Recognizing geochemical anomalies via stochastic simulation-based local singularity analysis

Abstract

The complexity of geochemical variation patterns and the limited number of samples that can be collected jointly make uncertainty an element that could not be ignored in modeling the geochemical distributions and further identifying geochemical anomalies. Local singularity analysis (LSA) of stochastically simulated geochemical distributions is carried out in this paper to identify geochemical anomalies in southwestern Fujian province of China, with the uncertainty at unsampled locations and its propagation being considered. A supervised way by analyzing the sensitivity of area under curve (AUC) to threshold values to determine an appropriate threshold is also presented. The case study further illustrated and validated the procedure, and several insights can be obtained thereby: (1) the uncertainty of values at unsampled locations could significantly affect the grid-based LSA result, especially for those locations with weak singularities. An explanation from the LSA algorithm is that the estimation of singularity exponent involves multiple local neighborhoods of the location to study, uncertain values at any location within the neighborhood would affect the result. Locations with weak singularities are quite sensitive to such uncertainties in that enrichment or depletion patterns identified are likely to change from one to the other. (2) Sensitivity analysis of AUC to threshold values provides a feasible way to determine an appropriate threshold to derive the anomaly probability map, which can enhance the geochemical signatures with respect to the interpolation-based singularity map. The anomaly probability accounts for the local distribution structure of singularity exponents, hence providing a more discriminative measure to enhance geochemical signatures with respect to the single singularity exponent from interpolation-based LSA, which mainly reveals the trend of local singularities. (3) Several targets have been preliminarily delineated based on the anomaly probabilities and favorable geological conditions in the study area, which should be validated by other types of information available in near future.