Page 277 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS

280 Chapter 8

between deposit-type locations and geological features as depicted by variations in
values of wC ji (see equation (8.2)) of C ji classes of individual X i spatial evidential
features with respect to D. These artifacts, which represent systemic (or procedural)
errors in predictive modeling of geo-objects, are manifested as abrupt (or ‘noisy’) rather
than gradual variations in the values wC ji because some data classes coincide with many
deposit-type locations, some data classes coincide with few deposit-type locations and
some data classes do not coincide with any deposit-type locations. This problem is
intrinsic not only in the application of data-driven modeling with EBFs but also in the
application of weights-of-evidence (WofE) modeling (see Bonham-Carter, 1994, pp.
319-321). In WofE modeling the problem is overcome by calculating values representing
wC ji based on cumulative classes of data. However, in accordance with the theory of
evidential belief (Dempster 1967, 1968; Shafer, 1976), classes of data are treated as
discrete and independent bodies of evidence so that equations (8.8) and (8.9) apply to
data-driven estimation of EBFs of non-cumulative classes of data. Nevertheless, non-
noisy variations of data-driven EBFs can be achieved by performing model calibration
experiments with different class intervals of data. Let us illustrate this with some
examples in the case study area.
In one example, the integrated PC2 and PC3 scores obtained from the catchment
basin analysis of stream sediment geochemical data (see Chapter 5, Fig. 5-12) are
classified into a map of classes of integrated PC2 and PC3 scores, which is then crossed
with (or overlaid on) the map of locations of epithermal Au deposits (Fig. 8-11A). In one
calibration experiment, the integrated PC2 and PC3 scores are classified into more-or-
less 10-percentile intervals (Fig. 8-11B). Only one class does not coincide with any
epithermal Au deposit location. However, because the class intervals (i.e., number of
class pixels) are not constant (which is common in raster-based analysis with a
somewhat coarse pixel size), the graph of the values of Bel versus the upper limits of
classes of integrated PC2 and PC3 scores (Fig. 8-12A) is somewhat noisy, indicating that
the data-driven EBFs are somewhat improperly calibrated. In a second experiment, the
eight classes of more-or-less 10-percentile intervals in the first experiment are re-
classified into four classes of more-or-less 20-percentile intervals (Fig. 8-11C). The new
curve of the values of Bel versus the upper limits of classes of integrated PC2 and PC3
scores is smooth (Fig. 8-12A), which indicates that the data-driven EBFs in the second
experiment are properly calibrated. This is so because the new curve of the values of Bel
versus the upper limits of classes of integrated PC2 and PC3 scores shows that the
epithermal Au deposits are associated spatially with high integrated PC2 and PC3 scores
and that the threshold score separating anomalous and background integrated PC2 and
PC3 scores is about 0.3. This result is consistent with the result of the distance
distribution analysis of the spatial association between the integrated PC2 and PC3
scores obtained from the catchment basin analysis of stream sediment geochemical data
(see Chapter 5, Fig. 5-12) and the known locations of epithermal Au deposits in the
study area (see Chapter 6, Figs. 6-12E and 6-12F).