Extending life cycle assessment normalization factors and use of machine learning - A Slovenian case study

•Regional normalization factors (NF) can greatly deviate from European NFs.•Machine learning was used to model missing toxic chemicals characterization factors.•Toxic chemicals properties were correlated with characterization factors.•Three additional impact categories were added and provided with results.

Within a life cycle assessment (LCA), normalization is an essential part for interpretation. In Europe, only the European Union normalization factors (EU NFs), with 2000 as the reference year, are available for LCA practitioners, although they work on a regional level. The hypothesis of this research was based on the assumption that some regional NFs deviate from the EU NFs due to unique regional profiles, or because of previously omitted or generalized human impact. In this particular case study set in Slovenia has been tested. By working on the ReCiPe 1.08 life cycle impact assessment methodology, 18 impact categories were investigated, and an additional 3 were added: electromagnetic radiation, light pollution and electric use. To meet practitioners' needs, the most up-to-date inventory data was used with the reference years of 2007-2012. Out of 440 environmental interventions that were investigated, 139 had no characterization factors (CFs), 97 were estimated using machine learning and 42 had to be omitted. The final result confirmed our hypothesis. Twenty NFs were compared with the EU NFs, and the results have shown that on average, the NFs differ by a factor of 9.76 (median = 1.65). The reasons for the high deviation are due to natural land transformation, and ionising radiation, toxicological and ecotoxicological impact categories; where there are major data gaps in the CFs. The primary concern of the research was data availability for toxicological and ecotoxicological parameters for toxicity-related emissions, and the fact that original CFs covered only 50.25% of plant protection products used in Slovenia. Toxicological and ecotoxicological uncertainties were illustrated by comparing four different results. Future studies should be focused on the use of machine learning to provide the next generation of CFs and to go beyond the CFs' damage-oriented assessment. Remediation should be the new endpoint category and its units should be Joules.