Identifying best existing practice for characterization modeling in life cycle impact assessment

TLDR: In this article, the authors performed a study for the Joint Research Centre of the European Commission (JRC) to identify the best among existing characterization models and provide recommendations to the LCA practitioner.

Abstract: Life cycle impact assessment (LCIA) is a field of active development. The last decade has seen prolific publication of new impact assessment methods covering many different impact categories and providing characterization factors that often deviate from each other for the same substance and impact. The LCA standard ISO 14044 is rather general and unspecific in its requirements and offers little help to the LCA practitioner who needs to make a choice. With the aim to identify the best among existing characterization models and provide recommendations to the LCA practitioner, a study was performed for the Joint Research Centre of the European Commission (JRC). Existing LCIA methods were collected and their individual characterization models identified at both midpoint and endpoint levels and supplemented with other environmental models of potential use for LCIA. No new developments of characterization models or factors were done in the project. From a total of 156 models, 91 were short listed as possible candidates for a recommendation within their impact category. Criteria were developed for analyzing the models within each impact category. The criteria addressed both scientific qualities and stakeholder acceptance. The criteria were reviewed by external experts and stakeholders and applied in a comprehensive analysis of the short-listed characterization models (the total number of criteria varied between 35 and 50 per impact category). For each impact category, the analysis concluded with identification of the best among the existing characterization models. If the identified model was of sufficient quality, it was recommended by the JRC. Analysis and recommendation process involved hearing of both scientific experts and stakeholders. Recommendations were developed for 14 impact categories at midpoint level, and among these recommendations, three were classified as “satisfactory” while ten were “in need of some improvements” and one was so weak that it has “to be applied with caution.” For some of the impact categories, the classification of the recommended model varied with the type of substance. At endpoint level, recommendations were only found relevant for three impact categories. For the rest, the quality of the existing methods was too weak, and the methods that came out best in the analysis were classified as “interim,” i.e., not recommended by the JRC but suitable to provide an initial basis for further development. The level of characterization modeling at midpoint level has improved considerably over the last decade and now also considers important aspects like geographical differentiation and combination of midpoint and endpoint characterization, although the latter is in clear need for further development. With the realization of the potential importance of geographical differentiation comes the need for characterization models that are able to produce characterization factors that are representative for different continents and still support aggregation of impact scores over the whole life cycle. For the impact categories human toxicity and ecotoxicity, we are now able to recommend a model, but the number of chemical substances in common use is so high that there is a need to address the substance data shortage and calculate characterization factors for many new substances. Another unresolved issue is the need for quantitative information about the uncertainties that accompany the characterization factors. This is still only adequately addressed for one or two impact categories at midpoint, and this should be a focus point in future research. The dynamic character of LCIA research means that what is best practice will change quickly in time. The characterization methods presented in this paper represent what was best practice in 2008–2009.