Andreas Ciroth and colleagues have refined the pedigree matrix by deriving uncertainty factors from empirical data. I am excited that the empirical evidence underlines that the pedigree matrix can be a reasonable basis for estimating uncertainties in LCI data.

Current Approach

In ecoinvent, the product life cycle inventory is described by single figures per input or output flow. These numbers contain a level of uncertainty, for instance, because of temporal or spatial approximations. Often the extent of uncertainty cannot be derived directly from the available information. For these cases, a simplified standard procedure was developed to derive uncertainty factors from a qualitative assessment of the data.

The procedure is based on a pedigree matrix composed of five data quality indicators – reliability, completeness, temporal correlation, geographical correlation, and further technological correlation – each with a score of 1 to 5. A score of 1 means that the data is of high quality with regard to that particular indicator (e.g. ‘data from the area under study’ for the indicator geographical correlation); a score of 5 means the data quality for that indicator is low (e.g. ‘non-qualified estimate’ for the indicator reliability).

Each combination of indicator and score gives an uncertainty factor. These uncertainty factors are aggregated into a standard deviation. The aggregation formula, however, is valid for log-normally distributed data only. For the full report on the pedigree matrix approach, see Weidema et al. 2013. The pedigree matrix uncertainty is only one part of the overall uncertainty in the ecoinvent pedigree approach. The additional basic uncertainty, specific to flow and sector, has not been revised by Ciroth and colleagues.

Refinement of the Pedigree Matrix

An important drawback of the current approach is that the pedigree matrix-derived uncertainty factors rely heavily on expert judgment and are not based on a documented empirical foundation. To correct this drawback, Andreas Ciroth and colleagues refined the ecoinvent pedigree approach. Their goal was to support the pedigree matrix uncertainty factors with reasonable, empirical values. Empirical is defined here as ‘derived from experiment and observation rather than theory and expert guesses’. Additionally, they sought to investigate how to apply the pedigree approach to other distributions than the log-normal distribution, but this goal will not be discussed in this article.

To provide empirical values for pedigree matrix uncertainty factors, the authors analyzed studies and datasets in the LCA domain and performed specific measurements on industrial processes. Starting from these data, they performed a stepwise analysis per indicator to provide improved values for the uncertainty factors in the pedigree matrix. Per indicator, the analysis constrained its values by filtering out more and more data sets from the data source. As a result, the data sample became increasingly precise regarding the investigated indicator. Each indicator’s calculated geometrical standard deviation reflects this and is different for each of the filter steps.

Indicator 1: Reliability

To refine the reliability indicator, data from two different databases were analyzed. The E-PRTR database (European Pollutant Release and Transfer Register) provides information concerning the amounts of pollutants released to air, water, and land as well as off-site transfers of waste and of pollutants in wastewater from a list of 91 key pollutants. The GEMIS database (Global Emissions Model for Integrated Systems) contains approximately 10,000 processes in energy and material flows, including transports. The authors detected variation in the contributions from the two databases to the geometric standard deviation of the reliability indicator per flow. To be on the safe side, they recommend using the higher geometric standard deviation contribution value as the uncertainty factor.

Indicator 2: Completeness

For the indicator completeness, data from a yogurt cup study were used. In the study, the cups were weighed and divided into samples. The variance of the samples was calculated and compared to the variance of all samples taken together. As this study is the only data source used to refine the completeness indicator, the authors propose tentative uncertainty factors.

Indicator 3: Temporal Correlation

To determine the temporal correlation indicator, the Tremod database (Transport Emission Model) was studied. This is a database that contains vehicle emissions gathered from 1990 to 2010, used for emission calculations in Germany. Note that the uncertainty factors should be applied for situations where a variation over time can be expected that is not related to technology. If the variation is related to technology, that uncertainty is already described by Indicator 5: Further Technological Correlation.

Indicator 4: Geographical Correlation

To derive new values for the indicator of geographical correlation, the E-PRTR database was used; the results of this analysis are proposed as tentative uncertainty factor values. The E-PRTR database includes emissions in European countries and lists them per year and per industrial facility.

Indicator 5: Further technological correlation

The tentative uncertainty factor values for the indicator of further technological correlation are based on an analysis of the Tremod database. This database contains information on the technology of different means of transport.

Concluding remarks

Ciroth and colleagues derived reasonable uncertainty factors from empirical data for almost all the cases. The work performed here was phase zero of a larger project aiming to refine the pedigree approach in ecoinvent. Updating the pedigree uncertainty factors is a valuable improvement, even though this study did not seek to refine new basic uncertainty values, which are also needed to calculate overall uncertainty. I am convinced that this empirical approach will be of tremendous benefit to ecoinvent users.

Before the new uncertainty factors will be implemented in ecoinvent and SimaPro, Ciroth and colleagues suggest a broader analysis on different sectors. After all, the current study focussed to a large extent on transport databases, and it is important to quantify possible differences between different sectors. Nevertheless, it is exciting that the field of LCA now has empirical evidence for pedigree uncertainty factors. This is an excellent step forward.

Even though the refined pedigree matrix approach is an interesting development, both pedigree matrix uncertainty factors and basic uncertainty factors are, of course, backup options. I suggest that you calculate the uncertainty yourself whenever sufficient data are available.