Data–driven prospectivity modelling of sediment–hosted Zn–Pb mineral systems and their critical raw materials

Demand for critical raw materials is expected to accelerate over the next few decades due to continued population growth and the shifting consumption patterns of the global economy. Sedimentary basins are important sources for critical raw materials and new discoveries of sediment–hosted Mississippi Valley–type (MVT) and/or clastic–dominated (CD) Zn–Pb deposits are likely required to mitigate future supply chain disruptions for Zn, Pb, Ag, Cd, Ga, Ge, Sb, and In. Herein we integrate public geoscience datasets using a discrete global grid to system to model the mineral potential for MVT and CD deposits across Canada, the United States of America, and Australia. Statistical analysis of the model results demonstrates that surface–wave tomography and derivative products from satellite gravity datasets can be used to map the most favourable paleo–tectonic settings of MVT and CD deposits inboard of orogenic belts and at the rifted edges of cratonic lithosphere, respectively. Basin development at pre–existing crustal boundaries was likely important for maintaining the low geothermal–gradients that are favourable for metal transport and generating the crustal fluid pathways that were reactivated during ore–formation, as suggested by the statistical association of both sediment–hosted mineral deposit types with the edges of upward–continued gravity and long–wavelength magnetic anomalies. Multivariate statistical analysis demonstrates that the most prospective combination of these geophysical datasets varies for each geological region and deposit type. We further demonstrate that maximum and minimum geological ages, coupled with Phanerozoic paleogeographic reconstructions, represent mappable proxies for the availability of oxidized, brine–generating regions that are the most likely source of ore–forming fluids (e.g., low– to mid–latitude carbonate platforms and evaporites). Ore deposition was likely controlled by interaction between oxidized, low–temperature brines and sulfidic and/or carbonaceous rocks, which, in some cases, can be mapped at the exposed surface or identified using the available rock descriptions. Baseline weights–of–evidence models are based on regional geophysics and are the least impacted by missing surface information but yield relatively poor results, as demonstrated by the low area–under–the–curve (AUC) for the spatially independent test set on the success–rate plot (AUC = 0.787 for MVT and AUC = 0.870 for CD). Model performance can be improved by: (1) using advanced methods that were trained and validated during a series of semi–automated machine learning competitions; and/or (2) incorporating geological and geophysical datasets that are proxies for each component of the mineral system. The best–performing gradient boosting machine models yield higher AUC for the test set (AUC = 0.983 for MVT and AUC = 0.991 for CD) and reduce the search space by >94%. The model results highlight the potential benefits of mapping sediment–hosted mineral systems at continental scale to improve mineral exploration targeting for critical raw materials.