready for use in generating global Grade Tonnage curves.
