Enviroscore: normalization, weighting, and categorization algorithm to evaluate the relative environmental impact of food and drink products

Food systems are major drivers of environmental degradation, contributing significantly to climate change, water scarcity, acidification, eutrophication, and resource use. In the EU, food production accounts for an estimated 20–30% of overall human-caused environmental impacts. The sector is heterogeneous, with environmental performance varying widely by origin, production practices, company size, seasonality, and supply-chain logistics. Life Cycle Assessment (LCA) is a robust method to evaluate environmental impacts across the life cycle of products and identify improvement opportunities. However, communicating LCA results to consumers faces several limitations: communication often focuses on a single impact (e.g., carbon footprint), methods are frequently product-specific limiting cross-product comparisons, and existing labels sometimes lack robust, peer-reviewed methodologies. There is also limited consumer trust in environmental claims. To address these issues, normalization and weighting across multiple impact categories are needed to produce a single index that is interpretable, comparable, and science-based. The EC’s Product Environmental Footprint (PEF) proposes an aggregated Single Score across goods and services, but this broad scope reduces sensitivity for benchmarking food products specifically. Therefore, this study aims to develop food- and drink-specific normalization and weighting factors to create a European Food Environmental Footprint Single Index (EFSI) and to define thresholds for a 5-level Enviroscore (A–E) label to communicate relative environmental impact, motivate sustainable choices by consumers, and encourage improvements by agri-food businesses.

The paper situates its work within literature on LCA-based communication and labeling. Prior studies show carbon labeling can reduce the carbon footprint of food purchases, but single-impact communication risks burden shifting to other categories. Product-specific labels (e.g., milk carbon labels) limit comparability across products and categories. Existing systems like Eco-Score, while grounded in LCA averages, add bonus–malus heuristics (e.g., origin, certifications) and lack peer-reviewed validation. Consumers report limited trust in environmental performance claims. The authors highlight the need for scientifically robust normalization and weighting across multiple impact categories, referencing the EC’s PEF Single Score approach and its limitations for food-specific benchmarking. This motivates developing a food-basket-referenced single index and categorical score for clearer consumer communication.

Study design followed a stepwise approach: (1) development of a food-specific single index (EFSI), (2) validation of variability captured by EFSI versus EC Single Score, and (3) definition and validation of Enviroscore thresholds.

1. Development of EFSI

• Reference system: European Food Basket representing ~90% of EU-28 food supply (FAO Food Balance Sheets, 2013), plus packed water; 23 representative food items (N1=23). Each item reflects common practices in origin, production, distribution, consumption, and end-of-life.
• Inventory and modeling: Product Environmental Footprint (PEF) methodology was applied. Data sources included Ecoinvent 3.5 and Agri-footprint datasets for primary production; processing data from literature and databases; packaging from Eurostat/industry sources; international distribution modeled when >20% imported (routes via Rotterdam/Frankfurt/Rungis as relevant); national distribution assumed 500 km by lorry; retail, consumer, and end-of-life stages from sector rules and guidance; background datasets for energy, transport, waste from Ecoinvent 3.5. EC PEFCR-based LCA results were used for certain items (milk, packed water, dry pasta, beer, wine). Modeling used SimaPro 9.0 with ILCD-recommended impact categories.
• Impact assessment and normalization: 13 ILCD impact categories included: climate change, ozone depletion, ionizing radiation, photochemical ozone formation, respiratory inorganics, acidification (terrestrial and freshwater), eutrophication (freshwater, marine, terrestrial), land use, water scarcity, resource use (energy carriers), and resource use (minerals and metals). Normalization factors (per capita) were calculated using the European Food Basket characterization results: NF(f) = Σ[FC(i) × e(fi)] / Population, where FC(i) is annual EU consumption of food i, e(fi) is impact per kg for category f, and Population is EU-28 population.
• Weighting: EC-recommended weighting factors (Sala et al., 2018) were applied, excluding toxicity-related categories due to methodological uncertainties. Aggregation of normalized, weighted impacts yielded the EFSI, a dimensionless single index relative to average per-capita European food consumption impacts.

2. Relative validation of EFSI

• Dataset: 21 food products with 149 hypothetical items (N2=149) representing variations in origin, transportation, and production methods relevant to the EU market. Detailed inventories provided in supplementary materials.
• Analyses: Distribution of EFSI versus EC Single Score across items via boxplots (median, IQR) to assess between- and within-product variability. Correlation heatmap and matrix computed between impact categories and each single index to assess sensitivity to individual impacts and inter-impact correlations (RStudio and MATLAB used; customizable heatmaps toolkit).

3. Enviroscore thresholds and validation

• Threshold setting: Based on the distribution of EFSI, assembled an additional dataset of 22 representative food items (N3=22), including 12 from the European Food Basket plus 10 items reflecting growing market trends. Thresholds for five categories (A–E) were defined to reflect distribution patterns: A <4×10^-4; B ≥4×10^-4; C ≥1.45×10^-3; D ≥2.00×10^-3; E ≥1.00×10^-2.
• Accuracy validation (Delphi): Seven LCA/food-environment experts participated in a three-round Delphi (Feb–Apr 2019). Experts categorized 22 items into five impact levels; items retained when ≥80% agreement. Agreement between Enviroscore and expert categorization assessed via weighted Kappa and contingency tables.
• Performance validation: Compared Enviroscore of each representative product to its hypothetical variants (N2=149; overlapping items N=130 after exclusions). Agreement and deviations analyzed (including impact of transport mode, especially air), with weighted Kappa statistics and scenario excluding air transport.
• Software and stats: SimaPro 9.0 for LCA; RStudio (v1.1.463) and MATLAB 2017b for statistics and visualization.

• Food-basket characterization and weights: Animal-based items are 28% of total food consumption yet contribute ~37% of environmental impact in the European Food Basket. Within animal-based foods, milk is most consumed (27%), while beef contributes the highest environmental impacts (31% of total impact reported in the study context).
• EFSI variability across products: EFSI values differentiated product groups clearly. Plant-based products had lower EFSI (median 1.30, IQR 1.81) than animal-based products (median 2.47, IQR 4.21). Lowest median EFSI: sugar from sugar beet (0.379, IQR 0.197). Highest: beef (11.51, IQR 4.48). EC Single Score showed a similar trend but different rankings and lower within-product variability than EFSI.
• Correlation analysis: Overall EFSI and EC Single Score correlation was high (r=0.79). EFSI had medium correlations (≤0.67) with most impact categories except water scarcity (high, r=0.75). EC Single Score correlated strongly with several categories (e.g., climate change r=0.90; acidification r=0.91; eutrophication categories r=0.75–0.90) but weakly with water scarcity (r=0.27). Water scarcity showed low correlation with most other impacts, underscoring its distinctiveness and importance for foods.
• Country of origin and water stress: EFSI better captured differences related to water stress than EC Single Score (example correlation with water stress differences: EFSI r=0.624 vs EC Single Score r=0.228).
• Enviroscore thresholds and examples: Thresholds categorized EFSI into A (<4×10^-4), B (≥4×10^-4), C (≥1.45×10^-3), D (≥2.00×10^-3), E (≥1.00×10^-2). Examples: A (very low) includes orange, rye flour, soybean beverage; B includes pasta, grapes, potato; C includes fruit juices, refined sunflower oil; D includes avocado, chicken meat, pig meat; E includes beef, canned tuna.
• Accuracy vs expert consensus: Weighted Kappa for Enviroscore vs Delphi experts was 0.642 (p=0.0025), indicating good agreement. Perfect agreement (100%) for categories A and E; intermediate categories showed some misclassification (e.g., B and D correspondence 64% and 57%, respectively; C had no agreement in the contingency table example), typically within one level. Notable outlier: strawberry (B by experts vs D by Enviroscore).
• Performance vs product variants: Agreement between representative products and their hypothetical variants yielded weighted Kappa 0.45 (p<0.05). Best agreement for B (62%) and D (78%); more deviations for A (17%) and C (25%). Small one-level deviations were influenced by production method (27%) and transport (19%). Large deviations (≥2 levels) were mainly due to air (66%) and long-distance terrestrial transport. Excluding air-transported items improved agreement (weighted Kappa 0.71).
• Sensitivity to transport, production, and origin: Enviroscore captured differences by transport mode (local vs international air), production environment (greenhouse vs field), and origin (water-stressed countries), and distinguished beef from dairy-cattle vs beef-cattle systems.

The study demonstrates that a food-specific normalized and weighted single index (EFSI) can effectively aggregate multi-impact LCA results while retaining sensitivity to differences relevant for foods (e.g., water scarcity, transport mode, production system). Compared with the EC-wide Single Score, EFSI better reflects within- and between-product variability in foods, largely due to referencing normalization to the European Food Basket and capturing distinct impacts like water scarcity that are critical for agriculture and livestock. The Enviroscore thresholding translates complex LCA outputs into an intuitive five-tier label, facilitating consumer understanding and potential behavior change while enabling industry benchmarking. The high agreement with expert categorization (Kappa 0.642) supports the label’s face validity, and performance analyses confirm that the scoring system is responsive to real-world variations in origin, transport, and production practices. Policy-wise, the approach aligns with the EU’s PEF initiative, providing a communication layer currently lacking, and may incentivize environmental improvements in the food sector.

This work develops and validates the European Food Environmental Footprint Single Index (EFSI) and an associated five-level Enviroscore for food and drink products. Using European Food Basket-based normalization and EC-recommended weighting, EFSI aggregates 13 ILCD impact categories into a single, comparable metric that better captures food-specific variability (notably water scarcity) than the generic EC Single Score. Thresholds for Enviroscore (A–E) translate EFSI into an accessible label with good expert agreement and demonstrated sensitivity to differences in origin, transport, and production. The method offers a transparent, science-based framework for front-of-pack communication and industry benchmarking, potentially guiding consumers toward lower-impact choices and stimulating supply-chain improvements. Future research should broaden case studies across more products and datasets, refine thresholds as markets evolve, and explore integration with policy and digital product information systems.

• Reference basket composition: Normalization is based on a specific European Food Basket selection (e.g., excludes cheese, butter, salmon) which can influence category magnitudes, though differences versus another published EU basket were small for most categories; larger differences observed for some (e.g., terrestrial eutrophication, ionizing radiation, photochemical ozone formation) due to product selection.
• Impact category scope: Toxicity-related categories were excluded due to methodological uncertainties in weighting; results thus do not reflect these impacts.
• Water scarcity treatment: While a strength for food relevance, water scarcity is a stand-alone, weakly correlated category; differences with EC Single Score arise largely from stronger EFSI sensitivity to water scarcity.
• Functional unit: A uniform mass-based functional unit of 1 kg was used across products for comparability; alternative functional units (e.g., portion, nutrient-based) could yield different comparisons.
• Data and representativeness: Limited number of products and scenarios; hypothetical variants may not capture full market diversity. Some parameters (e.g., transport distances/modes, retailer/consumer behaviors) rely on generic assumptions.
• Communication thresholds: Thresholds are derived from observed distributions; intermediate categories showed more disagreement with experts, and certain products (e.g., strawberry) deviated notably.
• Transport influence: Air transport strongly affects scores; market dynamics in logistics could change performance over time.