Regional-scale nickel sulfide prospectivity mapping of the Yilgarn Craton, Western Australia A. Porwal 1, 2 , I. González-Álvarez 1 , V. Markwitz 1 , T.C. McCuaig 1 , A. Mamuse 1,2 1 Centre for Exploration Targeting, The University of Western Australia, Crawley, WA 6009, Australia 2 Western Australian School of Mines, Curtin University of Technology, Bentley, WA 6102, Australia Abstract: Bayesian weight-of-evidence and logistic regression models are implemented in a GIS environment for regional-scale prospectivity mapping of nickel sulfide deposits in the Yilgarn Craton, Western Australia. The input variables for the models consisted of GIS layers that were used as proxies for the mappable exploration criteria of nickel sulfide deposits in Yilgarn. About 70% of the 169 known deposits of the Craton were used to train the models; the remaining 30% were considered “undiscovered” and used to validate the models. The output continuous scale prospectivity maps were reclassified into binary prospectivity maps based on the threshold values extracted from area versus prospectivity curves. The weights-of-evidence and logistic regression models, respectively, classify 81% and 86% deposits in high prospectivity zones that occupy about 8% of the total area of the Craton. The superior performance of the logistic regression model is attributed to its capability to accommodate conditional dependence of the input predictor maps. Keywords: Yilgarn Craton, GIS-based prospectivity mapping, Weights-of-evidence, Logistic regression, Magmatic nickel sulfide deposits. 1 Introduction Australia is the largest global producer of nickel after Russia and Canada, contributing about 12.8% to the global production of the metal (Hoatson et al, 2006). The entire production of Australian nickel comes from nickel-sulfide (about 82%) and laterite (about 18%) deposits of Western Australia (Abeysinghe and Flint, 2007). The Yilgarn Craton of Western Australia hosts several major nickel-sulfide deposits, including the world-class deposits at Mt Keith, Perseverance, Kambalda, Yakabindie, and Honeymoon Well (Hoatson et al., 2006 and references therein). All major deposits of the Yilgarn Craton are associated with komatiite sequences in Archaean greenstone belts, which have significant potential for the discovery of new deposits in Archaean granite–greenstone terranes, especially in the areas where mafic-ultramafic rocks are poorly exposed (Hoatson et al., 2006). 2 Spatial-mathematical-model-based prospectivity mapping GIS-based mineral prospectivity mapping of an area involves demarcation of potentially mineralized zones based on geologic features that exhibit significant spatial associations with the targeted mineral deposits. These features, which are termed deposit recognition (or exploration) criteria, are spatial features indicative of various genetic earth processes that acted conjunctively to form the deposits in the area. Exploration criteria are sometimes directly observable; more often, their presence is inferred from their responses in various exploration datasets that are appropriately processed in order to enhance and extract predictor-map layers. In model-based mineral prospectivity mapping, these predictor-map layers are used as proxies for exploration criteria. A spatial mathematical model for mineral prospectivity mapping can be defined as a simplified mathematical representation of the spatial association between exploration criteria and the targeted mineral deposits. Several models have been documented in literature, for example, Bayesian probabilistic, logistic regression, fuzzy, neural network, hybrid neuro-fuzzy models (see Porwal, 2006, and references therein). Because of its intuitive approach and ease of implementation, the Bayesian weights-of-evidence is probably the most widely used model in mineral prospectivity mapping. However, this model is based on a (log)-linear implementation of Bayes’ equation under the assumption of conditional independence of the input data, which may not hold in many real-world geological settings. Alternate approaches such as logistic regression, neural networks and Bayesian classifiers can accommodate conditional dependence of the input data, and hence yield unbiased output (Singer and Kouda, 1999; Brown et al., 2000; Porwal et al., 2006). In this study, we applied a weights-of-evidence (WofE) and a logistic regression (LR) model for nickel- sulfide prospectivity mapping of the Yilgarn Craton. 3 Exploration criteria for nickel sulphide deposits in Yilgarn In general, nickel-sulfide deposits are associated with mafic/ultramafic rocks (Naldrett, 1997). These deposits are formed by the saturation of nickel-rich, mantle- derived, mafic/ultramafic magmas with respect to sulfides, which results in segregation of immiscible nickel-sulfide liquid. The sulfide saturation is considered to be brought about by interactions with crustal rocks (Arndt et al., 2005), although alternate models have been proposed by various workers (e.g., Fiorentini et al., 2008). According to Hoatson et al. (2006), most fertile komatiite sequences in Yilgarn temporally cluster at ~2700 or ~1900 Ma. These ages are interpreted as periods of juvenile crustal growth and production of large volumes of primitive komatiitic flows, which are the main sources of nickel in the Craton. Trans- lithospheric faults, which have close spatial and genetic association with nickel-sulfide deposits, are potential pathways that allow the intruding magma to reach higher levels of the crust, thus promoting crustal contamination,