Multi-process and multi-scale spatial predictive analysis of an orogenic Archean gold system, Rio das Velhas Greenstone Belt, Brazil

摘要

There has always been a need for new methodologies and research to improve the decision-making process at the early stages of mineral exploration. This article presents a novel approach to integrating geodata in support of a mineral systems-based spatial analysis of orogenic gold deposits in the Rio das Velhas Greenstone Belt (RVGB), Quadrilkero Ferrffero Province, Brazil. The gold mineralization in the RVGB is spatially associated with thrust faults and shear zones and mainly hosted by iron-rich rocks such as mafic-ultramafic sequences and banded iron formations. To best represent the targeting elements of this mineral system spatially, a knowledge-based fuzzy logic method was employed to map the expressions of the gold depositional processes at the province (1:500,000), district (1:100,000) and camp (1:50,000) scales. At each scale, multivariate statistical techniques served to enhance multiple geological, geophysical, and geochemical datasets and extract from these data spatial proxies of the gold depositional processes. The results of this multi-scale predictive analysis were as follows: The first, province-scale model (M1) identified the entire gold prospective tract and the areas within it that may be of greatest relevance to future exploration. The second, district-scale model (M2) identified the different gold camps within the prospective tract and mapped the areas of gold favorability in a more detailed manner. The third, camp-scale model (M3) identified areas that, based on the current knowledge and distribution of high resolution geodata, are the most favorable whilst also being small enough as to permit target testing using conventional mineral exploration tools such as geophysics, geochemistry and/or drilling. The results obtained from our predictive models were validated by comparing them against the known gold occurrences using ROC (receiver operating characteristics) curves and AUC (area under the curve) graphs. According to these validations, model Ml scored an accuracy of 93.38%, whereas models M2 and M3 scored accuracies of 88.31% and 93.38%, respectively. A key observation made in the course of this study is that the gold prospective area as predicted by models M1, M2 and M3 varies according to the scale of the analysis. A novel factor in our approach is that we aimed assess the targeting criteria and spatial datasets that underpin them according to their spatial resolution and presented the results in form of integrated maps. In addition, the tools developed in this study have the capacity to reduce the cost of direct detection technologies regarding the transition from broad regional to camp scale at the early stages of mineral exploration, where the most initial decisions in search and area reduction are critical.