Modelling target maps of future gold occurrences with combination of categorical and continuous conditionally-dependent supporting patterns

This contribution discusses procedures to obtain target maps of future events such as the likelihood of discovering new gold occurrences knowing location and spatial context of a set of genetically-related gold occurrences. A target map is generated iterating a modelling process using a subset of the known occurrences and cross-validating its prediction pattern with the distribution of the remaining subset. The target map is obtained integrating the prediction patterns from all iterations. Strategies are proposed for 37 gold occurrences in the Red Lake study area, northern Ontario, Canada: sequential elimination of 1 out of 37, and random selection of 24. In either case 37 prediction patterns are generated whose associated statistics from cross-validation provides a target pattern and associated uncertainty of class membership. Two prediction models used are: one based on a fuzzy set function and one on a logistic discriminant function. Target and uncertainty maps are based on spatial relationships between the distribution of 37 gold occurrences and that of categorical and continuous-value geophysical, geological and geochemical maps, many of which are conditionally dependent. Using rank-based statistics only marginal effects of conditional dependence appear to affect the results of the modelling.