Multi-model assessment identifies livestock grazing as a major contributor to variation in European Union land and water footprints

The study addresses how the choice of accounting methods and models influences estimates of the EU’s land footprint (LF) and water footprint (WF) of food consumption—key indicators for policies under the European Green Deal and the Farm to Fork Strategy. Environmental footprints can be computed via process-based and environmentally extended multi-regional input–output (EE-MRIO) approaches, but methodological decisions and assumptions, especially regarding livestock grazing, yield divergent results. The authors aim to quantify the EU27’s LF and blue/green WF of food (and contrast with non-food uses) using multiple models and to harmonize the treatment of grazing land by aligning LF accounting with the established WF approach that counts only grass eaten. They further place EU consumption in the context of global agricultural land and water use and evaluate model differences by product groups and regions of origin.

Prior work has shown that environmental footprint estimates vary across modeling frameworks (process-based vs MRIO) and due to assumptions like proportionality between monetary and physical flows, sector aggregation, and discrepancies between physical agricultural statistics and macroeconomic accounts. MRIO models such as EXIOBASE and FABIO have been widely used to trace embodied resource flows, yet their robustness for physical biomass flows is debated. Water footprint assessments commonly count only grass actually eaten by livestock, whereas land-use accounting often attributes entire grazing areas to livestock, creating inconsistencies across footprint families. The paper situates its contribution within this debate, referencing studies that argue for moving towards more detailed physical data, highlight the importance of product and country disaggregation, and document the sensitivity of land-use footprints to model structure and classifications.

Study scope and data: The analysis targets the EU27 in the reference year 2012 (population 440,905,186), harmonizing inputs across models using FAOSTAT data. Footprint definitions: LF quantifies cropland and grazing land use; WF quantifies consumptive use of blue water (surface/groundwater) and green water (soil moisture from precipitation). Seven model variations were applied to track EU consumption through global trade to final demand: one physical bilateral trade model (PHYS) and three MRIO models with two variants each—EXIOBASE-min and EXIOBASE-max (monetary MRIO), FABIO-mass and FABIO-value (physical MRIO with alternative by-product allocation), and HYBRID-mass and HYBRID-value (mixed-unit MRIO integrating FABIO and EXIOBASE). PHYS uses physical bilateral trade flows (191 primary agricultural products, 223 countries) with origin tracing and conversion to primary equivalents; EXIOBASE v3.6 covers 200 products/services (19 agricultural) for 44 countries and 5 rest-of-world regions; FABIO v1.1 covers 125 agriculture/food products for 191 countries; HYBRID integrates FABIO for agri-food physical flows and EXIOBASE for non-food/service monetary flows. Model variants: EXIOBASE-min classifies as food only clearly food-related product groups; EXIOBASE-max additionally includes product groups/services that may include food. FABIO variants differ by mass vs value allocation for by-products such as soybean oil and cake. Harmonized grazing LF: The LF for grazing follows WF standards—only grass eaten is counted. Required grazed biomass is translated into area using country-specific grassland productivity derived by overlaying a spatially explicit pastureland layer with vegetation productivity, assuming a maximum 75% of aboveground NPP utilisable by livestock; productivity averages per country then convert grazed biomass to area. In very low productivity contexts, the grazing LF was capped at 80% of available land. This yields a hypothetical area equivalent if grazing lands were used at maximum intensity given current productivity. LF in PHYS: Trade flows (FAOSTAT) are converted to primary crop equivalents; re-exports are handled; feed embodied in traded animal products is computed using FAO commodity balances; primary product flows (tonnes) are converted to harvested area via producing-country yields. Blue/green WF accounting: Crop WF (blue and green) for food and feed uses comes from Mekonnen and Hoekstra’s international database (analyzed for 1996–2005). Green WF of grazing uses country-average ET from grazed pastures (assumed fully green) averaged over 2000–2009 (Schyns et al.) simulated with LPJmL under grazing density that maximizes grass yield over grazed areas (permanent meadows/pastures minus harvested fodder grasses, in grid cells with mapped grazing livestock). Attribution to final demand: Footprints are tracked through supply chains to final products/services, distinguishing food vs non-food uses. Non-food uses are assessed for context (biofuels, hides, textiles, detergents, hotel/restaurant services, etc., model-dependent). The study reports totals and disaggregation by product groups and by origin regions.

- EU food consumption LF: 140.3–222.4 Mha yr−1 (0.318–0.504 ha person−1 yr−1). Six models cluster at 140.3–152.9 Mha yr−1; EXIOBASE-max is the outlier at the high end. - EU non-food LF: Very high in EXIOBASE (133.3 Mha yr−1 in EXIOBASE-min; 63.7 Mha yr−1 in EXIOBASE-max), yielding total LF for EXIOBASE around 286.1 Mha yr−1 (0.649 ha person−1 yr−1). HYBRID and FABIO non-food LF are 21.5–55.9 Mha yr−1; PHYS 13.3 Mha yr−1. - EU food total WF: 569.3–917.8 km3 yr−1 (3,538–5,703 l person−1 d−1). Six models cluster at 569.3–674.3 km3 yr−1 (3,538–4,190 l person−1 d−1); EXIOBASE-max is highest. - EU non-food total WF: EXIOBASE 468.3 km3 yr−1 (min) and 222.4 km3 yr−1 (max), giving total WF 1,140.2 km3 yr−1 (7,085 l person−1 d−1). FABIO/HYBRID 120.9–213.6 km3 yr−1; PHYS 87 km3 yr−1 (540 l person−1 d−1). - EU food blue WF: 29.7–70.0 km3 yr−1 (184–435 l person−1 d−1); PHYS lowest (excludes processing and livestock drinking water), EXIOBASE highest. FABIO/HYBRID ~41.2–41.4 km3 yr−1 (~256–257 l person−1 d−1). - EU non-food blue WF: EXIOBASE 43.8 (min) and 24.8 (max) km3 yr−1, totaling 94.7 km3 yr−1 (589 l person−1 d−1); HYBRID 13.9–15.4; FABIO 4.8–5.1; PHYS 3.7 km3 yr−1. - Product group contributions: Animal products dominate LF—61% (PHYS), 67% (FABIO-mass/HYBRID-mass), 66% (FABIO-value/HYBRID-value). For blue WF, animal products contribute relatively less (PHYS 39%; EXIOBASE-min 20%; EXIOBASE-max 15%; FABIO/HYBRID 41%). Vegetables, fruit and nuts account for a large share of blue WF (PHYS 28%; EXIOBASE-min 36%; EXIOBASE-max 26%; FABIO/HYBRID 21%). - Origin of impacts: In PHYS/FABIO/HYBRID, EU-produced shares are high—LF 64–75%, blue WF 71–74%, green WF 58–64%. Imports contributing most to LF/green WF originate from Latin America (via animal products, coffee, cocoa). For blue WF, main imports come from Asia (pork, rice). EXIOBASE shows much lower EU-produced shares (LF 47–53%, blue WF 42–47%, green WF 46–50%) and very large imports from Asia and Africa, with substantial contributions from ‘Food products nec’ and ‘Hotel and restaurant services’ categories. - Global LF: Six models (excluding PHYS) estimate 3,014 Mha yr−1 total (cropland 1,357; grazing 1,657). PHYS estimates 2,809 Mha yr−1 (cropland 1,397; grazing 1,411). These totals are 59–63% of FAOSTAT’s 2012 agricultural land (4,773 Mha). Cropland is consistent with prior estimates (1,200–1,621 Mha), while grazing LF is considerably lower than FAOSTAT permanent pastures due to the ‘grass eaten’ approach. - Policy context: EU food LF corresponds to 5–7% of global agricultural LF and EU food WF to 6–10% of global agricultural WF, with the EU comprising ~6% of global population.

The multi-model assessment demonstrates that methodological choices significantly affect footprint estimates and their interpretation. While six model variants converge for EU LF and green WF totals, EXIOBASE-max yields substantially higher values because it attributes footprints from product groups and services that may include food, likely overestimating food-related footprints. Blue WF estimates vary more across models, with PHYS lower (exclusion of processing and drinking water) and EXIOBASE higher. Differences by product category and region of origin are pronounced due to divergent system boundaries, product classifications, and allocation rules. Linking livestock feed to animal products is a major uncertainty and differs across models (e.g., FABIO’s region- and species-specific feed requirements vs PHYS’s global weighting based on US systems). The harmonized grazing LF, aligned with WF practice, reduces over-attribution of extensive/low-intensity grazing areas by counting only grass eaten and translating to a hypothetical area based on productivity. This approach yields lower grazing LF than FAOSTAT’s permanent pasture area and brings LF and WF into a consistent footprint family framework. For policy, especially monitoring Farm to Fork Strategy targets, consistent model selection (or explicit multi-model assessments) and harmonized accounting are essential. Increased product and country resolution reduces uncertainty, enabling more reliable scenario analyses for diet shifts and food loss and waste reduction.

The study provides a harmonized, multi-model quantification of the EU27’s land and water footprints of food consumption and a novel, consistent approach to accounting for grazing land based on grass actually eaten. EU food LF is 140–222 Mha yr−1 and WF 569–918 km3 yr−1; globally, agricultural LF is 2,809–3,014 Mha yr−1 with grazing accounting for 1,411–1,657 Mha yr−1. Results highlight that model choice and accounting assumptions, particularly for livestock feed and grazing, drive substantial variation in footprint outcomes and regional/product attributions. For robust monitoring and policy design under the European Green Deal’s Farm to Fork Strategy, using consistent methods or a transparent multi-model approach is critical. Future work should: (1) increase product and country disaggregation (including processed products) in MRIO systems; (2) improve representation of livestock feed mixes and supply chains; (3) regularly update models to recent years; and (4) apply standardized methods to scenario analyses of healthy sustainable diets and reductions in food loss and waste.

Key limitations include differences in system boundaries and assumptions across models (monetary vs physical MRIO; treatment of non-food supply chains; inclusion of live animal trade; aggregation of product categories) that affect comparability and attribution. PHYS currently excludes water for food processing and livestock drinking, lowering blue WF. The WF crop dataset (1996–2005) and grazing WF data (2000–2009) are temporally mismatched with the 2012 FAOSTAT base year. Linking livestock feed to products is uncertain and model-specific (e.g., PHYS’s global weighting may over-assign feed to beef in the EU; FABIO’s ‘Meat other’ attribution may include horse feed). The grazing LF is hypothetical, based on assumptions (75% utilisable NPP; cap at 80% of available land) and does not measure actual grazed area or land-use change. The EXIOBASE product aggregation creates uncertainty in classifying food vs non-food (large spread between EXIOBASE-min and EXIOBASE-max). Analysis is limited to 2012; dynamic changes over time are not assessed.