Mineral Occurrence Target Mapping: A General Iterative Strategy in Prediction Modeling for Mineral Exploration

Abstract

Procedures are discussed to construct target maps ranking the likelihood of future discoveries: for instance, of gold occurrences, knowing location and spatial context, of a set of genetically related gold vein deposits. A favorability modeling process is iterated with a subset of the known occurrences, i.e., the locations of the deposits. The resulting prediction patterns are cross-validated with the distribution of the left-out occurrences, considered as representing the future discoveries. The target map originates from integration of all prediction patterns from the iterations. Rank-based statistics related to the target maps provides measures of quality, robustness and uncertainty of the classification of a study area into likelihood of discovery. Much of this is a relatively new area of research, so that to interpret such uncertainty is still a challenge. Four critical questions are formulated that identify areas in need of extensive research for any modeling procedure. They relate to the quality of prediction patterns, their associated uncertainty of class membership, their sensitivity to redundancy and to congruity within the database. A spatial database developed for advanced training is used to generate target maps. It comes from a study in the Red Lake area in northern Ontario, Canada. It contains information on 37 gold vein deposits. Their neighborhood distribution is instrumental to establishing spatial relationships with the units of categorical thematic maps and continuous field maps. The experimental results point at the extraction of significant properties of the spatial data that cannot be ignored but that we have yet to master to substantiate the reliability of prediction patterns.