Interpretation of geochemical anomalies and domains using Gaussian mixture models

The element's concentrations in stream sediments are frequently the product of complex geological processes and a variety of lithological sources. This variety of processes dilutions or concentrates the trace elements in the sediments concerning the original concentration presented in the source rock. When analyzing the geochemical data of stream sediments, it is common to find skewed distributions. This skewness can be attributed to multiplicative effects involved in dilution processes that result in lognormal distributions. It also can imply a mixture of populations that results from the variety of geological processes and/or the diversity of lithological sources. In this study, a method is proposed to classify geochemical samples of sediments into data sets or geochemical domains that follow normal, lognormal or log-ratio normal distributions. This methodology can be based on the classification of samples according to geological features or use Gaussian mixture models to determine the geochemical domains. To mixture modeling, CLR transformation and the expectation-maximization (EM) algorithm were used. The results obtained include mixture models of 2-3 populations associated with main geochemical processes and/or dominant lithological sources. The proposed methodology defines threshold values for each domain and allows to set a more appropriate background. An analysis carried out with data from northern Mexico identified anomalies that would not be defined without the separation of previous domains.