Author(s): Ana Catarina Rocha; Carla Palma; Ricardo J. N. Bettencourt Da Silva
Linked Author(s):
Keywords: No Keywords
Abstract: Oil spills in marine environments are common and represent a significant environmental concern. Chemical analyses of spill samples (Sp) and suspected sources (SS), collected in the spill and potential sources, are used in court to assign legal liability. These analyses rely on characterizing the chemical composition of oil samples (oil fingerprints), using a set of ratios between abundances of oil-discriminating compounds, known as diagnostic ratios (DR), and comparing the fingerprints using statistical approaches. The Nordtest guideline (Faksness et al., 2002) and the standard EN 15522-2 (European Committee for Standardization, 2023) are two analytical methodologies that have been employed for the comparison of DR observed in two samples using distinct DR comparison conditions and approaches, i. e., t-Student test (S-t) and a single criterion (SC) applied to the comparison of all DR of the fingerprint, respectively. These approaches assume assumptions or approximations, such as the normality of the DR probability distributions, that can lead to a higher probability of inaccurate findings. However, it is known that the ratio of independent or correlated random variables, such as oil component abundances, can deviate significantly from normality (Bettencourt da Silva, 2016). Therefore, it is necessary to develop new methods for interpreting results that are based on suitable statistical approaches and assumptions in order to guarantee the quality of the identifications. Moreover, the weight of the chemical analysis presented as evidence (E), often expressed as a likelihood ratio (LR), must be demonstrated in court. The LR is a ratio of conditional probabilities that assesses the evidence's contribution to the prosecution and defense hypotheses, namely, that the perpetrator of the oil pollution is the defendant and the perpetrator of the oil pollution is someone other than the defendant, and its determination is challenging and requires databases or statistical models (Thompson et al., 2012). This work introduces a novel approach to accurately model DR probability distributions, using simulation by Monte Carlo Method (MCM), providing statistically sound criteria for DR comparison. The developed set of computational tools compare samples' fingerprint, and estimate the probabilities of true (TD) and false (FD) declaration of oil fingerprints equivalence used to determine LR. This allowed compare the performance and the quality of the identifications obtained by the new approach with the S-t and SC approaches.
Year: 2024