Page 299 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

302 Chapter 8

Comparing and contrasting the performance of the map of discriminant scores in Fig.
8-22A with the map of integrated Bel in Fig. 8-19A, both of which are created using 86
coherent proxy-deposit-type locations, show the following. It is apparent that the former
is better than the latter because, if 40% of the case study area is considered prospective,
then the map of discriminant scores in Fig. 8-22A delineates correctly 100% of the
training coherent proxy deposit-type locations and 100% of the testing deposit-type
locations (Fig. 8-22B), whereas the map of integrated Bel in Fig. 8-19A delineates
correctly 93% of the training coherent proxy deposit-type locations and 85% of the
testing deposit-type locations (Fig. 8-19B). However, If 10-20% of the case study area is
considered prospective, then the map of integrated Bel in Fig. 8-19A delineates correctly
35-65% of the training coherent proxy deposit-type locations and 39-54% of the testing
deposit-type locations (Fig. 8-19B), whereas the map of discriminant scores in Fig. 8-
22A delineated correctly 20-54% of the training coherent proxy deposit-type locations
and 42-50% of the testing deposit-type locations (Fig. 8-22B). If 5% of the study area is
considered prospective, then the map of integrated Bel in Fig. 8-19A delineates correctly
26% of the training coherent proxy deposit-type locations and 31% of the testing
deposit-type locations (Fig. 8-19B), whereas the map of discriminant scores in Fig. 8-
22A delineated correctly 10% of the training coherent proxy deposit-type locations and
23% of the testing deposit-type locations (Fig. 8-22B). Therefore, because mineral
prospectivity mapping aims to constrain the sizes of exploration targets in order to
increase the chance of mineral deposit discovery, the cross-validation results indicate
that the map of integrated Bel in Fig. 8-19A is a better predictive model of epithermal
Au prospectivity in the case study area compared to the map of discriminant scores in
Fig. 8-22A.
The poorer model performance of the map of discriminant scores in Fig. 8-22A
compared to the map of integrated Bel in Fig. 8-19A can probably be ascribed to the use
of training data sets with (almost) equal numbers of deposit-type locations and non-
deposit locations in the application of LDA. In contrast, note that the data-driven
estimates of EBFs are based on all non-deposit locations (see equations (8.8b) and
(8.9b)). In addition, Skabar (2005) demonstrated that, in contrast to the findings of
Brown et al. (2000) and Porwal et al (2003a), using a training set of known deposit-type
and all known non-deposit locations optimises the performance of artificial neural
networks in data-driven modeling of prospectivity. Further experiments, explained in the
following paragraphs, were performed in order to show that the arguments of Skabar
(2005) for the application of artificial neural networks to data-driven modeling of
mineral prospectivity are also valid for the application of LDA to data-driven modeling
of mineral prospectivity.
The part of the study area with data for all predictor variables consists of 9719 unit
cells (each measuring 100×100 m). Thus, training data set A is modified to training data
set AA, which now consists of 79 randomly-selected proxy deposit-type locations and
9640 non-deposit locations; whilst training set B is modified to training set BB, which
now consists of 86 coherent proxy deposit-type locations and 9633 non-deposit