Uncertainty and Variability in Product Carbon Footprinting

Abstract

Summary Recent years have seen increasing interest in life cycle greenhouse gas emissions accounting, also known as carbon footprinting, due to drivers such as transportation fuels policy and climate‐related eco‐labels, sometimes called carbon labels. However, it remains unclear whether applications of greenhouse gas accounting, such as carbon labels, are supportable given the level of precision that is possible with current methodology and data. The goal of this work is to further the understanding of quantitative uncertainty assessment in carbon footprinting through a case study of a rackmount electronic server. Production phase uncertainty was found to be moderate (±15%), though with a high likelihood of being significantly underestimated given the limitations in available data for assessing uncertainty associated with temporal variability and technological specificity. Individual components or subassemblies showed varying levels of uncertainty due to differences in parameter uncertainty (i.e., agreement between data sets) and variability between production or use regions. The use phase displayed a considerably higher uncertainty (±50%) than production due to uncertainty in the useful lifetime of the server, variability in electricity mixes in different market regions, and use profile uncertainty. Overall model uncertainty was found to be ±35% for the whole life cycle, a substantial amount given that the method is already being used to set policy and make comparative environmental product declarations. Future work should continue to combine the increasing volume of available data to ensure consistency and maximize the credibility of the methods of life cycle assessment (LCA) and carbon footprinting. However, for some energy‐using products it may make more sense to increase focus on energy efficiency and use phase emissions reductions rather than attempting to quantify and reduce the uncertainty of the relatively small production phase.

Redefining scope: the true environmental impact of smartphones?

“…The main environmental impact of a smartphone in terms of GHG emissions is in the manufacture phase, whereas the server creates greatest environmental impact during the use phase. Both Stutz et al (2012) and Weber (2012) reports 94 % of all GHG emissions attributed to the use phase. By comparison, the use phase of a smartphone accounts for less than 18 % of the total GHG emissions of the life cycle.…”

“…2 is 47.5 kg CO 2 -eq. Taking the average of the total GHG emissions from the studies by Weber (2012) and Stutz et al (2012) to be 6484 kg CO 2 -eq, the average emission relating to a smartphone equates to approximately 1/140 of this. This is significantly larger than the 1/400 estimated by Bryant (2013).…”

“…They report total GHG emissions of 6360 kg CO 2 -eq for a Dell PowerEdge R710 (Stutz et al 2012), assuming use in the US and associated electricity grid mix, and a mean of 6607 kg CO 2 -eq (Weber 2012) over the lifetime of a study looking at the variation in life cycle GHG emissions from servers. Weber (2012) used regional electricity mixes.…”

Uncertainty Quantification in Life Cycle Assessments: Interindividual Variability and Sensitivity Analysis in LCA of Air‐Conditioning Systems

“…; Noshadravan et al. ; Weber ). In highly complex system models, application of Monte Carlo analysis can lead to excessive computational burden, necessitating development of surrogate models (Allaire and Willcox ).…”

“…Lloyd and Ries (2007) noted that application of some techniques may lead to unreliable outcomes and overconfidence in results; however, stochastic modeling is effective for assessing the probability of outcomes in LCA. Application of Monte Carlo analysis for stochastic modeling, implementing probability distributions in place of static input parameters, has been successfully implemented in many studies (Mattinen et al 2014;Niero et al 2014;Noshadravan et al 2015;Weber 2012). In highly complex system models, application of Monte Carlo analysis can lead to excessive computational burden, necessitating development of surrogate models (Allaire and Willcox 2010).…”