Handling Uncertainty in Systems Analysis

It has been suggested that not all the methods of statistical analyses of uncertainty can be applied to LCA, primarily because the underlying LCA data are not based on random samples, that is, we are not strictly dealing with random variables that follow some known frequency distribution. However, one can apply subjective probability distributions to quantify the uncertainties involved. The established procedure to achieve this consists of the following five steps: selection of essential parameters, determination of probability distributions, Monte Carlo simulation, significance analysis, and interpretation of the results.^[1]'⁶¹

Recent approaches have tried to integrate consideration of both technical and valuation uncertainties during decision making on the basis of the results provided by LCA. These are to be used in conjunction with established decision support tools based on multiple criteria decision analysis since it has been demonstrated that the structure of LCA has parallels with a decision analysis approach to decision making.¹⁷¹ Key elements of this approach include “distinguish ability analysis” to determine whether the uncertainty in the performance information is likely to make it impossible to distinguish between the activities under consideration, and the use of a multivariate statistical analysis approach, such as principal component analysis. The latter enables rapid analysis of large numbers of parallel sets of results, thereby allowing for the identification of choices (options) that lead to similar and/or opposite evaluation of activities.

Application of Subjective Probability Distributions Approach

This method of uncertainty analysis in LCA has inherent attributes since they are an alternative to simple point estimates, which allow for the use of a range within which we expect the true value to lie rather than using a single number to estimate the results of some real-world quantity (e.g., tons of sheet of steel required to make 1 ton of an output product). In this manner, by developing subjective probability distribution, the methods of uncertainty analysis can be used in LCA to account for the uncertainty in the results. This further enables accounting for the reliability of the decisions reached on the basis of the LCA outcomes, and if the reliability of the conclusions is not sufficient for our decision-making needs, then uncertainty analysis helps identify which data uncertainties are most significant, that is, influential in the process chain. Furthermore, applying the inverse method, this step can help the decision maker determine the levels of reduction in data uncertainty required to reach a specific level of confidence in the results. For example, to attain “90% confidence” in the results, one would need to assure that using the subjective probability distribution approach if the LCA analysis of a process is repeated several times, each time using new and equally probable randomly selected values for the uncertain input quantities, the conclusions would be correct 90% of the time.

Uncertainty Analysis versus Data Quality Characterization

Owing to a limited number of sources providing information while generating the LCI data, quite often the uncertainty related to the amount of a specific input or output cannot be derived from the available information. For such circumstances Frischknecht et al.^|8] have developed a simplified standard procedure to quantify the uncertainty. This simplified approach includes a qualitative assessment of data quality indicators on the basis of a pedigree matrix from published literature. Basic uncertainty factors are used for the kind of input and output considered. For example, it is assumed that C0₂ emissions generally show a much lower uncertainty as compared with CO emissions since the former is usually calculated from fuel input, whereas the latter depends on numerous parameters such as boiler characteristics, engine maintenance, and load factors. Table 1 provides a list of proposed uncertainty factors in a pedigree matrix, which have been based on expert judgments.

TABLE 1 Examples of Basic Uncertainty Factors (Dimensionless) Applied for Technosphere Inputs and Outputs and for Elementary Flows

Source: Data from Frischknecht et al.^|s|.

c, combustion emissions; p, process emissions; a, agricultural emissions.

• [1] The LCA is very dependent on data of good quality. In most cases, LCA practitioners rely on genericdatabases provided by different sources. The commonly available databases to date are the EuropeanUnion’s (EU) European Reference Life Cycle Data System (ELCD), Swiss National LCI Database (ecoin-vent), the U.S. LCI Database created by the National Renewable Energy Laboratory and its partners, theCanadian Raw Materials Database (CRMD), the Swedish National LCA Database (SPINE@ CPM), theDanish Food Database, and the Korean National LCI Database (KNCPC). Apart from these, currentlyAustralia and Japan also have ongoing initiatives to generate national LCA databases.