Page 10 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

Predictive Modeling of Mineral Exploration Targets 5

Predictive modeling is relevant to mapping of geochemical anomalies and
prospective areas because these real-world systems are complex and indirectly
observable. The objective of predictive modeling of geochemical anomalies and/or
prospective areas is to represent or map them as discrete spatial entities or geo-objects,
i.e., with conceivable boundaries. This entails various forms of data analysis in order to
derive and integrate pieces of information that allow description, representation or
prediction of geochemical anomalies and/or prospective areas. The term representation
or prediction is equivalent to mapping, which embodies results of analyses and
interpretations of various relevant geoscience spatial data sets according to hypotheses or
propositions about geochemical anomalies and/or prospective areas. Therefore, a map of
geochemical anomalies and/or prospective areas is a predictive model of where mineral
deposits of interest are likely to exist.

Approaches to predictive modeling

There are two approaches to predictive modeling – induction and deduction.
Induction is the process of making generalisations about particular instances in a set of
observations or data. A generalisation is derived by studying patterns in a data set.
Deduction – the inverse of induction – is the process of confirming particular instances
based on a generalisation about patterns in a set of observations or data. The
confirmation of particular instances is made by testing a generalisation against every
observation or datum. The distinction between induction and deduction can be illustrated
via a hypothetical example in Fig. 1-1 (cf. Bonham-Carter, 1994, pp. 180). An initial
visual analysis of the distributions of Fe contents in soil vis-à-vis the distributions of
lithologic units (Fig. 1-1A), may lead one to hypothesise that soils in areas underlain by
basalts have higher Fe contents than soils in areas underlain by other lithologic units.
The hypothesis may be supported by a simple linear model of Fe contents generally
decaying with distance from the basalt unit (Fig. 1-1B). The map of spatial distributions
of Fe contents and the simple linear model may further lead one to make a generalisation
that basalts influence Fe contents in soils and that Fe data in soils are useful aids to
lithologic mapping. Up to this point, one has performed an induction because a
generalisation was made based on particular instances in a set of observations or data.
To test the generalisation, because there is always an ‘exception to the rule’, one has to
perform deduction. To do so, one may use the simple linear model to predict Fe contents
in soil as function of distance to basalt and then map the spatial distributions of residual
Fe contents (Fig. 1-1C). The presence of enriched Fe in soils (i.e., positive residuals) in
certain parts of the area may lead one to hypothesise that there are unmapped basalt
units. Confirmation of this hypothesis requires re-visiting the sample sites (i.e., every
particular instance), which could result in updating of the lithologic map (Fig. 1-1C).
Induction and deduction are therefore complementary to each other, such that
switching from induction to deduction or vice versa at intermediate steps in predictive
modeling could provide better description, understanding and discovery of the system of
interest. Thus, despite the approach in the preceding hypothetical example, it is not
necessary to initiate predictive modeling with induction. The evolution of scientific