Spatio-Geologically Informed Fuzzy Classification: An Innovative Method for Recognition of Mineralization-Related Patterns by Integration of Elemental, 3D Spatial, and Geological Information

Abstract

Recognition and mapping of mineralization-related patterns in geochemical data is a key computational analysis to achieve a predictive model of prospectivity for mineral deposit occurrence. This contribution describes a spatio-geologically informed fuzzy classification (SGIFC) for portraying the spatial-frequency distribution of mineralization through integration of geochemical concentrations, 3D spatial properties, and geological knowledge contained in surface samples. A spatio-geological interaction model (SGIM) was defined to constitute an SGIFC by modulating fuzzy memberships in each iteration of standard fuzzy c-means (FCM) according to the incorporation of both spatial and geological inter-sample similarities. This strategy was adapted to the compositional nature of multi-elemental data and implemented by programming over packages supplied within the R language environment. A comparative experiment on soil samples collected in a porphyry Cu–Au system was subjected to understand how the SGIM affects standard memberships spatially, statistically, and geostatistically. Spatial autocorrelation analysis of fuzzy memberships through Moran’s I calculation and Monte Carlo simulation indicated that SGIFC leads to more robust spatially connected patterns compared to FCM models. Moreover, variogram analysis illustrated that SGIM orients the spatial continuity of mineralization-related memberships along the diffusion bearing of hydrothermal fluids across geological structures of the study area. Rock samples collected along trenches, as well as Cu volumetric productivities, were used as benchmarks to evaluate the integrity of the derived predictive models. The results reveal that the distribution pattern of mineralization-related memberships for SGIFC, compared to that of FCM, is more consistent with sub-outcropping and deep metallogenic realities and describes a more significant quantitative association with subsurface mineral content within the study region. The incorporation of SGIM into fuzzy classification has demonstrated the sufficiency to synthesize the comprehensive information of samples to predict effectively potential targets for follow-up exploration.