UC Santa Barbara

Life cycle assessment (LCA) aims to support corporate and public policy decisions by quantifying the environmental performance of a product. Understanding uncertainties in LCA results is therefore important for making informed decisions. Monte Carlo simulation (MCS), which uses random samples of the parameters from pre-determined probability distribution, has been widely utilized to characterize uncertainties in LCA. However, as the size of an LCA database grows, running a full MCS is becoming increasingly challenging. Furthermore, uncertainty literature in LCA has focused on life cycle inventory (LCI), while the uncertainties from the remaining steps—including characterization, normalization, and weighting—have not been addressed, despite their perceived relevance in overall uncertainty characterization in LCA.

The objectives of my dissertation are: (1) to develop a new method to improve the computational efficiency of large-scale MCS in LCA, (2) to empirically test the reproducibility of comparative decisions obtained using the method, and (3) to develop and test an analytical method to decompose the overall uncertainty in LCA into its constituents. The new method for uncertainty characterization in LCA involves pre-calculating and storing the distribution profiles of the most widely used LCA database, ecoinvent. Using parallel computing, I have generated the distribution functions for 22 million life cycle inventory (LCI) items of the database. I then tested 20,000 randomly selected comparative LCI cases, and showed that pre-calculated uncertainty values can be used as a proxy for understanding the uncertainty and variability in a comparative LCA study without compromising the ability to reproduce the comparative results.

Another key barrier to conducting uncertainty analysis in LCA occurs in life cycle impact assessment (LCIA), an important step of LCA calculation flowered LCI phase, because characterization models for LCIA do not typically provide uncertainty information for the input parameters, and lack detailed information about the relationships between those inputs. A Pedigree matrix for characterization factor in LCIA was developed to fill in the gap in the uncertainty characterization in LCA. Expert opinions of the use of Pedigree method in estimating uncertainty in LCIA and the Pedigree scores for both LCI and LCIA were collected through an online survey.

Finally, I demonstrated a new method to decompose the overall uncertainties of an LCA result over the contributing factors including those from LCI, characterization, normalization, and weighting, which are the steps involved in LCA calculation. To do so, I adopted the logarithmic mean Divisia index (LMDI) decomposition method into MCS parse out the overall uncertainty into its constituents.

I believe that my research helps improve the efficiency and analytical power of uncertainty analysis in LCA. The findings can be applied to other problems outside of LCA that utilize MCS.