Pattern Recognition Methods in Environmental Radioactivity Studies

Abstract

Pattern recognition methods provide powerful tools for the analysis and interpretation of large environmental data sets generated within environmental monitoring programmes. Most of these data sets consist of trace elements and/or trace organic pollutants patterns. Only a few studies have been done on employing the pattern recognition methods to describe the correlation between locations and radioactive contamination. This multivariate approach has been used predominantly for the identification of radioactive isotopes, quantitative gamma-ray spectrometry analysis and for optimization of gamma-ray spectrometric measurements. Spatial data analysis based on radioactive contamination of diverse regions is complex for many reasons. These include the uncertainty of the source term, high spatial and temporal variability of pollution patterns, spatial and temporal non-stationary and the multivariate nature of the phenomenon with linearly and non-linearly correlated variables. Our studies show that the geographic origin can be recognized with minimum effort if the relevant constituents are analyzed and the results are included in data analysis algorithms. Five common pattern-recognition techniques, artificial neural networks (ANN), principal component analysis (PCA), linear discriminant analysis (LDA), k-nearest neighbors (kNN) and soft independent modeling of class analogy (SIMCA) were employed to classify soil and bioindicator samples (mosses and lichens) according to their geographical origin, based on their content of radionuclides from different sources (members of the natural uranium and thorium decay chains, cesium isotopes originating from the Chernobyl power plant accident and cosmogenic beryllium), determined by gamma-ray spectrometry. The ability of the ANN to extract hidden features from the input signals was found to be particularly useful for this kind of monitoring, when the data sets with complex correlation structures had to be analyzed, or when data sets contained series of many highly inter correlated variables.