Two Giant Green Databases, One Answer: Carbon Footprint Uncertainty Holds Steady Across Scales

Every time a company claims its product is greener than a rival’s, or a government bets public money on a new technology because its carbon footprint looks small, an invisible army of numbers stands behind that claim. Life cycle assessment, the accounting framework that tallies the environmental costs of everything from sewage sludge to smartphones, depends on vast background databases describing the emissions and resources embedded in electricity, steel, transport and thousands of other inputs. A new study published in the Journal of Industrial Ecology by Elisabetta Pigni of the University of Bologna and colleagues at Aalborg University tackles a question that has quietly nagged the field for years: when you calculate a carbon footprint with one of these databases instead of another, how much of the difference is real, and how much is simply uncertainty propagating through the mathematics?

The team focused on two of the most widely used data sources in the discipline, which approach the economy from fundamentally different angles. Ecoinvent is a process-based database, built bottom-up from detailed measurements of individual industrial activities, supply chains traced link by link. Exiobase, by contrast, is a multiregional input-output database, built top-down from national economic accounts and environmental statistics, capturing the entire global economy in one enormous matrix of monetary flows between sectors and countries. Process data are typically more precise for the activities they cover but risk truncation, missing distant upstream links; input-output data are complete by construction but highly aggregated, lumping many heterogeneous activities into single sectors. Analysts have long suspected that these structural differences translate into different levels of uncertainty in the final results, but rigorous, like-for-like comparisons have been scarce.

To make such a comparison possible, the researchers first had to solve a thorny problem: exiobase, unlike ecoinvent, does not ship with ready-made uncertainty estimates for its thousands of data points. They therefore tested two approaches to estimating data uncertainty in exiobase. The first assumed a uniform level of uncertainty across the entire database, a simple but crude benchmark. The second applied the pedigree matrix method, a well-established scheme in life cycle assessment that scores each data point on qualitative characteristics such as reliability, completeness, temporal correlation, geographical correlation and further technological correlation, and converts those scores into quantitative uncertainty factors. The pedigree approach, originally developed for ecoinvent, allowed the team to assign differentiated, defensible uncertainty distributions to exiobase entries rather than painting the whole database with one brush.

A second layer of uncertainty arises when a specific product system, the so-called foreground, is connected to the background database. Matching the activities of a novel technology to representative background datasets involves judgment calls, and each mismatch introduces model uncertainty. The researchers estimated this matching uncertainty too, again using the pedigree matrix, so that both the data quality of the background numbers and the quality of the coupling between foreground and background entered the final error budget. This dual treatment is one of the study’s technical contributions, because most previous comparisons of the two database families considered data uncertainty alone and ignored the additional noise injected by the matching step.

The case study chosen to stress-test the framework was an emerging wastewater treatment technology developed under the European ToSynFuel project, which turns sewage sludge into biofuels and hydrogen. Emerging technologies are exactly where uncertainty matters most: there are no mature inventories for them, foreground data are provisional, and the choice of background database can swing the headline climate impact. The team built the technology’s life cycle model and then ran stochastic error propagation, using Monte Carlo simulation in the open-source Brightway Python framework, to compute climate impact scores as probability distributions rather than single numbers. Ad hoc simulations were designed for both databases, drawing thousands of random samples from the uncertainty distributions of the underlying inventory data and propagating them through the full linear model each time.

The results are striking in their symmetry. When uncertainty in exiobase was estimated with the pedigree matrix, the analysis confirmed a proportional relationship between input and output uncertainty: uncertainty in the database entries propagates through the input-output calculations in a predictable, roughly proportional fashion. That finding matters because it suggests the aggregation embedded in input-output tables does not wildly amplify or dampen error in the way some analysts feared. The error structure of a highly aggregated global database turns out to be more tractable than its sprawling matrix appearance might suggest.

Even more consequential is the head-to-head comparison of the final climate scores. Calculated through exiobase, the wastewater treatment technology’s climate impact came out with a median of 0.146 million tonnes of carbon dioxide equivalent, with a 5th to 95th percentile range of 0.140 to 0.150 million tonnes. Calculated through ecoinvent, the median was 0.148 million tonnes of carbon dioxide equivalent, with a range of 0.143 to 0.152 million tonnes. The two distributions overlap almost completely. The medians differ by less than two percent, and the uncertainty bands are nearly identical in width. On that basis the authors ruled out the hypothesis that, when pedigree-based uncertainty estimates are used, one database systematically yields more uncertain results than the other, at least for this technology and this impact category.

For practitioners, the practical message is one of reassurance with caveats. Analysts who prefer ecoinvent for its process-level detail and those who turn to exiobase for its economy-wide completeness can now point to evidence that, once uncertainty is handled consistently with the pedigree approach, the two routes deliver climate impact estimates of comparable reliability. The choice between them can be driven by scope, coverage and the question at hand, rather than by fear that one path is statistically shakier. The study also supplies the tools for others to repeat the exercise: the scripts used in the work are archived on Zenodo, and the team released a general-purpose library, hosted on GitHub, that performs the same type of computation on other cases, potentially lowering the barrier for uncertainty analysis across the field.

The work also feeds a broader methodological debate about how uncertainty should be handled in comparative life cycle assessment. The authors engage with a literature that questions whether classical statistical testing and confidence intervals are appropriate at all when results come from Monte Carlo simulations of overlapping distributions, citing arguments that such tests can mislead when distributions are not independent. By reporting medians and percentile ranges rather than verdicts of statistical significance, the study models a reporting style that lets readers judge the overlap for themselves. It likewise builds on a lineage of work stretching from early error analyses of input-output inventories to recent estimates of greenhouse gas uncertainty in global multiregional models, positioning itself as the first, or among the first, to propagate uncertainty through both database families at different aggregation levels within a single consistent framework.

Limitations remain, and the authors are careful about scope. The comparison rests on one emerging technology and one impact category, climate change, so extending the conclusions to other technologies, to toxicity or water impacts, or to other database versions will require further work. The pedigree scores themselves embed expert judgment, and different scoring choices could shift the distributions. Yet the core result stands as an unusually clean empirical answer to a question the field has debated largely on theoretical grounds: at different levels of aggregation, from sector-level economic matrices to plant-level process chains, the uncertainty of a carbon footprint is less a property of the database you choose and more a property of how honestly you characterize the numbers inside it. For a discipline whose outputs increasingly steer billions of euros of investment, that is a finding worth propagating.

Subject of Research: Uncertainty propagation in life cycle assessment using input-output and process-based inventory databases

Article Title: Uncertainty propagation of input–output and process-based life cycle inventories at different aggregation levels

Article References: Pigni, E., An, N., Righi, S., Marazza, D., Balugani, E., & Pizzol, M. (2026). Uncertainty propagation of input–output and process-based life cycle inventories at different aggregation levels. Journal of Industrial Ecology, 30(4), 1671-1682. https://doi.org/10.1007/s44498-026-00113-0

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00113-0

Keywords: life cycle assessment, uncertainty propagation, exiobase, ecoinvent, Monte Carlo simulation, pedigree matrix, input-output analysis, carbon footprint, wastewater treatment, Brightway, climate impact, Journal of Industrial Ecology