Food systems are major drivers of environmental pressures, particularly concerning land and water resources. The European Green Deal and its Farm to Fork Strategy aim for a sustainable food system, necessitating accurate quantification of land and water footprints (LF and WF) for effective policymaking. Existing methodologies for calculating LF and WF vary significantly, leading to discrepancies in results. Different approaches, ranging from process-based methods to environmentally extended multi-regional input-output (EE-MRIO) models, yield diverse outcomes, even for the same regions. Discrepancies also arise from varying assumptions, such as considering only grass eaten by livestock in WF assessments versus attributing all grazing land to the LF of livestock. EE-MRIO models, while widely used, face challenges regarding the accurate representation of physical product flows due to price variations, limited detail in monetary input-output tables, and discrepancies between agricultural statistics in physical and monetary units. This research addresses these challenges by employing a multi-model approach to analyze EU food consumption's LF and WF.

Numerous studies have used EE-MRIO models to analyze the physical flows in the global economy. However, the robustness of these models in accurately representing global physical biomass flows has been questioned. Concerns include the potential for over- or underestimation due to price variations affecting the proportionality between monetary and physical flows. The limited detail of these tables often necessitates grouping diverse products into homogeneous sectors, which can lead to inaccuracies. Furthermore, inconsistencies exist between agricultural and forestry statistics reported in physical and monetary units. To mitigate these uncertainties, several studies have advocated for shifting from sector-level economic data towards more detailed physical data. This research contributes to this movement by implementing a multi-model approach, incorporating both physical trade data and several MRIO models to enhance the accuracy and reliability of LF and WF estimations for EU food consumption.

This study employed a multi-model approach using one physical trade model (PHYS) and three global MRIO models: EXIOBASE, FABIO, and a hybrid model combining both (HYBRID). PHYS utilizes physical bilateral trade-flow data with an origin-tracing algorithm, covering 191 primary agricultural products for 223 countries. EXIOBASE, a well-established MRIO model, covers 200 products and services, including 19 agricultural ones, for 44 countries and five aggregate regions. FABIO, the Food and Agriculture Biomass Input-Output model, focuses on physical units and covers 130 commodities, predominantly agricultural and food products, for 191 countries. Each MRIO model utilized two set-ups, resulting in seven model variations (including PHYS). EXIOBASE's set-ups (EXIOBASE-min and EXIOBASE-max) represented extreme scenarios regarding the classification of products as food or non-food. FABIO's variations (FABIO-mass and FABIO-value) differed in their allocation methods for by-products. The study also employed a harmonized approach for both the LF and WF of grazing, accounting only for grass eaten by livestock, converting grass consumption into area-based estimates using remotely sensed grassland productivity data. The LF and WF were calculated for the EU27 in 2012, with a harmonized input data and a population of 441 million. Footprints for non-food uses of agricultural products were also calculated. LF quantifies cropland and grazing land use, while WF assesses the consumptive use of blue (surface and groundwater) and green (soil moisture) water resources. Irrigated agriculture uses both blue and green water, while rainfed agriculture uses only green water. The livestock WF includes blue and green water in feed and green water in grazing.

The EU food consumption LF ranged from 140.3 to 222.4 Mha yr⁻¹ (0.318 to 0.504 ha person⁻¹ yr⁻¹), with EXIOBASE-max showing significantly higher values than the other models. The LF of non-food agricultural products was substantial, particularly for EXIOBASE. The EU food consumption total WF ranged from 569.3 km³ yr⁻¹ (PHYS) to 917.8 km³ yr⁻¹ (EXIOBASE-max) (3,538 to 5,703 l person⁻¹ d⁻¹). Similar to LF, the total WF of non-food products was high for EXIOBASE. The green WF was the largest component of the total WF. The blue WF showed more variation across models. Animal products comprised a large portion (61-67%) of the total LF across models. The proportion of animal products in the total WF varied considerably among models (15-41% for blue WF and 39-64% for green WF). Analysis by product origin revealed that PHYS, FABIO, and HYBRID showed a relatively high proportion of EU-produced food in total footprints (64-75% for LF, 71-74% for blue WF, and 58-64% for green WF). EXIOBASE exhibited much lower proportions of EU-produced food. A global LF calculation, excluding PHYS, showed a total of 3,014 Mha yr⁻¹, with 1,357 Mha yr⁻¹ from crop production and 1,657 Mha yr⁻¹ from grazing. PHYS showed a global LF of 2,809 Mha yr⁻¹. This estimate is 59–63% of the FAOSTAT total agricultural land use in 2012. The difference is mainly attributable to the grazing LF, as cropland estimates are within the range of previous studies.

The study's multi-model analysis highlights the importance of selecting appropriate accounting methods. While six of the seven models showed agreement in total LF and green WF, EXIOBASE-max exhibited substantially higher values. Differences in system boundaries and key assumptions among models caused discrepancies in the values for product groups and product regions of origin. For instance, models differ in how livestock feed is linked to livestock products; some overestimate feed use for beef within the EU, and consequently underestimate feed use for pork, for example. Consistent model selection is crucial for scenario analyses. The high disaggregation of products and countries in certain models reduces uncertainty in total food footprints. Increasing country resolution improves accuracy. Scenario analyses for reducing EU food consumption LF and WF include shifting to sustainable diets and reducing food loss and waste. Detailed product and country-of-origin information is essential for such analyses and resulting policy.

This study demonstrates the importance of consistent methodology in assessing the land and water footprints of food systems. The multi-model approach reveals significant variations depending on the chosen model and accounting method. High food product and country resolution enhances accuracy, enabling effective scenario analysis for promoting sustainable food systems in line with the EU's Farm to Fork Strategy. Future research should focus on harmonizing methodologies and improving data resolution for more precise and reliable assessments.

The study's limitations include the use of data from 2012, which may not fully reflect current trends. Different models use varying system boundaries and assumptions, leading to potential biases in results. The harmonized approach for grazing, while improving consistency, might not capture the full complexity of grazing practices and their impact on the land. Finally, the analysis focuses on land use and does not consider land-use change, which is another significant environmental impact.