Using multivariate compositional data analysis (CoDA) and clustering to establish geochemical backgrounds in stream sediments of an onshore oil deposits area. The Agri River basin (Italy) case study

Establishing the natural background levels of chemical elements is very often extremely complicated. This is even more true especially for the more anthropized areas, where the concentration of an element in the environmental matrices is conditioned not only by the natural (geogenic) context but also by anthropic activities. In this study, the natural background levels (NBLs) of several chemical elements in the stream sediments of the High Agri River basin have been calculated using a combination of multivariate statistical analysis (MSA) and machine learning techniques in the frame of compositional data analysis (CoDA). Specifically, data clustering and robust principal component analysis (rPCA) have allowed us to recognize and isolate three different data populations, belonging to three different geological domains. A first population of data is clearly related to calcareous lithologies, while the second one is associated to siliciclastic lithologies. A third one is of mixed origin, partly linked to the volcanoclastic nature of the sediments and due to the presence of co-precipitation phenomena. The three data populations have been later separated into three different databases and, whereas needed, the outliers have been eliminated. Based on the obtained results, the NBLs have been calculated using the US-EPA's (United States Environmental Protection Agency) ProUCL software. The results derived from this new approach have been later compared with those obtained using the spectral analysis (S-A) method to evaluate its advantages and versatility. This approach allowed calculating more reliable natural background values.