Anomaly recognition for multi-element geochemical data — A background characterization approach

Techniques for recognizing populations and defining anomalous samples in geochemical surveys are of two types: objective and subjective. Subjective techniques involve defining samples as anomalous if they have concentrations greater than a ‘certain percentile’ or greater than the mean plus ‘some multiple’ of the standard deviation. In both cases, the number of anomalous samples defined is dependent upon a subjective ‘rule’. Objective methods for defining geochemical populations include probability plots and the gap statistic. In both cases a model for the distribution of the data must first be assumed, and then the actual data distribution defines what samples may be anomalous. The Background Characterization Approach to anomaly recognition is an objective method designed for cases where significant overlap of multiple populations exists, or where no anomalous population can be recognized (unimodal distributions). It involves the determination of a background regression model for a pathfinder element using only background samples, and then applying this model to all possibly anomalous samples. Samples with element concentrations predicted successfully by the background model are classified as background, while those ‘reacting’ differently to the function are deemed to be from another population and therefore anomalous. A multi-element stream sediment exploration survey for stratabound Cu-Ag deposits in the Belt Basin of Montana is used to demonstrate the technique when the data are unimodally distributed.