Page 306 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

Data-Driven Modeling of Mineral Prospectivity 309

selection of a suitable unit cell size for spatial representation of deposit-type locations
requires further testing in GIS-based data-driven modeling of mineral prospectivity.
The experiments of data-driven modeling of mineral prospectivity presented in this
chapter show that using coherent deposit-type (or proxy deposit-type) locations results in
better predictive models than using just any or randomly-selected deposit-type (or proxy
deposit-type) locations. The results presented here, therefore, show further the usefulness
of the two-stage methodology proposed by Carranza et al. (2008b) and explained here
for objective selection of coherent deposit-type (or proxy deposit-type) locations. The
results of the experiments presented here also show that using not just any but coherent
proxy deposit-type locations (i.e., unit cells immediately surrounding a unit cell
representing a deposit-type location) is useful in cases where the number of occurrences
of mineral deposits of the type sought is considered and/or found insufficient to derive a
proper (e.g., statistically significant) data-driven model of mineral prospectivity.
The experiments presented here were tested through a combination of N–n and
deposit-type classification strategies for cross-validation of predictive models of mineral
prospectivity. In practice, one must adopt or adapt a cross-validation strategy depending
on factors like (a) number of known occurrences of mineral deposits of the type sought
in a study area, (b) relevant attributes of deposit-type locations (e.g., grade, tonnage,
mine status, etc.) and (c) spatial (geological) coherence of different parts of a
moderately- to well-sampled mineralised landscape under investigation. The cross-
validation strategy adopted or adapted must enable comparison of predictive models,
derived via either one method or different methods, in order to determine the best
predictive model. Finding the ‘best’ predictive model of mineral prospectivity in the case
study area was, however, not an objective of this chapter. The cross-validation strategies
adopted here were, nonetheless, necessary in order to demonstrate the utility of coherent
(proxy) deposit-type locations in data-driven modeling of mineral prospectivity via at
least two different methods – evidential belief modeling and linear discriminant analysis.
The methods for data-driven evidential belief modeling of mineral prospectivity
presented here are relatively new, whilst the techniques for linear discriminant analysis
for mineral prospectivity mapping were first demonstrated more than three decades ago.
The latter techniques are able to handle both quantitative and qualitative predictor
variables, whilst the former method is able to handle qualitative predictor variables
(although some of these variables are derived via classification of quantitative variables).
Because evidential belief modeling and discriminant analysis are very different methods,
a scheme of spatial representation of qualitative evidential data (Fig. 8-20) was devised
so that (a) the same predictor variables used in evidential belief modeling can be used in
linear discriminant analysis and (b) testing of the utility of coherent (proxy) deposit-type
locations using the two different methods is objective. However, the scheme of spatial
representation of qualitative evidential data presented here is not only useful for the sorts
of analyses performed in this volume. In fact, Carranza and Castro (2006) employed this
scheme of spatial representation of qualitative evidential data in the application of
weights-of-evidence modeling, evidential belief modeling and logistic regression for