Assessment of chemical emissions in life cycle impact assessment - focus on low substance data availability and

Life cycle assessment (LCA) studies on products or services seem generally to be carried out without a proper inclusion of potential toxic impacts from emissions of chemicals. The first goal of the thesis is to investigate this statement and to clarify whether or not the outcome of an LCA can be significantly dependent on the inclusion of toxicity- or chemical-related impact categories. The two main reasons for poor coverage of potential toxic impacts from chemical emissions in LCA studies are lack of available data on upstream emissions (e.g. emissions during production of raw materials) and lack of substance data on known emissions. To be able to characterize the potential toxic impacts on humans and the environment of chemical emissions, substance data on fate and effect are needed. The second goal of this thesis is to investigate how to deal with low substance data availability on especially effect data within the context of LCA, when the aim is to improve the inclusion of toxicity- or chemical-related impact categories. The first goal regarding the significance of potential toxic impacts in LCA is investigated by carrying out a full LCA case study on printed matter and putting special emphasis on the inclusion of chemical emissions. The second goal regarding low data availability is addressed in two ways. First by introducing selection methods, which are chemical screening methods designed to select the most significant chemical emissions on a low to very low data availability. Secondly by developing a low data demand ecotoxicity effect indicator to be used together with a fate indicator, when estimating the potential impact of chemical emissions. The results of the case study document that for LCAs on printed matter, the inclusion of chemical-related impact categories can be decisive for the outcome, and it shows that chemical-related impact categories are poorly or not at all included in previous studies. The share for the total environmental impact of for example the printing process in the case study is reduced from 41% to 10%, if the chemical-related impact categories are excluded. So, the basis for defining for example ecolabelling criteria (typically based on life cycle thinking) on printed matter is substantially different depending on whether or not the chemical-related impact categories are (properly) included. The investigation on selection methods shows that only three chemical screening methods, associated with a characterisation method, and therefore here defined as selection methods, actually exist to day. Selection method performance criteria are developed including demands on consistency in prioritisation with associated characterisation method, applicability to different chemical groups, high data availability combined with low data demand, data useable in characterisation, user friendliness and transparency. A mainly qualitative evaluation of the existing selection methods against these performance criteria shows that none of these score high on all criteria, and this indicates the need for development of new selection methods. Recommendations on which components to include, which issues to address and general principles for developing selections methods are therefore given. A quantitative evaluation of the consistency in chemical ranking between the existing selection methods (EDIP-selection, Priofactor and CPM-selection), the risk ranking method EURAM, and the characterisation methods EDIP97 and CPM, is performed. The result of this evaluation shows a good correlation between the ranking of all the tested methods, but strongest between the EDIP97 method and its two associated selection methods EDIP-selection (revised version) and Priofactor. A statistical test of correlation in ranking between EDIP97, Priofactor, CPM and EURAM shows significant correlation in all cases. The main reason for this result is that a common perception of what makes a substance ecotoxicologically problematic underlies all four methods. Nevertheless, some outliers as compared to the EDIP97 ranking are identified. These outliers are due to specific characteristics of each of the methods which for certain combinations of substance properties may result in false negatives or false positives as compared to EDIP97. These characteristics include the influence of data availability on the size of assessment factors for conversion of acute effect data to chronic values, and whether or not mode of entry is taken into account in the fate modelling. Further, the reversing of the effect of toxicity on ranking by negative logKow values is observed when logKow is a direct factor in the expression, and there is a significant influence of the way in which the BCF is estimated and included. The second part of the second goal of this thesis, which deals with low availability of substance effect data, is addressed by carrying out an inventory of existing ecotoxicity effect indicator approaches, including a qualitative evaluation based on developed performance criteria. Both impact approaches, and damage approaches, which are all at an early development stage, are included. The evaluation of the existing impact approaches, i.e. the assessment factor-based PNEC approach and the PAF-based approach, shows pros and cons for both. However, taking the comparative nature of LCA and its aim for best estimate into account, and combining this with the possibilities for reducing the data demand of an EC50-based PAF approach, and further including the (at least theoretical) connection to damage approaches, leads to the choice of an effect-based average PAF ecotoxicity effect indicator expressed by 0.5/HC50EC50 for further development. The most reasonable way to estimate the hazardous concentration for 50% of the included species (HC50) based on only three acute laboratory effect data is hereafter investigated by testing and discussing different ways of estimating averages (e.g. median and geometric mean), different data selection strategies and different ways of estimating uncertainty (confidence) limits around the HC50EC50 value. The results of this investigation show that the geometric mean is the most robust estimator for small data sets. Seeking the coverage of many chemicals in LCA and considering the fact that the main part of the useable single species laboratory test data (EC50) is on algae, crustacean and fish, which in practice represent the trophic levels primary producers, primary consumers and secondary consumers, the use of a minimum of three acute EC50 values from each of these three throphic levels is recommended when estimating HC50EC50. Due to the comparative nature of LCA, the possible bias from severe unequal species representation and inclusion of erroneous data, due to bad non-standardised test conditions, should be avoided by only including tests on standard organisms fulfilling certain defined test criteria on durations and endpoints. Further, in order to avoid the effect of possible haphazard or regulatory determined species representations in the data set used, which may be decisive for the weighting of each trophic level in the estimation, the geometric mean based on the average of the averages within each trophic level is chosen for the ecotoxicity effect indicator GM-troph. Hereby, it is consciously chosen to put equal weight on each throphic level. The statistical confidence limits around the GM-troph are in most cases too wide, because the average is based on only three data values, making a statistically significant differentiation between the different toxicants nearly impossible. However, test on fictitious three data value test sets based on combinations of max and min values from a larger ‘mother’ data set indicates that the use of the min and max value among the three data value GM-troph data set (i.e. average within algae, average within crustacean and average within fish) as max-min limits around GM-troph gives a reasonable (and as good as confidence limits) certainty that the ‘true’ GM-troph value (based on the full ‘mother’ data set) lies within the interval. The inclusion of the toxicity-related impact categories in LCA at a similar level as the better established impact categories, like global warming, is far from achieved yet. This thesis point at relatively well functioning selection methods and defines the framework including performance criteria and recommendations on how to improve existing selection methods and how to develop new ones. By introducing the GM-troph this thesis contributes with a robust, low data demanding effect part of the ecotoxicity characterisation factor. The GM-troph has the potential of facilitating a high number of characterisation factors which are robust with relatively low uncertainty if combined with a ‘fate part’ of equal strength. The way for further improvement of the involvement of toxicity-related impact categories in LCA is hereby facilitated.