Page 305 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

308 Chapter 8

1991; Reddy et al., 1992; Chinn and Ascough, 1997; Chinn, 2006). GES- and AI-based
techniques for mineral prospectivity mapping are beyond the scope of this volume.
Readers are referred, nonetheless, to Yatabe and Fabbri (1988, 1989) for review of
geoscience applications of AI methods and to Pan and Harris (2000, pp. 422-438) for a
review of expert systems and AI methods applicable to mineral exploration.

DISCUSSION AND CONCLUSIONS
There are various bivariate and multivariate methods of GIS-based data-driven
modeling of mineral prospectivity and these methods are now mostly well-established.
This chapter has demonstrated two methods for GIS-based data-driven modeling of
mineral prospectivity (one bivariate method, data-driven evidential belief modeling, and
one multivariate method, linear discriminant analysis) and two techniques supporting
these methods in order to properly create and integrate predictor maps in determining
new targets for further exploration of undiscovered occurrences of mineral deposit of the
type sought in a study area. These two latter techniques each address the issues of (a)
objective selection of a suitable unit cell size for data-driven modeling of mineral
prospectivity and (b) selection of coherent deposit-type locations for data-driven
modeling of mineral prospectivity. In addition, a brief review of cross-validation
strategies in data-driven modeling of mineral prospectivity has been provided here.
Until now, the choice of a suitable unit cell (or pixel) size [denoted as N(•)] for
raster-based-GIS data-driven modeling of mineral prospectivity has been subjective.
Point pattern analysis (Boots and Getis, 1988; Rowlingson and Diggle, 1993) of the
range of distances in which there is zero probability of one neighbour deposit-type
location situated next to another deposit-type location is useful in deriving a preliminary
set of choices for a suitable N(•). Then, using this range of distances, the analysis of the
rate of increase in the ratio [N(D)] : [N(T)–N(D)] as function of equal-interval change in
N(•) is robust, regardless of the number of deposit-type locations and the size of a study
area, in determining the most suitable N(•). Although these propositions have not been
demonstrated through formal testing, the empirical spatial associations between deposit-
type locations and indicative geological features in the case study area as quantified via
data-driven estimation of EBFs (this chapter) and as quantified via the distance
distribution method and distance correlation method (Chapter 6) are very similar. On the
one hand, the pixel size used in the spatial association analyses in Chapter 6 is, in fact,
10 m. This small pixel size was necessary in the analysis in Chapter 6 because we know
(or have learned from Chapter 4) that spatial measurements (e.g., distances) are fractals
so that an infinitesimally small pixel or unit cell is necessary for accurate spatial
measurements. On the other hand, the pixel size used in all the experiments of data-
driven modeling of mineral prospectivity presented in this chapter is 100 m, as
determined objectively via the proposed techniques. A pixel size of 100 m is a practical
spatial representation of the epithermal Au deposit occurrences in the case study area,
whilst a pixel size of 10 m is not. Nevertheless, the proposed techniques for objective