In a cellular 3d grid model, various properties can be modeled (also named Common Earth Model). Each cell stores a property vector, which is obtained either by direct property measurements at sampling locations, or estimation, or simulation. Cells representing exploration targets, like orebodies, can feature characteristic property patterns and spatial relationships, like proximity to faults.
A probabilistic
approach can provide a quantitative
objective measure of the mineral potential based on data and thereby aid the expert
in finding target locations. Three data sources can be considered:
i) expert knowledge, ii) known occurences in the CEM, and iii) occurence
data from similar geological situations and exploration model (may be
stored in a GIS or borehole database).
Several Methods have been successfully used for GIS-based mineral potential
favourability analysis, namely Weights of Evidence, Boolean and Fuzzy
Logic, Logistic Regression. Predict currently implements Weights of Evidence as the most well-known
and widely used Bayesian methods.
A case study has been conducted by Martina Boehme using a comprehensive data set from the famous Noranda VMS district.
» view pdf slides (presented at the MiraGeosciences Gocad Meeting, Montreal 2006)
Documentation
0. Data analysis
Predict Mode - Explore bivariate data - Pearson
CorrCoef
Predict Mode - Explore bivariate data - Spearman
Rank CorrCoef
Remarks. These functions work for all object types with
properties (Voxets, PointsSets, ...). Result will be print to the history
and status bar. Note: a conditional independence test is always carried out during Weights of evidence computation.
1. Weights of Evidence Method
Predict Mode - Weights of Evidence - Create
evidential properties.
This command provides functionality for data preparation.
The WofE implementation requires a set of binary properties
which we have to generate from our numerical and categorical property values. We require
a Voxet with a set of evidential properties as input data. From each property, a new property (property
name + "newPropertySuffix") is generated. Here the value "0" means that the property anomaly is absent, "1" means anomaly occurs, "2" means no data available. The checkbox
parameters "useMin, minVal useMax, maxVal" define the tresholds to compute the property. For example, if "useMin=true", then all cells with property
values > minVal will be assigned the value "1". If you check both useMin and useMax an interval is defined.
If you wish to use categorial variables like rock types, you can check the box Categories and input the integer code values. For example, if you are looking for sediment-hosted deposits, you may select here all sedimentary rocks. All cell with these values will the be assigned the value "1", else "0".
If you have only partial data coverage in your Voxet, then you can work just in a Region, which is set to "everywhere" by default. Outside the model region the cells will be assigned the value "2" - these will not be used for further computations.
Predict Mode - Weights of Evidence - Create evidential properties from quantiles.
If you want to
consider multiple classes of one property
you have to create one binary property for each class. This can be cumbersome - so you might better use this dialog and define a number quantiles to compute a set of treshold values and corresponding evidential properties. The resulting properties contain the and treshold value quantile number in their name, e.g. Cu_val0.523_q2.
Predict Mode - Weights of Evidence - Posterior
probability from training data
This will compute the posterior probability property which represents the probability of a deposit to occur based on the evidential properties and a training region which represents known
occurrences in a Voxet.
The weight statistic and the result of the New Omnibus test for conditional independence between evidential properties will be saved as a HTML-document in your
Gocad project directory and also written as Info to the Gocad history.
Recommended introductory reference for the
weights of evidence method:
Bonham-Carter, G.F. 1994: Geographic Information Systems for
Geoscientists. Elsevier Science, 398pp.
Predict Mode - Weights of Evidence - Posterior probability from known weights
In case that the Voxet does not contain enough known occurrences for the computation of reliable weights, it is possible to use estimated (expert) weights, or pre-computed weights from other CEM or GIS databases with similar geological setting. Here, the posterior probability of an occurrence is computed from an overall prior probability and given weights assigned to a set of evidential properties.
