A FUNDAMENTAL APPROACH TO THRESHOLD ESTIMATION IN EXPLORATION GEOCHEMISTRY - PROBABILITY PLOTS REVISITED

Several threshold estimation (and thus anomaly recognition) procedures are in use in exploration geochemistry. Experimental methods rely on absolute values in graphs or tables and are highly subjective in being dependent of the variable experience of explorationists. Model-based subjective techniques of threshold determination, including the mean plus two standard deviations, are arbitrary and inefficient; thus, they are not suitable despite the widespread use they have found in the past. Model-based objective methods include the gap statistic and the probability graph approaches, the latter finding much greater acceptance with the greatly increased ease of application using a recently available microcomputer software package that can treat many variables easily and rapidly. Many critical decisions in exploration geochemistry require a comprehensive interpretation of available data, including clear insight into the recognition of anomalous and background samples. Several decisions cannot be made in a rigorous and confident manner unless based on a fundamental approach to threshold selection. Examples include: (1) element zoning in geochemistry; (2) absolute estimation of geochemical contrast; (3) the recognition of the isotropic or anisotropic nature of anomalies; and (4) an estimation of areal extent of anomalies of various elements. Methods of threshold estimation which incorporate the philosophy that anomalous and background data are each characterized by their own probability density functions will be most successful in deriving a fundamental approach to threshold estimation. In the simplest general case, there are two overlapping populations, the overlapping character leading naturally to the extension of the single threshold concept of the definition of two thresholds that delimit the range of overlap. Such a concept, easily conceived and applied to individual variables, can be extended to the n-dimensional case. Univariate approaches will continue to dominate practical applications in the foreseeable future except in special circumstances.