A Methodology for Similarity Area Searching Using Statistical Distance Measures: Application to Geological Exploration

Abstract

Mineral exploration combined with prospectivity mapping has become the standard process for utilising mineral exploration data. Nowadays, most techniques integrate multiple layers of information and use machine learning for both data-driven and knowledge-driven approaches. This study introduces a novel and generalised methodology for comparing different layers of information by using superpixels instead of pixels to identify similarities. This methodology provides an enhanced statistical representation of regions, facilitating and enabling effective comparisons. Three different statistical distance measures were considered: Kullback–Leibler divergence, Wasserstein distance and total variation distance. We apply the proposed process to data from the Antofagasta region of northern Chile, a well-known area for metallogenic belts, that contain notable copper reserves. Each metric was used and compared, resulting in different similarity maps highlighting interesting mineral exploration areas. The study results lead to the conclusion that the proposed methodology can be applied at different scales and helps in the identification of areas with similar characteristics.