Mineral Prospectivity Mapping Using a Combination of Cell-Based Association and Gradient Boosting Methods: Investigating Gold Occurrences in French Brittany
Résumé
Mineral prospectivity mainly focuses on measuring the statistical link between a mineral occurrence data set (points) and a geological map (polygons) to point out favorable zones in terms of mining potential. However, usual favorability methods may not yield satisfactory results due to input data quality (e.g., clustered points, mixed and scarce data, approximate location) or some assumptions that are considered unreasonable (e.g., map areas relevance, conditional independence). To address this issue, the Cell-Based Association (CBA) method has been developed by Tourlière et al. (2015). This method consists in replacing polygons of geological units with a square cell grid, each cell recording the presence (1) or absence (0) of lithological units from the geological map. The synthetic map of lithological associations can then be coupled to the mineral occurrence data set to generate a favorability map. In this study, we combine the CBA method with a supervised machine learning technique based on an ensemble of decision trees (gradient boosting) to investigate Promine gold occurrences in French Brittany. We compare the performance of this approach with a baseline model based on the classical Weight of Evidence method. We show that the use of such machine learning approach 1) significantly improves the predictivity of the model and 2) eases the process of adding valuable information (e.g., faults, streams, etc.) that increases further its performance.