Companies use life cycle assessment and corporate carbon accounting to understand the emissions associated with the products they make, the services they provide, and the goods they purchase. These methods help sustainability teams measure impacts, identify hotspots, set targets, and decide where better data or emissions reductions can have the greatest effect.
But the method used to estimate a footprint can change both its size and its apparent source. Product-level assessments may follow materials and processes through a defined life cycle. Corporate inventories organize emissions into scopes and categories. Spend-based methods use industry-average emissions factors when supplier-specific activity data is unavailable. Environmentally extended input-output models connect economic transactions to emissions, while multiregional models also account for where production occurs and how goods move through international supply chains.
Our research examines how these choices affect corporate Scope 3 estimates. In particular, it shows why accounting for the regions where goods and services are produced can materially improve the accuracy and usefulness of corporate footprints.
For related work, see our research brief on open data and our research brief on AI for sustainability.
Watershed research on life cycle assessment and corporate footprints
Corporate accounting: Why regional resolution matters for Scope 3
The importance of multiregional accounting for corporate carbon emissions Nature Communications, 2026. This study compares single-region and multiregional environmentally extended input-output models for estimating upstream corporate emissions. Using the Comprehensive Environmental Data Archive (CEDA), the researchers assess how regional and sectoral resolution changes emissions estimates, hotspot rankings, and priorities for supplier data collection and decarbonization.
The multiregional model covers approximately 400 industry sectors across 65 countries representing more than 90% of global GDP, plus a rest-of-world region. The study includes authors from Stanford University, Watershed, World Wildlife Fund, and CDP.
Data and methods: Building more representative footprint models
CEDA is a multi-region environmentally extended input-output database for spend-based emissions data. The latest version includes approximately 60,000 emissions factors across 400 industries and 148 countries and regions, covering 95% of global GDP.
The Open CEDA database makes global spend-based emissions data available for broader use. The Cornerstone U.S. model methodology documents how the U.S. model is built, including its relationship to USEEIO and CEDA.
Product footprints: Testing each step of the assessment
PCFBench: A diagnostic benchmark for product carbon footprint estimation tests how AI systems perform each step of building a product carbon footprint, from identifying materials to matching them with emissions data. The open benchmark includes 614 expert-labeled examples, along with a public dataset and evaluation code.
Our evaluation criteria for AI-assisted product carbon footprinting systems describe the evidence a credible system should provide, including benchmark performance, uncertainty indicators, and an inspectable record of how it reached its conclusions.
Hybrid methods: Combining industry averages with supplier data
The paper’s findings connect to research on hybrid approaches in life cycle assessment, which combine input-output data with process-level and supplier-specific data. Hybrid methods can reduce the risk of missing upstream emissions while focusing detailed data collection on the activities that contribute most to uncertainty.
For additional work on product footprint data quality and emissions-factor mapping, see Quality-aware automation for LCI database mapping and our conversation with ecoinvent on getting to more actionable product footprints.
Key findings
The findings below come from The importance of multi-regional accounting for corporate carbon emissions. The study compares a U.S.-based single-region EEIO model with a multi-regional CEDA model that represents production across 65 countries and approximately 400 industry sectors.
Single-region models can underestimate upstream emissions
- Multiregional accounting estimated 2.0 gigatons more emissions. Across companies reporting to CDP, the multi-regional model estimated 2.0 gigatons of CO₂-equivalent emissions more than the comparable U.S.-based single-region model, an increase of approximately 9.7%.
- The difference reflects production and trade, not changes in company activity. The two models use the same general accounting framework. The key difference is that the multi-regional model accounts for where goods and services are produced and how they move through international supply chains.
- The aggregate estimate may include multiple counting. The same emissions can appear in the Scope 3 inventories of multiple companies connected through a value chain, so the 2.0 gigaton figure should not be interpreted as a unique global emissions total.
The effect is largest in manufacturing and international supply chains
- Manufacturing sectors showed the largest increases. Multi-regional estimates were 71.3% higher for structural products, 69.7% higher for construction machinery, 50.6% higher for fabricated metal products, and 39.3% higher for electronic components.
- The difference was smaller in some service and consumer product sectors. The multi-regional model estimated increases of 10.1% for business support services, 10.5% for truck transportation, 13.3% for software publishing, and 6.8% for financial investments. Differences were 2.2% for chemical products and 1.2% for plastics.
- China was the largest source of additional emissions identified. The multi-regional model attributed approximately 973 million additional tonnes of CO₂e to China, reflecting both the country’s role in global manufacturing and the emissions intensity of production in some sectors.
Regional and sectoral resolution both matter
- Emissions intensity varies widely across regions. For key supplier sectors, the standard deviation of emissions intensity across regions exceeded 180 grams of CO₂e per dollar. A U.S.-based emissions factor can therefore substantially overestimate or underestimate the intensity of production elsewhere.
- The largest differences were concentrated in moderate-intensity sectors. Across sectors with multi-regional emissions intensities of approximately 0.4 to 0.8 kg CO₂e per dollar, multi-regional estimates were roughly 30% higher than single-region estimates.
- Single-region models do not always underestimate. The multi-regional estimate was higher for most sectors, but the single-region estimate was higher for 29 of the 400 sectors studied. These differences can occur where production relies more heavily on low-carbon energy sources, such as nuclear power, hydropower, or bioenergy.
Model choice can change emissions reduction priorities
- Multi-regional accounting changes the ranking of upstream hotspots. The model did not simply increase the emissions associated with existing hotspots. It often changed which sectors appeared to contribute most to a company’s upstream footprint.
- Additional hotspots appeared in energy-intensive manufacturing. The multi-regional results increased the prominence of manufacturing and equipment-related sectors, including emissions associated with steel, iron, and resin production in some regions.
- The model can change where companies focus primary data collection. A company using a single-region model may prioritize the wrong suppliers or sectors for more detailed data. Multi-regional accounting can redirect attention toward international suppliers and production processes with higher emissions intensity.
Better accounting does not eliminate the need for primary data
- EEIO models provide industry averages. They are useful when supplier-specific activity data is unavailable, but they do not describe the performance of an individual supplier or facility.
- Hybrid methods can improve accuracy. Supplier-specific activity data can be used to replace or refine industry-average estimates, especially for the largest upstream hotspots.
- More detail does not resolve every source of uncertainty. Land-use change, data quality, sector definitions, allocation choices, and the distinction between attributional and consequential accounting can all affect results.
What this means for sustainability teams
Use the method that matches the decision. Product carbon footprints, corporate inventories, supplier assessments, and consequential studies answer different questions. Before comparing results, document the goal of the analysis, the system boundary, the accounting approach, the data sources, and the level of geographic and sectoral detail.
Treat geographic resolution as a material modeling choice. If a company purchases goods from international suppliers but applies a single-region emissions factor, its Scope 3 estimate may not reflect the emissions intensity of actual production. Multi-regional models can provide a more representative starting point for global supply chains.
Use secondary data to prioritize primary data collection. Spend-based and EEIO estimates can identify likely hotspots across thousands of suppliers. Companies can then focus supplier engagement and activity-data collection where better information is most likely to change decisions. Research on hybrid LCA methods suggests that improving data for a relatively small number of high-uncertainty activities can substantially improve the overall result.
Track model versions and rebaseline carefully. A change from a single-region to a multiregional model can increase reported emissions without any change in the company’s operations or supply chain. Recalculate prior years using the same method when possible, and clearly separate methodological changes from operational reductions. The CEDA 2024 update provides an example of how changes in underlying emissions data can affect footprint calculations.
Evaluate product footprint systems step by step. A final product footprint can appear reasonable even when errors occur in the bill of materials, material mapping, emissions-factor selection, or supply chain model. PCFBench and the evaluation criteria for AI-assisted product carbon footprinting provide frameworks for assessing the reliability of these intermediate steps.
Pair attributional accounting with action-oriented analysis. Attributional footprints describe emissions associated with a company’s existing products and supply chain. They do not necessarily predict how total global emissions will change if a company switches suppliers, changes products, or shifts demand. For major procurement and design decisions, consequential analysis can provide an important complement.
About the researchers
Steven J. Davis: Professor of Earth System Science in the Stanford Doerr School of Sustainability and leads the Sustainable Solutions Lab, a research group dedicated to quantifying how different human activities are affecting climate and air quality, how those environmental changes in turn jeopardize human wellbeing, and the relative priority of solutions.
Andrew Dumit: AI Engineer. Builds AI systems that measure product carbon footprints and supply chain emissions. Research: Quality-aware automation, ATLAS, PCFBench, Evaluation Criteria; Build evals you can sustain.
Mo Li: Climate Science Lead at Watershed. She leads CEDA and works on the Cornerstone Sustainability Data Initiative, an open collaboration with Stanford and ERG. She earlier helped develop open-source data infrastructure for the U.S. EPA's USEEIO modeling framework. She holds a Ph.D. and M.S. in Natural Resources Science and Management from the University of Minnesota.
Yohanna Maldonado: Director of the Cornerstone Sustainability Data Initiative and Head of Climate Data at Watershed.
Sangwon Suh: Tsinghua University, research advisor to Watershed. Research: Evaluation Criteria; ATLAS; emissions reduction levers library.
The study was led by researchers from Stanford University and Watershed, with contributions from World Wildlife Fund and CDP. Read the full paper, “The importance of multi-regional accounting for corporate carbon emissions,” in Nature Communications.
