Page 295 - Geochemical Anomaly and Mineral Prospectivity Mapping in GIS
P. 295
298 Chapter 8
TABLE 8-V
Two models of discriminant functions for predictive mapping of epithermal Au prospectivity,
Aroroy district (Philippines) based on training sets each with nearly balanced numbers of proxy
deposit-type and non-deposit locations. Values in bold represent predictor variables, per set of
spatial evidence, with strong positive spatial associations with the training deposit-type locations.
Discriminant analysis using training set A 1 Discriminant analysis using training set B 2
(Wilks’ lambda = 0.192; α=0.0001) (Wilks’ lambda = 0.172; α=0.0001)
Predictor Function coefficients Predictor Function coefficients
3
3
variables Standardised Unstandardised variables Standardised Unstandardised
NNW1 0.239 0.092 NNW1 0.244 0.088
NNW2 0.512 0.017 NNW2 0.727 0.024
NNW3 0.554 0.021 NNW3 0.713 0.026
NNW4 0.326 0.009 NNW4 0.537 0.014
NW1 -0.229 -0.006 NW1 -0.221 -0.006
NW2 0.094 0.003 NW2 0.069 0.002
NW3 -0.071 -0.002 NW3 -0.069 -0.002
NW4 -0.047 -0.001 NW4 -0.080 -0.002
FI1 1.497 0.043 FI1 1.282 0.037
FI2 1.006 0.028 FI2 0.939 0.025
FI3 0.467 0.013 FI3 0.467 0.012
FI4 -0.017 -0.001 FI4 -0.044 -0.001
ANOM1 0.524 0.013 ANOM1 0.517 0.013
ANOM2 0.163 0.004 ANOM2 0.233 0.006
ANOM3 0.113 0.003 ANOM3 0.145 0.004
Constant - -3.017 Constant - -3.323
1 Consists of 79 randomly-selected proxy deposit-type locations (Fig. 8-8) and 81 non-deposit
locations with lowest predicted mineral occurrences (from set 2 non-deposit locations; Figs. 8-4
2
and 8-7B). Consists of 86 coherent proxy deposit-type locations (Fig. 8-8) and the same 81 non-
3
deposit locations in training data set A. Statistically significant predictor variables in the
discriminant models (see ‘class code’ columns in Table 8-IV for explanations of variable names).
variables in the discriminant model based on training set A are not so much higher than
the standardised function coefficients of the same predictor variables in the discriminant
model based on training set B. These results mean that, on the one hand, the
contributions of the ‘FI’ predictor variables are more important to the discrimination
between the randomly-selected proxy deposit-type locations and non-deposit locations in
training set A than to the discrimination between the coherent proxy deposit-type
locations and non-deposit locations in training set B. This implies that any of the proxy
deposit-type locations chosen randomly has stronger spatial association with
intersections of NNW- and NW-trending faults/fractures than to either of these
individual sets of faults/fractures. On the other hand, in addition to the contributions of
the ‘FI’ predictor variables, the contributions of the ‘NNW’ variables are more important
to the discrimination between the coherent proxy deposit-type locations and non-deposit