Translation of mineral system components into time step-based ore-forming events and evidence maps for mineral exploration: Intelligent mineral prospectivity mapping through adaptation of recurrent neural networks and random forest algorithm
Soran Qaderi, Abbas Maghsoudi, Mahyar Yousefi, Amin Beiranvand Pour
Abstract
• A time-step-based mineral system components is introduced. • Temporal-based mineral prospectivity mapping (TMPM) is proposed. • TMPM considers the temporal dependencies of evidence layers. • Recurrent neural networks (RNNs) are adapted for MPM. • TMPM using the adapted RNNs recognizes reliable exploration targets. In the integration step of conventional mineral prospectivity analysis approaches chronology of ore-forming subsystems is ignored leading to less reliable predictions. In this paper, we design and adapt recurrent neural network architectures, which have the ability of modelling sequence-related natural events, and a random forest algorithm to bring the temporal nature of ore-forming subsystems into prospectivity analysis procedure and to mitigate the aforementioned issue. A dataset of Pb-Zn mineralization in Semnan Province, Iran, is used to illustrate the procedure. The exploration targets in the prospectivity maps show excellent agreement with the deposit locations, demonstrating the importance of incorporating the chronology of ore-forming geological processes in targeting mineral deposits. This study links our understanding of the chronology of mineral system parameters to predictive modeling to support decision-making in mineral exploration targeting.