A new worldwide trend
Mineral potential assessment via integration of
spatially georeferenced data is an approach that is now used on
a global scale, and it keeps generating more and more interest.
The growing number of articles dealing with this technique confirms
this. This new trend was strongly influenced by the development
of new software making the process possible on a personal computer,
and by the availability of new digital databases.
Formerly, zones favourable to the discovery of new deposits were
generally identified by overlapping several layers of polyester,
each illustrating a geological feature potentially indicating the
presence of a given type of ore deposit (fault, geophysical anomaly,
geochemical anomaly, etc.). Target areas were selected among those
showing the largest number of overlapping features. The digital
integration of georeferenced data revolutionized this approach,
by making it possible to target, with pinpoint accuracy, the type
of data (or theme) to process, to adapt the processing based on
data-driven or knowledge-driven methods,
and finally, to combine the results according to the relative importance
of each theme.
The various methods used to process the data are grouped into two
categories: first, data-driven methods that rely on
statistical analysis to determine relevant spatial associations
of various geological, geophysical or geochemical elements with
known ore deposits; second, knowledge-driven methods
which, based on the expert assessment of the processing geologist,
rely on the definition and spatial processing of relevant metallogenic
parameters for a selected model.
Data-driven methods use a training set generally derived
from a digital map where a sufficient number of known deposits corresponding
to the selected ore deposit model are located in the study area.
The processing consists in calculating the statistical correlation
between this training set and a certain number of digital maps each
representing a parameter (or theme) relevant to the model (map of
felsic intrusions, iron alteration, synvolcanic dykes, etc.). Data-driven
methods essentially use three techniques: regression analysis, neural
networks and weight of evidence.
Knowledge-driven methods (used to produce MPMPS maps)
are based on an ore deposit model that is relevant for the study
area. This model is based on geological parameters (heat source,
favourable lithologies, faults allowing fluid circulation, etc.)
which are considered as essential or favourable elements for the
formation of deposits. To these parameters are also added discrete
indicators (mineral occurrences, alteration indicator minerals,
geochemical anomalies, etc.) signalling the presence of mineralization.
In each case, these elements constitute a modelling theme. Data
processing is guided by the expert’s evaluation of the role
and importance of each theme in the emplacement of potential deposits.
The weight attributed to the various themes and their combination
may be derived either from the arbitrary assessment of an ore deposit
expert, or from combination methods based on concepts such as fuzzy
logic, or the Dempster-Shafer belief theory.
Data-driven methods that require a significant number
of occurrences related to the selected model can only be used in
areas that have already been extensively explored. The advantage
of these methods lies in the fact that they are based on a factual
approach that may be demonstrated and verified. However, they are
based on occasionally complex statistical correlations which, from
a geological standpoint, are not always meaningful. Knowledge-driven
methods are more suitable in poorly explored areas where little
or no favourable mineralization has been recognized to date. Since
the evolution of the product is guided by the expert judgment of
a metallogenist, this method yields a product that closely resembles
ore deposit models already known and studied by exploration companies.
On the other hand, the arbitrary weight attributed to parameters
in this approach is determined by the expert preferences of the
metallogenist, which may vary from one expert to another.