The Impact of Compositional Data in Environmental Risk Assessment through Information Theory

Abstract

A water system impacted by mining activities was assessed to determine the extent of contamination, in the Trimpancho River mining system, in Spain. This system is in the Iberian Pyritic Belt, a metallogenic province in the southwest region of the Iberian Peninsula. Related pollution has been stud-ied by multiple authors in recent decades. However, a pollution geochemical signature is not yet defined, even if, a few elements such as Cd, Cr, Cu, Fe, Hg, Mn, Pb, and Zn reach critical values, much above legislation for surface waters. Mercury is responsible for the highest level of hazard and therefore is central to defining water pollution signatures associated with acid drainage. Water samples were collected at the surface level of the streams, acidified with nitric solution, and stored in dark glass (only for Hg) and polyethylene containers at 4°C. Samples were digested with nitric and hydrochloric solu-tions in a high-pressure microwave unit and analyzed in ICP-OES for the majority of metals. Hg was directly analyzed in a mercury analyzer (NIC MA-3000). Since the chemical element concentration is compositional, an analysis was conducted to quantify how the uncertainty of the states of a to-be-predicted variable (mercury) is influenced by using both raw and centered log-ratio transformation (CLR) data. For that purpose, a methodology based on information theory (IT) and implemented through a Bayesian approach was used to about the obtained results, the normalized entropy decreased from 43% (raw data) to 33% (compositional data), and a Contingency Table Fit of 21% (raw data) was obtained compared to 71% (compositional data).