Translating a mineral systems model into continuous and data-driven targeting models: An example from the Dolatabad chromite district, southeastern Iran

Abstract

The Dolatabad chromite district, a well-endowed mineralized tract located in SE Iran, hosts numerous podiform chromite deposits. Its favorable geological setting and known chromite endowment suggest that the district has good potential for the discovery of additional chromite mineralization. However, the processes that controlled the development and preservation of the chromite ores are poorly known. As such, a better understanding of the ore-forming processes will be critical for the success of future exploration activities in the Dolatabad chromite district. Recognition of these critical genetic processes and the mapping of their spatial expressions are the precursors to mineral prospectivity mapping (MPM) as well as any mineral exploration activities on the ground. This study adopted a mineral systems approach to MPM, which entailed the formulation of a probabilistic framework to the recognition of the critical genetic processes and translation of these processes to mappable evidence maps. Here, we developed three proxies, representing trap, deposition, and preservation processes, that were subjected to a range of weighting procedures, including continuous (i.e., fuzzy gamma, geometric average, and data-driven index overlay) and data-driven machine-learning (i.e., multilayer perceptron neural network and random forests) methods. The machine-learning procedures outperformed the continuous procedures, suggesting that the former approaches are more reliable in targeting mineralized zones. The targets delineated by random forest algorithm, the superior model generated in this study, predicted c. 71% of the known chromite mineralization in c. 6% of the study area.