Prediction of Occurrence of Discrete Events - Geomathematics: Theoretical Foundations, Applications and Future Developments - page 140

igneous intrusives, aeromagnetic data, relative age, and topographic elevation. The
Kolmogorov-Smirnov test is used for testing goodness of fit.

Keywords Discrete event prediction • Weights-of-evidence method • Contrast
• Missing data • Weighted logistic regression • Meguma terrain gold deposits
• Greater Toronto area flowing wells • Gowganda gold occurrences • Pacific rise
volcanic vents • Conditional independence test

5.1 Weights-of-Evidence Modeling

In the preceding chapter, stepwise multiple regression analysis was used to estimate
a probability index for occurrence of large copper deposits in the Abitibi area on the
Canadian Shield. The input for explanatory variables primarily consisted of infor-
mation on rock types systematically quantified for cells measuring 10 km on a side.
Geophysical field data at cell centers were used for gravity (Bouguer) and regional
aeromagnetic anomaly. Cross-products of variables provided better results than
scores for individual variables. The dependent variable used in the multivariate
linear model was logarithmically transformed total amount of copper in one or
more copper deposits per input cell. Neither of the two geophysical input variables
made a significant contribution to the magnitudes of the probabilities that were
being estimated. However, in a separate computer experiment using binary
(presence-absence) input data only, the variable most strongly correlated with
occurrence of large copper deposits was a combination of presence of felsic
volcanics at the surface of bedrock and higher than average Bouguer anomaly.
This result could be interpreted in terms of a mineral deposit model, because nearly
all large copper deposits in Abitibi are of the volcanogenic massive sulphide (VMS)
type and were formed near volcanic centers in association with acidic (felsic)
volcanics, while a relatively high Bouguer anomaly on the Canadian Shield indi-
cates relatively large amounts of mafic volcanic rocks with above average specific
gravity at greater depths.
The probabilities of occurrence estimated for (10 km 10 km) cells in the last

chapter were added for larger (40 km 40 km) unit areas to produce a regionally

based prognostic contour map for expected numbers of known and unknown copper
deposits (Agterberg et al. 1972 ). Later, Assad and Favini ( 1980 ) conducted a
statistical mineral exploration study for the eastern part of the Abitibi region
using localized geophysical (aeromagnetic, gravity and terrain elevation) anoma-
lies. It could be assumed that localized geophysical and geochemical anomalies can
be superimposed on broader regionally based probabilities of mineral occurrence.
This prompted an experiment of using the probabilities previously estimated in the
Abitibi case study as priors in weights-of-evidence (WofE) modeling followed by
later incorporating Favini and Assad's map layers that had been based on local
geophysical anomalies only, in order to obtain updated posterior probabilities of
occurrence (Agterberg 1989b ). A somewhat similar method of updating prognostic