Multi-driver biodiversity impact assessment reveals hidden hotspots in institutional consumption

Human activity has now substantially altered most of the world’s ecosystems, driving a widespread decline in species abundance and richness across biomes^{1,2,3,4,5,6,7}. Habitat loss and degradation are the leading threats to global biodiversity and cause of elevated extinction rates of wild species¹. Agriculture occupies roughly 43–46% of habitable land⁸, and over the past half-century, agricultural expansion and intensification have driven about 80% of global land‑use change, largely replacing forests, grasslands and wetlands with cropland and pasture^1,9,10,11,12. At the global scale, food systems mainly affect biodiversity through the conversion of natural habitats and by intensifying farming on existing fields, which increases pollution (including eutrophication and pesticide use) and resource demands (e.g., irrigation water)^3,4,13. In many countries, food consumption is a predominant driver of biodiversity impact¹² and rising demand, driven by population growth and dietary shifts^14,15 towards more resource-intensive foods, is projected to increase global consumption by 35–60% by 2050, further amplifying pressures on ecosystems^{3,14,15,16,17}.

As human demand for food is increasingly decoupled from the location of production, the geographic origin of commodities becomes important for biodiversity outcomes¹⁸. A substantial proportion of species threats in Latin America and Africa have been attributed to foreign demand, particularly from high‑income countries^{5,19,20,21,22,23}. Biodiversity responses to pressures are highly site‑specific: severity and spatial patterns vary according to management, biome, landscape history, and local environmental conditions^{6,9,24,25,26,27,28,29,30}. Crop type and origin are among the strongest determinants of biodiversity pressure³¹. Imported fruits, for example, often impose greater biodiversity pressure than domestically grown equivalents, and tropical commodities such as cocoa, coffee, and palm oil can have outsized impacts when sourced from high‑biodiversity regions^32,33,34. Equally, identical land‑use classes (e.g. pasture, irrigated cropland) can produce very different outcomes depending on local management, stocking rates, fertilizer inputs, landscape fragmentation, and proximity to protected or critical habitats^35,36,37,38.

To assign responsibility for displaced pressures, regionalized consumption‑based assessments that link production locations to final demand have been recommended^{36,39,40,41,42,43,44}. Multi-Regional Input Output (MRIO) analysis can trace resource flows back to countries of primary production^29,45,46 and several approaches have revealed substantial telecoupling effects (e.g. land-use–mediated species losses occurring far from consumption centers)^{18,29,33,47,48}. However, many applications rely on global-average datasets in the inventory and/or impact characterization^49,50 (see e.g. Bidoglio et al.³¹ for the risks of using average values) or obscure spatial heterogeneity by relying on monetary transactions^{12,19,20,42,51,52,53,54,55,56,57} (see e.g. introduction of Vanham et al.⁵⁸ for limitations), and therefore fail to track geographic origin or within-sector variation^{58,59,60,61,62,63} potentially over‑ or under‑estimating impacts^52,64. Prior work centred around the production and trade windows of 2000–2013^20,32,65 and predominant focused on land-use characterisation methods^{50,51,65,66,67,68,69} (many of which follow the previous guidance of UNEP–SETAC Life Cycle Initiative⁷⁰ using Chaudhary and Brooks^38,70,71) with limitations in species habitat preferences and spatial granularity^,. From this three persistent limitations emerge: (i) monetary MRIOs aggregate agriculture into broad categories and mask material-flow heterogeneity and provenance effects (ii) most analyses emphasiz land use while under‑representing other drivers such as water use, eutrophication, and climate change; and (iii) widely used CFs and spatial datasets are often dated or coarse, limiting spatial and taxonomic specificity. These constraints can distort both the magnitude and geographic distribution of biodiversity impacts and reduce the usefulness of results for targeted supply- and demand-side mitigation. Additionally, previous consumption-based biodiversity assessments have operated predominantly at national or global scale, characterising the footprints of entire economies or dietary patterns^20,32,51,65. Whether regionalized, multi-driver assessment meaningfully changes impact profiles and intervention priorities for individual organizations remains largely unexplored, despite such organizations representing a tractable and concrete entry point for supply chain intervention that national or global analyses cannot directly inform.

We assess the impact of one full year of consumption at an institutional canteen in Germany, covering 42 food products. We couple this data with the FABIO (Food and Agriculture Biomass Input‑Output) model, a physical multi‑regional input–output model that provides origin‑resolved flows for 123 agricultural and forestry commodities in physical units⁶⁴. Our trade-linked physical inventory traces agricultural production along global supply chains and quantifies country‑of‑origin pressure profiles that incorporate agronomic heterogeneity (e.g. country- and crop- specific intensities, yield differences, irrigation, nutrient applications) for each food commodity in 2021. We then quantify biodiversity impacts from agricultural production by applying spatially explicit (country-level) characterization factors from UNEP GLAM, which differentiate impact by location, taxa, and driver for (i) land occupation (land‑use type and country-specific intensity resolved), (ii) blue water use (ground and surface water), (iii) nitrogen and phosphorus driven freshwater eutrophication, and (iv) climate change impacts. Biodiversity impacts are expressed as the Potentially Disappeared Fraction of species per year (PDF.yr), a widely used indicator in life cycle impact assessment that quantifies the proportion of species in a given area that are lost locally lost relative to a natural or undisturbed reference state as a result of a specified pressure^36,72,73,74. The PDF indicator allows impacts from different drivers to be aggregated and compared within a single metric.

By combining physical, origin-resolved flows with regionalized, country-specific (intensity-weighted) CFs we (i) preserve spatial variation in species richness and vulnerability, (ii) integrate multiple pressures beyond land use, (iii) attribute biodiversity risk to specific sourcing regions, thereby enabling identification of commodity×region×driver hotspots and more actionable mitigation options, and (iv) quantify how regionalization changes impact rankings relative to a product-level global average. Applied to a German institutional canteen, regionalization substantially alters both the magnitude and ranking of impacts relative to global-average baselines. The assessment reveals that land-use impacts are concentrated in tropical sourcing regions, consistent with the broader literature, while the inclusion of water use impacts meaningfully shifts the hotspot structure toward water-stressed origins, a pattern that single-pressure approaches cannot capture. At the commodity level, animal products dominate in absolute terms, while tropical commodities such as cocoa, coffee, and palm oil rise sharply in per-unit rankings once spatial sensitivity to extinction risk is accounted for.

Estimating regionalized biodiversity impact of food consumption

We quantify the biodiversity footprint embodied in food production supplied to a German canteen over a single year. The hotspot countries, principal drivers and products we identify therefore correspond to the export-linked share of agricultural production associated with this specific food basket and do not represent the producing countries’ total agricultural or overall environmental burden. Results are presented at the product, country, driver, and product level, followed by a decomposition of regionalization effects.

Product comparison

For the canteen’s annual consumption, biodiversity impacts are dominated by land use (60%), followed by (blue) water use (irrigation) (28%), freshwater eutrophication (N, P) (9%) and climate change (3%). The top ten commodities together contribute roughly 85% of the canteen’s biodiversity footprint and include both high‑intensity items (e.g. Cocoa Beans and products; Coffee and Products) and high‑volume items (e.g. Vegetables, Other; Milk). Animal products account for the largest absolute biodiversity burden, driven principally by land use with a substantial secondary contribution from water use. ‘Beverages & Sweets’ also show a strong land‑use signal. By contrast, the category ‘Vegetables’ stand out due to their high water‑use impacts compared with other categories.

A land-dominated group with especially large absolute impacts comprises perennial and tropical commodities such as cocoa (≈90.5% land), palm oil (≈86.7% land) and coffee (≈86.1% land) but also bovine meat (≈72.0% land). A separate water-dominated cluster includes ‘Vegetables, Other’ (≈71.2% blue-water), ‘Fruits, Other’ (≈64.1% blue-water) and ‘Rice and products’ (≈55% blue-water). ‘Milk – Excluding Butter’ attains a large absolute impact mainly through the scale of consumption despite moderate per-unit intensity, while ‘Pigmeat’ occupies an intermediate, mixed-driver position (land ≈48.6%, water ≈32.6%, eutrophication ≈14.9%).

To convert these patterns into actionable priorities for reducing the canteens biodiversity footprint, we classify product by median thresholds of absolute impact (x) and per-unit intensity (y) and map each quadrant to appropriate intervention types for the canteen. This yields the categories Q1–Q4: high‑intensity opportunities (Q1), strategic hotspots (Q2), volume‑driven products (Q3), and lower‑priority items (Q4) (Fig. 1). Q1 contains products with high per‑kg footprints, many of which are water‑dominated, as well as some land‑ and eutrophication‑driven items (17% of products) which are suitable for running small, time-bounded pilots to test substitutes, alternative suppliers or to reformulate recipes. Q2 highlights the strategic biodiversity hotspots with a high absolute footprint and high per-unit intensity (33% of products), most of which are dominated by land‑use impacts (e.g. cocoa, coffee, palm oil, bovine meat, butter/ghee). These items are the highest priority for action through e.g. reducing frequency and replacing them with attractive lower-impact alternatives (plant-forward mains, legumes, pulses) as well as changing procurement where available to verified lower-impact or certified options. Q3 comprises high‑volume commodities (e.g., vegetables, milk, wheat, potatoes), where total impacts are driven primarily by quantity rather than per‑unit intensity (33%). These products are priority for demand management and menu strategy like portion control (e.g. smaller default servings) and where appropriate seasonal and local sourcing to reduce embodied pressure. Q4 contains products of relatively low priority for biodiversity interventions because both absolute and per‑unit pressures are small (33%). Q4 is best maintained under monitoring unless contextual risks increase.

Country and driver profile

Although food is consumed at a single site in Germany, the associated biodiversity pressures extend across all world regions and are highly unequally distributed. Middle/South America and the Asia–Pacific bear a disproportionate share, with the top 15 sourcing countries together responsible for roughly 76% of the total footprint (Fig. 2).

Land occupation is the dominant driver for most of these hotspot countries in our case study. In Indonesia, Ecuador, Germany, Myanmar and Côte d’Ivoire land-use shares are ~94–100%, and land use also dominates in Brazil (≈77%) and Australia (≈61%). This dominance is mirrored in the product mix. Impacts in these countries are primarily driven by plantation and commodity crops (notably palm oil in Indonesia and Malaysia, and cocoa in Ecuador and Côte d’Ivoire) and by animal products in several countries (e.g., Germany, Australia, Brazil, and Argentina) (Table 1).

A smaller group of hotspot countries is characterized by dominance of blue‑water impacts, notably Mexico (≈80% blue-water), United States of America (≈71%), Spain (≈57%) These driver signatures map onto product profiles: Mexico’s impacts are largely from irrigation-intensive fruits and vegetables, Spain’s from olive oil and vegetables, and United States of America’s water signal is driven mainly by animal products (especially pigmeat and milk), indicating a substantial water footprint embedded in livestock supply chains rather than solely in irrigated crops. Freshwater eutrophication (N, P) only rarely emerges as the primary driver, but it frequently attains a secondary rank and can be locally critical where nutrient loads are elevated.

Several heavily influential countries show mixed driver profiles. For example, Peru and India where blue‑water use (≈42; ≈51%) and land use (≈55; ≈46%) contributing almost equally, rather than a single near-exclusive driver.

Cross-country comparison

A comparison of key impacted countries (Mexico, United States of America, Peru, and Indonesia) in our case study, reveals contrasting profiles (Fig. 3). Mexico has the highest overall biodiversity impact and is dominated by water use, with most of this impact concentrated in a single product group namely ‘Vegetables, Other’, indicating a highly irrigation-driven footprint This pattern is consistent with documented freshwater pressures in Mexico, where agriculture accounts for the majority of withdrawals and contributes to overexploitation, salinization and contamination^75,76,77. United States of America also exhibits a water-use–led profile, but with a more mixed driver structure. Land use and freshwater eutrophication contribute more visibly than in Mexico, and the dominant-driver breakdown points to a strong role of animal products alongside other categories. In contrast, impacts in Indonesia is clearly land-use dominated, with the largest contributions stemming from oil crops (notably ‘Palm Oil’), while water use and freshwater eutrophication remain comparatively minor. Peru displays a more heterogeneous pattern than Mexico, with substantial contributions across multiple products (‘Vegetables, Other’, Cocoa Beans and products’, ‘Coffee and products’) consumed in the canteen, highlighting that impacts can arise from distinct combinations of products and environmental drivers.

Regionalized vs non-regionalized analysis

Overall, the regionalized analysis of biodiversity impacts of consumption in the canteen yields a different result in terms of magnitude, driver and product dominance than total the global-average baseline.

In the regionalized analysis of our case study canteen, biodiversity impacts are predominantly driven by land use (63%) followed by blue water use (31%). Freshwater eutrophication (3%) and climate change (3%) contribute only minor amounts. By contrast, the global-average profile is largely dominated by water use (92%) followed by land use (7%). Freshwater eutrophication and climate change each contribute less than 1% (Fig. 4). The regionalized results reduce water‑related biodiversity loss relative to global averages. Land use impacts are heterogenous, with several outliers, while freshwater eutrophication and climate change impacts remain similar.

At the commodity level, animal products clearly dominate the biodiversity footprint in the global-average case. In the regionalized view they still remain the dominant driver but their share shrinks, and the impact is more evenly spread across categories. In our case study, regionalization tends to lower the impact for several staples and animal products (largest reductions for ‘Milk – excluding Butter’ and ‘Pigmeat’, driven primarily by reduced water-use intensity and, to a lesser extent, land use) but increases it for many tropical plant supply chains (notably Coffee, Palm oil, Cocoa and some Fruits) mostly via land occupation, and from vegetables, with a mixed driver profile. The same is true when using global mean average values (see Supplementary Fig. 25). These shifts indicate that regionalization materially refines prioritization and can reassign management focus across commodities and sourcing regions.

Decomposing regionalization effects

We used a three-case decomposition to separate the effects of inventory and impact-assessment regionalization. Case 1 (g × g) serves as the baseline, using global-average inventory data (pressures) combined with global-average CFs. Case 2 (r × g) replaces global inventories with regionalized, product-specific data while retaining global CFs, thereby isolating the inventory effect, the change attributable solely to more spatially differentiated life cycle inventory (LCI) data. Case 3 (r × r) represents our regionalized inventory with spatially explicit CFs, representing the full regionalization scenario. The CF effect is then quantified as the difference between Case 3 and Case 2, capturing the influence of spatially differentiated impact assessment independently of inventory choices.

Land-use decomposition into inventory (g × g → r × g) and characterization‑factor (CF; r × g → r × r) components shows contrasting mechanisms across products (Fig. 5). Averaged over the top 15, inventory changes explain 53.2% of absolute contributions and CF changes 46.8%, but the balance differs by product. Most animal products (Bovine meat, Mutton & Goat, Poultry) are inventory‑dominated (inventory shares 66–99%), indicating that a regionalized inventory (production intensities, yields, area use) account for most of their reduction under full regionalization. Notably, ‘Milk – Excluding Butter’ behaves differently from the other major animal products. Inventory regionalization increases its land-use impact, but CF regionalization produces a much larger reduction. This pattern suggests that milk’s apparent improvement under full regionalization stems mainly from re-weighting of vulnerability (CFs) toward less-sensitive production locations in our case study.

By contrast, several perennial, tropical commodities (Coffee, Cocoa, Palm oil) are CF‑dominated (CF shares 63–89%) and show net increases under full regionalization. CF regionalization re‑weights remaining pressures toward production locations of higher ecological vulnerability, amplifying their land‑use biodiversity signal.

Regionalization reshapes the relative importance of commodities and drivers of biodiversity impact, with implications for which products represent the greatest intervention leverage. We compared each product’s share of total impact across the three scenarios (g × g, r × g, r × r) (a; Fig. 6) and identified four patterns. Pattern one, characterized as reinforcing reductions, applies for example to the category ‘Cereals and Grain’ and products such as wheat, rice, and potatoes. Both inventory and CF regionalization reduce or leave impacts unchanged relative to the global baseline. ‘Cereals & Grain’ accounts for 19.9% of total impact under g×g, declining to 13.1% under r×g and further to 4.2% under r×r. Wheat corroborates this trajectory, ranked #3 under g×g before dropping out of the Top 10 entirely under both r×g and r×r. These products remain lower priority under full regionalization.

Pattern two, characterized as diminishing reductions, applies to animal products like milk, pigmeat, poultry, and bovine meat. Inventory regionalization lowers impacts, whereas CF regionalization partially counteracts these gains by re‑weighting the remaining pressures according to the ecological sensitivity of production locations. Animal products remain major contributors across scenarios in our case study but decline in relative importance as regionalization increases. They account for 66.2% of total impact in the baseline (g × g), fall to 37.3% under regionalized inventories (r × g), and decline further to 31.5% under full regionalization (r × r). The continued but decelerating reduction suggests that while inventory regionalization captures the dominant share of the effect, CF regionalization continues to modestly reduce the relative contribution of animal products. Milk and pigmeat exemplify this pattern most clearly: milk is ranked #1 under g × g and #3 under both r × g and r × r, while pigmeat declines from #2 to #4 to #6. Bovine meat provides a partial exception, showing a degree of rank recovery (ranked #8 under g×g and #10 under r × g, it rises back to #7 under r × r), suggesting that CF regionalization partially re-elevates the relative importance of bovine meat by up-weighting impacts in ecologically sensitive production regions.

Pattern three, characterized as reinforcing increases, applies to the category ‘Beverages & Sweets’ including tropical plantation and perennial commodities like cocoa, palm oil, and coffee as well as ‘Oils & Fats’. For ‘Beverages & Sweets’, inventory regionalization produces a modest initial increase from 1.1% (g × g) to 6.0% (r × g), but CF regionalization drives a much larger surge to 20.4% under full regionalization (r × r), indicating that their elevated impacts stem predominantly from the ecological sensitivity of their production locations. ‘Oils & Fats’ shows a similar but less extreme trajectory, rising from 1.1% (g × g) to 8.5% (r × g) and 11.9% (r × r).

‘Vegetables’ and ‘Spices & Condiments’ share a distinct fourth dynamic, best characterized as an inventory-driven increase with partial CF reversal. Both categories rise strongly under inventory regionalization but are subsequently dampened by CF regionalization. ‘Vegetables’ rises from 7.1% (g × g) to 18.1% (r × g), before reducing slightly to 16.8% under r × r. ‘Spices & Condiments’ shows an even more pronounced reversal in relative terms, rising from 0.1% (g × g) to 5.2% (r × g) before falling back to 2.5% under r × r. At the product level, ‘Vegetables, Other’ rises from #5 to #1 under r × g and retains the top rank under r × r, while ‘Spices, Other’ appears at #6 exclusively under r × g before dropping out of the Top 10 entirely under r × r. This pattern indicates that for these categories the dominant driver of impact re-weighting is inventory regionalization, reflecting shifts in the geographic origin and volume of environmental pressures. Whereas CF regionalization, by re-weighting according to ecological sensitivity, partially offsets rather than reinforces these gains. This contrasts directly with ‘Beverages & Sweets’ and ‘Oils & Fats’ (Pattern three), where CF regionalization is the dominant amplifying step, illustrating that inventory and CF regionalization do not act symmetrically across commodity types.

Rank-shift analysis (b; Fig. 6) corroborates our findings. While ‘Milk – excluding Butter’ and ‘Pigmeat’ remain highly ranked under g × g and r × g, their positions decline under r × r. At the same time, ‘Vegetables, Other’ increases in relative importance and becomes the top ranked product under r × r. Several staple crops such as wheat and rice drop out of the top ten under full regionalization, whereas products cocoa (#2 under r × r), palm oil (#4), coffee (#5), and beans (#8) enter the ranking, indicating that full regionalization changes both the level and the composition of biodiversity impact hotspots. ‘Cocoa beans and products’ represent only 0.63% of total impact in the baseline (global average) and 1.46% under regionalized inventory. However, under full regionalization, cocoa surges to 11.3% of total impact, more than 17× increase from r × g driven almost entirely by the CF effect (720% increase). Palm oil demonstrates an even more extreme pattern, increasing from 0.42% of total impact in the baseline to 0.81% under regionalized inventory, but then surging to 8.9% under full regionalization.

International trade can enhance food security and resource efficiency^23,78,79,80, yet it also displaces environmental burdens to exporting regions^{46,47,81,82,83}. This study demonstrates that regionalized, multi-driver biodiversity footprint assessment substantially alters both the magnitude and structure of impacts compared to global-average baselines, and that this difference is consequential for identifying where and how to intervene. By coupling a physical, origin-resolved agriculture- specific MRIO, with spatially explicit (country level) characterization factors (CFs) from Life Cycle Impact Assessment (LCIA), our approach for biodiversity impact assessment reflects differing contexts from local production practices (crop- and country-specific land use types and intensities) and ecological conditions (biomes and species vulnerabilities) across multiple drivers (land use, blue water use, and nitrogen- and phosphorus- driven freshwater eutrophication, climate change). This produces commodity × country × driver footprints (i), reveals product–origin hotspots that differ from global‑average baselines (ii), and enables inference about the land requirements per unit output that underpin observed impact differentials, reflecting, for instance, the combined effect of below-average yields, high baseline biodiversity, and pressure intensity in specific sourcing regions (iii), even where mechanistic causes cannot be isolated at this level of aggregation^32,33. Together, these outputs provide a more complete picture of the pressure profile behind a given consumption pattern and support the design of origin-sensitive mitigation measures, ranging from demand-side and procurement interventions to cooperative yield improvement or conservation investments. The analytical value of this approach operates across three distinct dimensions, each with direct implications for mitigation targeting.

First, regionalization reveals that sourcing origin modifies impact magnitude. Lower yields in high-biodiversity regions imply more land per kilogram produced and, combined with higher baseline species richness and weaker regulatory enforcement, a greater biodiversity impact per hectare^20,84. Conversely, water-related impacts in some regions fall below global-average estimates where water scarcity is lower, illustrating that regionalization can reduce as well as amplify assessed impacts depending on driver and origin. Second, the multi-driver structure exposes geographic heterogeneity in pressure profiles and reveals that the relative ranking of commodities and sourcing regions can shift depending on which driver is considered. Water-related impacts, for instance, follow a different geographic pattern than land-use impacts, and their inclusion modifies the overall hotspot structure in ways that single-pressure approaches cannot capture. The results can indicate the mechanisms (e.g. irrigation efficiency in Mexico’s vegetable production) relevant for each hotspot and distinguishes single‑lever from multi‑lever strategies, enabling more precise, place‑based targets^27,40,43,85. Third, by operating at the organizational scale, a level at which concrete, recurring procurement decisions are made, this approach bridges the gap between national-scale footprint analyses and actionable supply chain intervention. To our knowledge, this represents the first application of a regionalized, multi-driver biodiversity footprint assessment at the level of an individual institutional food provider, demonstrating that such analyses are both technically feasible and practically informative at this scale.

Country level results in our case study of food consumption in a canteen in Germany are dominated by land‑use and water‑use impacts in Middle and Latin America as well as land use-induced impacts in Asia. This aligns with evidence that a substantial proportion of species threats in Latin America and Asia is linked to foreign demand^{20,22,41,86,87}. Previous biodiversity footprint studies have largely been confined to land use as the sole pressure driver. Land use impact hotspots such as Brazil, Indonesia, Côte d’Ivoire and Malaysia (key exporters of soybean, cocoa, palm oil and coffee) have previously been identified as particularly critical nodes in global biodiversity impact networks⁶⁵, a pattern that holds across different consuming entities, from nations³³ to cities²⁹ to individual organizations (this study). The inclusion of water use impacts in our assessment shifts the geographic distribution of hotspots meaningfully, elevating the relative contributions of water-stressed regions including Mexico, the United States, Spain, and India (compared to e.g. ref. ³³). At the commodity level, our results converge with the broader literature on product hotspots. Animal products are the largest contributors in absolute terms, consistent with evidence across consumption-based studies^{31,50,51,69,88}. However, several plant‑based commodities, mainly tropical commodities such as cocoa, coffee, and palm oil sourced from high‑biodiversity regions gain prominence when regionalizing assessments. For example, coffee, while modest in area terms, rises to the top of biodiversity impact rankings when spatial sensitivity to extinction risk is accounted for in line with^29,33,69. The agreement across methodologically diverse studies using different pressure scopes, characterization factors, trade models, and consuming entities, strengthens confidence in the qualitative robustness of our conclusions while the divergence in geographic hotspots introduced by multi-pressure accounting, particularly water use, highlights the added value of a more comprehensive driver framework for identifying intervention priorities at the institutional level.

For the canteen specifically, the results reveal a clear distinction between two intervention archetypes. First, (1) demand‑ or efficiency‑oriented strategies could reduce the footprint of volume‑driven staples with low-to-moderate intensity (e.g. vegetables, milk). These measures may include scale management via demand reduction (e.g. portion control and menu redesign) that lower aggregate demand as well as seasonal and local sourcing (Q4, Fig. 1). Second, (2) strategic hotspots with both high absolute and high per-unit impacts (e.g., cocoa, coffee, palm oil, bovine meat) require a different logic: demand reduction alone is insufficient, and durable mitigation requires simultaneous supply side intervention. Demand‑side actions could include pricing signals (for example modest surcharges on high‑impact items), portion‑size reductions and default plant‑based mains to discourage high‑impact choices, while supply‑side responses could include switching suppliers for products with both high per‑unit intensity and large absolute impacts. For land‑use‑dominated commodities, durable reductions require improved land management (for example agroforestry and biodiversity‑friendly intensification), deforestation‑free procurement, and sustained supplier engagement and certification schemes that secure verified, long‑term conservation outcomes. At the same time, management choices must engage the long-standing sparing-versus-sharing debate. Sparing (producing more on less land) may be appropriate where high yields and strict protection can prevent conversion of native habitats, whereas sharing (integrating biodiversity into production systems) can sustain within-landscape diversity but may reduce yield per area and require locally tailored practices^89,90,91. The appropriate mix of demand-side and supply-side measures therefore depends on local ecological context, food-security objectives and livelihood implications. Crucially, such interventions should be designed in alignment with dietary recommendations, for instance the DGE guidelines for Germany or the Planetary Health Diet, to ensure that footprint reductions do not compromise nutritional adequacy. Reporting impacts per unit of nutritional value, including calories, protein, or a nutrient-density index, would enable direct comparison of environmental cost against dietary contribution, aligning biodiversity footprints with planetary health frameworks and enabling more nuanced trade-off analysis between environmental harm and nutritional function^{81,82,92,93,94}.

The geographic specificity of regionalized results enables the canteen to move beyond generic sustainability commitments toward origin-sensitive procurement targets, identifying not just which commodities to prioritise but which sourcing regions and supply chain nodes represent the highest-leverage points for intervention. Regionalization does more than refine impact estimates, it provides a hierarchy of actionable strategies for the canteen and supplies information needed to follow the mitigation hierarchy from avoidance and minimization of impacts to restoration and compensation where impacts occur. The same spatial data could in theory be used to direct ecologically equivalent, like-for-like restoration within affected ecoregions, strengthening the credibility of nature-positive claims for organizations⁵⁰. At the same time, regionalization is a screening and prioritization tool rather than a farm-level prescription. National-average characterization factors cannot resolve management heterogeneity within countries, and the governance and land management conditions that determine whether sourcing shifts reduce or merely displace impacts require ground-truthing beyond what biodiversity footprint analysis alone can provide.

Several directions would substantially extend the utility of this approach. Expanding the system boundary to encompass processing, packaging, transport, and food loss would capture downstream supply chain hotspots that production-focused assessment cannot reveal. Complementary advances include incorporating more drivers (e.g. ocean and terrestrial acidification^95,96), comparing metrics and CF versions, testing the spatial and taxonomic scope (country vs ecoregion; plants, mammals, birds, etc.), exploring further scenario and parametric uncertainty (e.g. land-intensity assumptions, reference states). Linking inventory outputs to prospective economic models would allow the framework to reflect evolving trade structures and agricultural technology trajectories rather than a fixed baseline year. These extensions, combined with systematic comparison across biodiversity metric families and characterization factor frameworks, would progressively close the gap between the spatial precision now achievable in consumption-based biodiversity assessment and the ambition of the policy and procurement responses it is intended to inform.

Limitations

The interpretation of our findings is subject to several data and modelling constraints.

First, the analysis only covers food and beverages consumed within the canteen, which represents a single meal occasion (lunch) and selected snacks and drinks. Breakfast, dinner, and food consumed outside the canteen may differ substantially in composition and environmental impact but are not captured in the analysis. This means that the results reflect only a subset of the total dietary consumption of staff and students. Notably, items typical of institutional catering, such as coffee and hot drinks, are likely to be reasonably well represented, whereas staple foods and animal products predominantly consumed at home, particularly at dinner, are likely to be underrepresented. The representativeness of canteen meals for overall dietary patterns therefore remains uncertain, and the findings should be interpreted as being indicative of institutional lunch provision rather than total dietary impact. Additionally, grouping heterogeneous food items into broad commodity categories introduces classification uncertainty. While transparent concordance tables and, where feasible, recipe-level disaggregation mitigate this risk, they cannot eliminate it.

Second, FABIO resolves trade at the national level. Within‑country heterogeneity in yields, irrigation intensity and nutrient management is not captured, which has consequences for land- and water-related impacts in spatially diverse producing countries.

Third, our land-use accounting captures the ongoing occupation of land that has already been converted, but it excludes biodiversity losses associated with indirect land use change and habitat transformation. Consequently, products linked to frontier agriculture (e.g. cattle ranching in the Amazon, oil palm plantations on former rainforest) carry higher land-related impacts when considering land use change than is reflected here. While the updated country-level land-use intensity levels improve realism relative to defaults, they represent national averages (proxies) that mask within-country variation and system- and management-specific practices such as fertilization regimes (e.g. fertilization practices, see Coelho et al.⁶⁶ for a scenario analysis of different land use intensities). Reference-state uncertainty further complicates land-use assessments, as results are sensitive to whether occupation or conversion is used as the impact mechanism and to assumptions about the time since conversion.

Forth, eutrophication impacts were estimated using uniform field-scale leaching fractions applied globally. This approach does not account for in-stream nitrogen removal processes^97,98, ignores regional controls on leaching (soil properties, climate, hydrology)⁹⁹ and omits organic nitrogen sources including manure, biological fixation, and atmospheric deposition. Spatially differentiated nitrogen flow data have recently been incorporated into a subsequent FABIO version (Hinz et al., in review) but were unavailable for this study. These limitations are, however, of limited consequence for our primary conclusions: eutrophication contributes 9.2% of total assessed biodiversity impact, and sensitivity analyses ranging from complete removal to doubling of eutrophication impacts produced no rank changes among the top 10 sourcing countries (see Supplementary Tables 12, 13). Using climate- and soil-adjusted leaching factors would therefore enhance absolute quantification^48,100 but do not change commodity×region hotspot identification.

Fifth, the system boundary for climate change differs from that for the other drivers. FABIO does not yet provide emissions data. Therefore, greenhouse gas emissions are derived from product-level carbon footprint data rather than from the shared process-level inventory underlying the remaining indicators. The data cover all life cycle stages up to the supermarket checkout including land-use change via an attributional approach, and represent an average food product sold in Germany, with weighting applied according to domestic and imported proportions, countries of origin, cultivation methods, seasonality, and transport modes.

Due to the lack of country-resolved inventory data, climate change impacts could not be disaggregated at the country level. This contrasts with all the other indicators and highlights the need for future work to enable their inclusion in country-specific analyses. Sensitivity analyses across emissions assumptions (±50%), characterization factor variants (±100%), climate scenarios (RCP 2.6–8.5), and time horizons (100–1000 years) produced no changes to the top 10 rankings changes under standard scenarios. Even under the most extreme combined scenario (RCP 8.5, 1000-year horizon), in which the climate’s contribution to the total impact rises to the 22.6%, only 9 of 42 products shifted by ±1–2 positions (see Supplementary Tables 10, 11).

Sixth, our analysis and comparison are subject to temporal misalignment across three data layers: procurement data (2023), trade modeling (2021), and the global average benchmark (2000–2016). Trade structures, yields, and governance evolve over time. The procurement–trade gap may misattribute sourcing regions for commodities with volatile supply chains in the post-2021 period, including those affected by geopolitical disruptions (e.g. Wheat from the Ukraine) and COVID-19 recovery effects^101,102. There is a further methodological distinction between our production-focused regionalized assessment and the global-average benchmark used for comparison. The latter encompasses processing, packaging, food losses, and land-use change across the full supply chain, whereas our approach isolates agricultural production (apart from climate change analysis which includes the same stages). This system boundary mismatch has directional implications that differ by commodity type. Highly processed products and commodities with high loss rates such as refined oils, packaged foods, and dairy products are more strongly affected by boundary differences than minimally processed commodities, since downstream stages contribute disproportionately to their full life-cycle footprint. The exclusion of land-use change (LUC) in our land use analysis (regionalized assessment) introduces an additional bias. LUC impacts, which vary substantially by commodity, region, and time, are captured in the global baseline but only partially attributed in physical trade data. This means that our regionalized estimates may appear systematically lower and potentially underestimate the full effect of regionalization when applied to complete life cycles¹⁰³. However, the directionality is not unambiguous: where the global baseline averages high-impact production with lower-impact processing and distribution stages occurring in less biodiverse regions, its life-cycle-averaged values may in turn underestimate the severity of production-stage hotspots, which our approach reveals more directly by isolating agricultural stages³¹.

A related comparability limitation applies to water use. Our assessment quantifies blue water consumption (consumptive use of surface and groundwater for irrigation), whereas the global benchmark includes both blue and green water (rainfall and soil moisture). However, for biodiversity impact assessment, blue water is the ecologically relevant metric as it is consumptive blue water that drives river flow reduction, aquifer depletion, and wetland shrinkage, and characterization factors for water-related biodiversity impacts are accordingly constructed around blue water consumption and local scarcity indicators. Green water, consumed in situ through evapotranspiration, does not generate the flow deficits or resource competition that underpin biodiversity impact pathways. Our blue-water-only approach is therefore not simply narrower than the benchmark, it is better aligned with the underlying causal mechanism. Nevertheless, the metric mismatch means that direct numerical comparisons against the global benchmark for the water use pathway should be interpreted with caution, as the benchmark reflects a broader water budget than the biodiversity-relevant fraction captured here.

These system boundary and temporal differences are partially, but not fully, disentangled by our decomposition approach (holding characterization factors constant between cases 1 and 2; holding inventory constant between cases 2 and 3). Residual confounding between spatial disaggregation and temporal change cannot be eliminated within this analytical structure.

Seventh, the biodiversity characterization factors applied are spatially explicit at the national level and represent average conditions, capturing broad geographic variation but not site- or management-level heterogeneity. Global extinction probabilities (GEPs) are applied across all CF datasets to scale local species loss to global extinction risk. However, taxonomic coverage remains limited, with species acting as coarse proxies for broader biodiversity, and underlying data biased towards higher trophic levels and certain geographic regions¹⁰⁴. Vascular plants and vertebrates are comparatively well represented, whereas fungi, soil invertebrates, and many arthropod groups are largely absent from the underlying datasets, which likely leads to an underestimation of true biodiversity impact in taxa with high functional importance but limited occurrence data. The resulting PDF.yr value therefore represents not a count of species extinctions, but a probabilistic aggregate (=the expected proportional reduction in species richness, summed across all contributing pressures throughout the product system), and should be interpreted as a comparative index of biodiversity pressure rather than a direct measure of ecological outcome. Nevertheless, GEPs represent the most advanced method for currently available incorporating global-scale extinction risk when PDF-based indicators are applied in LCA. A further source of uncertainty lies in the choice of characterization factor family. Structural differences between frameworks, including modeling assumptions, taxonomic scope, and reference states, mean that divergences between CF families such as approaches derived from GLAM, LC-IMPACT⁷³, ImpactWorld+⁷⁴, and GLOBIO^105,106) cannot be attributed solely to differences in spatial resolution or inventory quality. They reflect, in part, genuinely different operationalizations of biodiversity loss. All characterization factors represent updated frameworks with improved spatial resolution and broader impact coverage relative to previously used methods. For example, land-use CFs follow Scherer et al.³⁷ building upon the work of Chaudhary and Brooks³⁸ with a higher spatial resolution and inclusion of fragmentation impacts. Eutrophication CFs follow Zhou et al.¹⁰⁷ extending previous approaches by covering both freshwater and marine eutrophication more comprehensively than hypoxia-only methods, incorporating both N and P simultaneously while accounting for limiting nutrient conditions, and providing substantially finer resolution. The uncertainty associated with each CF set used is documented in the respective underlying publications and is not reproduced here. However, users comparing results across studies should bear in mind that inter-CF variation may be as large as, or even larger than, the differences between commodity or regional hotspots identified within any single assessment framework.

Although regionalized CFs improve spatial resolution in hotspot identification, using them to guide sourcing decisions carries an inherent risk: shifting procurement toward nominally lower-intensity regions may displace rather than reduce impacts.

Another uncertainty relates to the choice of biodiversity metric. As this introduces framework-specific assumptions that affect absolute magnitudes and relative rankings of assessed impacts. PDF quantifies the relative loss of species richness compared to a natural reference state, while mean species abundance (MSA)⁷² expresses the mean remaining abundance of originally occurring species. As MSA is based on species abundance rather than occurrence, it is more sensitive to sub-extinction changes in ecological communities and may better capture early-stage biodiversity degradation and partial recovery than PDF⁷². Land-cover-based future extinction approaches such as land-cover-based future extinction metrics (LIFE)^69,108, estimates committed future extinctions from projected range contractions rather than current local disappearance and is likely to rank commodities linked to large-scale habitat conversion more severely than either PDF or MSA. Given that land use contributes ~60% of the total assessed impact in our analysis, metric choice within the land-use domain represents the largest source of framework-dependent uncertainty. Kuipers et al.⁷² have characterized the directional consequences for specific commodity × region rankings for PDF and MSA by and warrant systematic extension to extinction-committed metrics in future work. It is also important to note that biodiversity encompasses multiple dimensions (spanning genetic, species, interaction, and ecosystem diversity), so no single metric can fully represent it¹⁰⁹. While PDF does not account for all aspects, it provides a consistent, quantitative metric for comparing biodiversity impacts across drivers, regions, and products within a unified assessment framework.

Finally, a related and underappreciated limitation concerns the treatment of biodiversity within agricultural landscapes themselves. The PDF-based framework used here quantifies biodiversity loss relative to a natural reference state, implicitly treating agricultural land as a source of pressure rather than a potential contributor to biodiversity. However, human-modified agricultural environments can also support biodiversity: diverse farm landscapes, semi-natural field margins, and traditionally managed systems can harbor substantial species richness¹¹⁰. Biodiversity exists within and among food systems, not only beyond them. The current approach cannot differentiate between intensively managed monocultures and more biodiverse agroecological systems, nor can it capture positive biodiversity outcomes associated with habitat management, reduced-input practices, and the maintenance of landscape heterogeneity by food value chain actors¹¹¹. Characterization factors based on land-use intensity and species–area relationships are not designed to capture such actions, and their absence from the accounting framework means that biodiversity benefits of proactive stewardship are systematically unrepresented. Distinguishing production systems that actively counteract biodiversity decline from those that do not, and translating such distinctions into quantifiable impact differentials, remains an important methodological frontier.

Taken together, these limitations reflect the broader challenges inherent in life-cycle-based biodiversity assessment. LCA-derived footprints involve detailed estimation of pressures, yet they remain sensitive to system boundary definitions, characterization factor selection, and allocation choices. There is currently no universally accepted standard for any of these. This constrains direct comparability between studies and precludes simple summation of impacts across regions or products^40,43,69.

Our sensitivity analyses demonstrate, however, that while absolute magnitudes carry uncertainty, the rank order of commodity×region hotspots remain stable across the full range of tested methodological assumptions, including eutrophication weighting, climate characterization choices, and extreme combined scenarios. By contrast, absolute impact magnitudes carry compounding uncertainties arising from mass allocation, uniform leaching fractions, national-average characterization factors, and temporal misalignment across data layers. The results should thus be interpreted as order-of-magnitude indicators rather than precise quantities. Regionalization is therefore most appropriate as a tool for strategic screening and procurement prioritization, and findings should be applied at that scale rather than extrapolated to farm-level action or for direct numerical comparison with assessments employing alternative characterization frameworks.

The biodiversity impacts associated with various food products were assessed using a structured, multi-step methodology (Fig. 7):

• Step 1: Annual food consumption data for the canteen were obtained from the university’s central procurement system for one full year of operation (2023). These Procurement records were filtered and processed to identify the ingredients used. Food items were then grouped and categorized based on their primary ingredients, using the classification scheme provided by the FABIO v2 (Food and Agriculture Biomass Input-Output) database (Sect. 2.1).

• Step 2: For each product, we computed country-specific per‑kg intensities for land use, nitrogen and phosphorus application, and water use, using FABIO v2 (Sect. 2.2). An existing LCA datasets was deployed to calculate greenhouse gas emissions contributing to global warming (Sect. 2.3.2).

• Step 3: Potential biodiversity loss was assessed by linking results for the environmental pressures to ‘endpoint’ impacts on ecosystem quality. This translation applied spatially explicit (country level) characterization factors from GLAM, which account for regional differences in species vulnerability and ecosystem responses to environmental stressors (Sect. 2.3)

• Step 4: The resulting regionalized biodiversity impact assessment of food products consumed in a canteen in Germany was compared with a dataset representing global average impacts of the food items considered.

Step 1: Context and data cleaning

This study evaluated the biodiversity impacts associated with the consumption of various food products in Germany using detailed procurement records from the canteen at the University of Hohenheim as an example. The canteen regularly provides meals and beverages (hereafter referred to collectively as food) for ~8800 university students, ~2080 academic and administrative staff, as well as support and facilities personnel (institutional records). All meals are prepared in a central kitchen, with an adjacent cafeteria offering snacks and beverages.

The analysis is based on data derived from the central procurement system of the “Studierendenwerk Tübingen-Hohenheim”, the organization responsible for operating the canteen. This dataset comprised 1204 food items purchased over a one-year period with precise information on the quantities and units. A total of around 200 metric tons of food products were purchased in 2023, corresponding to the preparation of ~211,000 main meals as well as snacks and drinks in the adjoining cafeteria. For the purpose of this study, it was assumed that all procured items were used for food preparation within the same year. Importantly, the analysis focused exclusively on the biodiversity impacts arising from agricultural cultivation and feedstock production of these food products. Processes such as international and regional transportation, food processing, and preparation in the canteen kitchen are beyond the scope of this analysis and were not considered.

Product weights were calculated using the information provided on unit weight and container size. Subtotal sums were compiled to establish the overall quantities of food purchased. Given the prevalence of processed foods containing multiple ingredients in the canteen dataset, several pre-processing steps were necessary. First, the main ingredients of each purchased product were identified. For composite products, such as baked goods, each product was disaggregated into its constituent ingredients (wheat flour, eggs, butter, sugar etc.) using standardized proportions derived from common recipes. The quantity of each ingredient was then estimated by multiplying the total product weight by its relative proportion. For example, soft drinks were assumed to contain 10% sugar, and cheese production was attributed to a milk input ten times the weight of the final product. An overview of the data and recipes is provided in the Supplementary Information 1 (SN2.2).

The canteen’s procurement records were harmonized with the classification system used in the Food and Agriculture Biomass Input–Output (FABIO) database. This comprehensive database was developed to systematically map and quantify agricultural flows in physical units. FABIO enables the tracing of agricultural and food products with high spatial (186 countries) and product resolution (123 commodities)^64,112. The model integrates data on crop production, trade, and utilization in physical units, together with technical and metabolic conversion efficiencies, to capture the physical structures of agri-food conversion and distribution networks, offering higher resolution for agriculture and forestry⁶⁴.

These steps resulted in 42 food products being considered.

Step 2: Regionalization: product- and country-specific requirements

Primary crops reach their final consumers via complex global supply chains involving numerous processes and stakeholders that transform and move materials and energy, connecting regions of production with (often distant) regions of consumption. While the original procurement dataset occasionally included country-of-origin labels, these typically referred to the location of the final processing rather than the actual site of primary production. Therefore, for consistency and accuracy, the FABIO model (beta v2, provided by Martin Bruckner) was used to trace the origin of all food items and to calculate the environmental impacts (land use, blue water use, nitrogen and phosphorous application) of the 42 food products consumed. FABIO is a multi‑regional physical supply–use and input–output database for global agriculture and forestry. It integrates FAOSTAT data on crop production, trade, and utilization in physical units together with technical and metabolic conversion efficiency parameters to construct a consistent, balanced MRIO framework. FABIO therefore captures food system resource flows in the agricultural sector and the indirect crop feed impacts of livestock^64,112.

FABIO v2 provides inter‑commodity transactions in two variants: Z_mass (mass allocation) and Z_value (value allocation) as well as final demand Y disaggregated by use categories (including food use), total output X, Leontief inverses L_mass/L_value, and environmental extensions E for land use (ha), blue and green water (m³), and nitrogen and phosphorus application (kg) among others.

We quantify country-of-origin–resolved footprints from agricultural production (land use, water use, N and P application) of each product consumed by consistently applying the ‘food use’ final‑demand category for Germany in 2021 and using the mass‑allocation variant in FABIO. Different years for FABIO (2021) and consumption data (2023) are used due to data availability at the time of analysis. We allocate the origin‑specific pressure (Q_p,o,m) for each product (_p) and driver (_m) (m ∈ {land, blue water, N, P}) to German final demand. Dividing by Germany’s apparent consumption (CDE_p) (kg) yields origin‑specific intensities:

and the product-average intensity:

where (_p) denotes product, (_o) country or region of origin, and (_m) the driver (land use, blue water use, nitrogen emissions, phosphorus emissions). (Q_p,o,m) is the pressure attributable to national (German) consumption from origin (_o) for product (_p) (units: ha for land use, m³ for water use, kg N and kg P for fertilizer application). (CDE_p) is the national apparent consumption of product (_p) (kg), (i_p,o,m) is the origin-specific intensity (unit per kg), and (I_p,m) is the product-average (consumption-weighted) intensity (unit per kg).

Multiplying by the canteen’s purchased mass (K_p^canteen) (kg) gives absolute values by product and driver

with an origin breakdown for further analysis

Canteen-level totals are

We report land (cropland and grassland), blue water, and N/P application. Consistency checks include (left(6right),{varSigma }_{{{rm{o}}}}{Q}_{p,o,m}) matching the product’s total pressure and (left(7right),{varSigma }_{{{rm{o}}}}{F}_{p},o,{m}^{{{rm{canteen}}}}=,{F}_{p},{m}^{{{rm{canteen}}}}left.right).,)

For animal-based products, the analysis considered the type and origin of the animal feed (indirect crop feed impacts from FABIO) including cropland for fodder crops (soy, maize, alfalfa) and pasture/grazing land. FABIO disaggregates feed-related land use into distinct feed sourcing pathways while maintaining consistent land use type classifications. Grazing land is attributed based on GLEAM (Global Livestock Environmental Assessment Model)¹¹³ and Krausmann et al.‘s¹¹⁴ feed use estimates where only grass biomass consumed by livestock is converted to area equivalents based on spatially explicit productivity data. This methodology calculates a functionally allocated grassland area rather than total pasture extent which means that only the grass biomass actually consumed by livestock is attributed to production, not the entire pasture area (see Vanham et al.⁵⁸). Animal husbandry infrastructure (barns, feedlots, manure facilities) is therefore excluded. This conservative approach may underestimate biodiversity impacts for extensively managed rangeland systems where livestock utilize only a fraction of available biomass, as the full ecological footprint (trampling, altered vegetation structure, modified disturbance regimes) extends beyond consumed biomass. Conversely, the approach may overestimate impacts for degraded or overgrazed pastures. No distinction is made between cultivated pastures (‘fodder crops’ where grass or herbaceous biomass is harvested from pastures as fodder) and ‘grazing’ (direct consumption of standing biomass) in FABIO, with both aggregated under ‘pasture’ classifications (used synonymously).

The analysis encompasses upstream impacts from feed production, including irrigation water use and (synthetic) fertilizer inputs (N and P), which are allocated to livestock products via the respective environmental extension accounts. Water and fertilizers (N/P) used in the production of feed are included in the respective extensions.

FABIO’s nitrogen (N) and phosphorus (P) data represents inputs from synthetic (inorganic) fertilizer application only. Organic inputs (e.g., manure), biological N fixation, atmospheric deposition, and other N/P flow components are not included. An updated and more comprehensive N/P dataset is currently under development. To calculate freshwater eutrophication we calculated N and P leaching using PEF¹¹⁵ and IPCC¹¹⁶ estimations. For water use impacts we include blue water use which refers to freshwater withdrawal from surface and groundwater sources for agricultural irrigation. For greenhouse gas emissions, we used an existing inventory of environmental footprints of average food products sold in Germany including all life cycle stages until the supermarket checkout¹¹⁷.

Step 3: Biodiversity impact assessment

The resulting environmental pressure data was combined with spatially explicit (country level) characterization factors (CFs) from the UNEP Life Cycle Initiative’s GLAM framework¹¹⁸ to estimate potential biodiversity loss in Potentially Disappeared Fractions of Species (PDF.yr). These damage-oriented CFs follow a cause–effect chain that links environmental pressures to species loss through fate, exposure, and effect components, though not all are required for every impact driver. The GLAM CFs integrate high-resolution ecological data (species distributions, habitat conditions, and stressor-specific responses) and are available at regional and global scales.

The global PDF.yr (variously notated as: PDF.y, PDF·year, PDFy; units: PDF·yr·kg⁻¹) indicator denotes the proportion of species potentially lost due to an environmental pressure, linking the different environmental impacts quantified by the midpoint impact categories to the Area of Protection (AoP) ecosystem quality (see Supplementary Information 1 for extended description; SN2.3). PDF is an indicator of ecosystem integrity derived from species loss^12,73,74. It quantifies the relative reduction in species richness compared to reference conditions and is usually expressed on a scale from 0 to 1, where 0 indicates that all original species persist and 1 that all original species have been lost. In our case PDF·yr should not be interpreted as a fraction, but as an aggregated extinction risk integrated over area and time. PDF·yr can therefore exceed 1 analogous to how a probability integrated over time can exceed 1.

Characterisation factors (CFs) translate a unit of environmental pressure, e.g. one kilogram of emitted substance, one square meter of occupied land, or one degree of warming, into a predicted fractional loss of species, integrated over area and time. For example, a CF of 1 × 10⁻⁸ PDF·yr·kg⁻¹ implies that emitting one kilogram of a stressor is associated with the equivalent of one hundred-millionth of the local species pool disappearing for one year, or equivalently, a 10% reduction in species richness across ten square meters for one year.

Local and regional species loss estimates are scaled to global extinction risk through the application of Global Extinction Probabilities (GEPs) following Verones et al.¹⁰⁴. For each species present in a given region, the GEP integrates three quantities: the proportion of that species’ global range falling within the region, the certainty of its occurrence there, and a weight reflecting its IUCN Red List threat status. By construction, GEPs sum to 1 globally across all regions for each species group, meaning the metric directly expresses the fraction of global extinction risk concentrated in a given location. A region hosting many small-ranged or highly threatened species, such as Madagascar for vascular plants or the Congo Basin for freshwater fish, therefore carries a disproportionately high GEP relative to its area, and the same unit of local species loss there translates into substantially greater global extinction risk than an equivalent loss in a species-rich but range-abundant region. By multiplying regional PDF-based CFs by the corresponding GEP, local impacts are converted into estimates of potential irreversible global species extinction, providing an endpoint indicator for the AoP ecosystem quality. The GEP framework used here draws on 98,212 species across terrestrial, freshwater, and marine realms, spanning vascular plants, five vertebrate classes, and selected invertebrate and marine taxa¹⁰⁴.

Our analysis focused on: phosphorus and nitrogen-driven eutrophication impacts on freshwater fish biodiversity¹⁰⁷; water consumption impacts on riverine fish species¹¹⁹; greenhouse gas emissions impacts on terrestrial and marine species¹²⁰, and freshwater fish species¹²¹ as well as land use occupation impacts on vertebrates³⁷. The resulting PDF·yr value therefore represents not a count of species extinctions, but the expected proportional reduction in species richness, summed across all contributing pressures throughout the product system.

Following GLAM guidance, we applied average global-scale species-loss CFs to capture irreversible biodiversity damages^37,104. For each pressure and spatial unit, we computed pressure-specific impacts as a(p,i) · CF(p,i) (PDF·yr), where a(p,i) is the activity or pressure amount from Step 2 (e.g., m²·yr, kg P). Where country-level CFs were unavailable, we used proxies from CFs from ecologically similar neighboring countries based on biogeographic realm, dominant biome, and climatic similarity. Preference was given to countries within the same biogeographic realm to ensure comparable species pools and evolutionary history using e.g. spatial proximity or geographic similarity (e.g. climatic zone). Within each realm, countries were matched based on dominant biome composition, climatic conditions, and overall ecosystem structure (see Supplementary Table 5).

We summed impacts across spatial and taxonomic units and across pressures to produce a combined biodiversity impact. Results are reported as the aggregated total and disaggregated by pressure and origin to retain interpretability and identify dominant contributors.

The following outlines how the potential biodiversity loss was calculated for each environmental pressure. Further descriptions are provided in the Supplementary Information 1 and a full list of CFs is provided in the Supplementary Data 1.

Impacts on biodiversity from land use

Based on the consumption volumes (Step 1) of each food product, the land use occupation (in m²) was estimated using FABIO v2 (Step 2). The resulting country-specific land use requirements (cropland and pasture) for each food product were translated into a biodiversity impact using spatially explicit (country level) characterization factors (CFs), focusing on land occupation impacts from cropland, pastures and plantations³⁷. The CFs are based on species–habitat relationships, considering both intensity levels and habitat fragmentation to allow an estimate of potential species loss. The CFs cover five taxonomic groups (plants, amphibians, birds, mammals, and reptiles) and five broad land-use types (cropland, pasture, plantations, managed forests, and urban land) at three intensity levels (minimal, light, and intense) across terrestrial ecoregions and countries. For land use CFs, empirical species–response relationships derived from field surveys and macroecological datasets including the PREDICTS database, are used to estimate relative species richness under disturbed versus reference conditions across biomes and land use types, yielding a fractional local species loss per unit area. These regional loss estimates are then scaled to global extinction risk by multiplication with Global Extinction Probabilities (GEPs), which weight each region by the proportion of species’ global ranges present there, the certainty of their occurrence, and their IUCN Red List threat status, independently of any disturbance response.

For land occupation, we applied the average global species loss CFs (CFocc_avg_glo) provided by Scherer et al.³⁷. These characterization factors (CFs) quantify biodiversity loss as the relative species loss per unit area (PDF/m²), where the reference state represents a hypothetical biotic potential defined not as the original or future successional biodiversity state at the same location, but as the current natural habitat elsewhere in the same ecoregion³⁷. To integrate the characterization factors (CFs) with our life cycle inventory data, we assigned each food product to the appropriate land-use classes and intensity level (minimal, light and intense). For land use classes³⁷ follows the definitions of ref. ¹²², classifying orchards, shrubs and tree-like plants (palms and vines) as plantations rather than cropland. Therefore, coffee, palm oil, fruits and grapes were classified as plantations in this study. Cropland was the primary category used for most other products (see Supplementary Table 4).

For land use intensity, we used the country-specific intensity levels provided in the supplementary material of ref. ⁶⁶. In their approach⁶⁶, overlaid the intensity maps from ref. ³⁷ with the HILDA+ cropland dataset¹²³ and national boundaries from GADM¹²⁴ to derive country-specific intensity shares. These resulting shares were applied directly in our analysis using a weighted approach, i.e., each intensity category contributed proportionally to its country-specific share (see Supplementary Table 5). Country-specific intensity levels were unavailable for 41 countries and territories, requiring substitution with values from neighboring countries. However, as none of the top 40 sourcing countries by procurement volume or biodiversity impact required proxy values, and their combined contribution is 0.3% of total assessed impact, this gap does not affect primary hotspot identification.

For each food product, the CFs across all five taxonomic groups were totaled to generate a single biodiversity impact per country. This was then used in subsequent aggregation steps for the overall assessment.

Impacts on biodiversity from GHG emissions (climate change)

Based on the emission data provided in ref. ¹¹⁷, greenhouse gas (GHG) emissions associated with the production of each food product were calculated in CO₂eq. These data represent an average food product sold in Germany, weighted across domestic production and import shares, countries of origin, cultivation methods (open field and greenhouse) across all months of the year including seasonal and non-seasonal production, and respective transport modes (sea and air freight). The data covers CO₂, CH₄, and N₂O (summed as CO₂-equivalents) and including land-use change emissions via an attributional approach. To assess potential biodiversity impacts from the resulting greenhouse gas inventory, CFs were taken from ref. ¹²⁰ for terrestrial and marine impacts, and from ref. ¹²¹ for freshwater impacts. Both datasets are designed for application in footprint studies and are made publicly available at multiple levels of spatial (i.e., individual grid cells, biogeographical realms, or global averages) and taxonomic (i.e., individual or aggregated species groups) resolution, thereby facilitating their integration into commonly used assessment tools. Due to the lack of country-resolved inventory data, the global average level was applied in this study. Consequently, climate change impacts are not included in the country-specific analysis.

Iordan et al.¹²⁰ provide spatially and taxonomically specific CFs for the impacts of GHGs on terrestrial and marine biodiversity. They assessed the vulnerability of different species (n = 26,648) to a projected near-surface temperature increase to identify those species most sensitive to changes in temperature and climate-driven environmental conditions across various geographic regions. By combining species vulnerability with the spatial distribution of GHG emissions, the method generates impact estimates that vary according to both the location of emissions and the taxa affected. To operationalize this, the authors derived CFs for each combination of region and taxon. These CFs translate a given quantity of emissions (e.g. kilograms of CO₂-eq) into expected biodiversity impacts at grid cell level, reflecting differences in exposure, sensitivity, and adaptive capacity among species. Different global warming scenarios (RCP2.6, 4.5, and 8.5) were used to develop CFs for different years (2050;2100) in average potentially affected fraction (PAF) of species per taxonomic group, reflecting the fraction of species exposed to pressure levels exceeding their tolerance thresholds. For impacts on freshwater ecosystems, de Visser et al.¹²¹ provide CFs per unit of GHG emissions accounting for climate-driven changes in both streamflow and water temperature extremes. To construct these CFs, the authors integrated climate models (global mean temperature increases) projecting future streamflow and temperature changes with species distribution models and ecological vulnerability assessments. Average effect factors were derived from mapping the global extinction risk against corresponding global mean temperature increases under different climate scenarios, using global extinction risks based on more than 11,000 threatened species of riverine fish, or 76% of the total freshwater fish species. The derived CFs account for regional variations in climate change effects and differentiate between species with varying sensitivities to thermal and hydrological changes (taxonomically explicit). These can be used in LCIA to convert inventory data on GHG emissions into an estimate of the impact on freshwater fish biodiversity in PDF.yr·kg-1 (further methodological details are provided in the Supplementary Information 1; SN2.3).

In our study, we applied global average CFs in PDF.yr·kg⁻¹.

Impacts on biodiversity from freshwater eutrophication

Using the crop- and country-specific fertilizer application rates (P and N requirements) produced with FABIO (Step 2) we estimated N and P emissions to freshwater bodies, applying a fixed average leaching factor. The global IPCC default methodology for estimating these indirect N emissions assumes that a fixed fraction of N inputs (Frac_LEACH) is lost through leaching and runoff. The IPCC has recently updated estimates of Frac_LEACH (wet climates and dry climates where irrigation other than drip irrigation is used) to 0.24 kg for N (kg N input)⁻¹¹¹⁶. For P we assume leaching and runoff to water bodies of 5% (comparable to)^115,125. It is important to note that this approach represents average conditions and does not capture the heterogeneity of fertilization practices at different spatial scales.

To assess potential biodiversity impacts on freshwater fish species, we used spatially explicit (country level) CFs developed by Zhou et al.¹⁰⁷ to link these emissions to potential species loss. These characterization factors couple nutrient fate and concentration outputs with long-term observations of freshwater fish richness (~13,920 species over 41 years) to quantify species loss at high spatial resolution. The approach accounts for emission pathways (direct, diffuse, erosion), nutrient limitation (P- vs. N-limited regions), and integrates global extinction probabilities¹⁰⁴ to translate local impacts into global species loss. CFs express the Potentially Disappeared Fraction (PDF) of species per kg of nutrient emitted (PDF.yr·kg⁻¹), enabling country-level impact assessment (further methodological details are provided in the Supplementary Information 1; SN2.3). In our analysis, we used the country-level CFs considering P (or N, respectively)-limited regions (average CF for direct emissions in PDF.yr/kgP or N).

Impacts on biodiversity from water use

The country-specific blue water use (ground and surface water) associated with each food product consumed was calculated based on FABIO v2 (Step 2). Potential biodiversity impacts were then assessed by linking country-weighted water requirements to riverine fish species loss using the spatially explicit (country level) characterization factors (CFs) developed by Pierrat et al.¹¹⁹. These CFs quantify the downstream effects of water consumption on freshwater ecosystem quality, expressed as the Potentially Disappeared Fraction (PDF) of species per unit of water consumed (m³·yr⁻¹). The rationale behind the CFs is that water consumption can reduce streamflow, potentially leading to species loss within the basin affected. The CFs are based on a regionalized species–discharge relationship (SDR) model covering 2320 river basins and 11,450 fish species¹²⁶, which links reduced river discharge from upstream consumption to species loss. Fate factors represent changes in river discharge, while effect factors capture the resulting biodiversity loss, yielding river basin-specific CFs. To account for irreversible global extinctions, these regional CFs were then further weighted by Global Extinction Probabilities¹⁰⁴ (further methodological details are provided in the Supplementary Information 1; SN2.3).

In our analysis, we used the average country level (global) characterization factor value including the mean regression coefficient (CF_GLOB_A_m) in PDF.yr.m-3.

Step 4: Comparison with non-regionalized inventory

The resulting regionalized biodiversity impact assessment of food products consumed in Germany (Step 1–3) was benchmarked against a global dataset derived from ref. ¹²⁷, which has been previously used as a data source for multiple biodiversity impact assessment studies of food consumption^50,51,128. The dataset provides environmental impacts for commercially viable food production systems from a globally harmonized meta‑analysis covering 38,700 farms and 1600 processors (2000–2016).

For the comparison units (per kg at retail) were aligned (if necessary) and product classifications were harmonized. Where no exact match was available, the most representative product was used (see Supplementary Data 1). Products that were not available in the dataset at all, and for which no comparable items existed were excluded from the comparison (e.g. coconuts, spices, pepper), yielding n = 34 products for comparison. For these 34 products we consistently applied the median values provided in the supplementary material of ref. ¹²⁷. Biodiversity impacts were then derived using the same LCIA method sources of CFs (Step 3). In the regionalized approach, we applied spatially explicit (country‑level) CFs, while for the global‑average benchmark we used the corresponding global‑average CFs provided in the same datasets.

To systematically isolate the contributions of inventory regionalization versus characterization factor regionalization to observed differences in biodiversity impact results, we implemented a three-case decomposition analysis. Case 1 (g × g) combines the global-average inventory data from Poore and Nemecek¹²⁷ with global-average characterization factors, serving as the baseline ‘global × global’ scenario. For animal products in Case 1, we applied a weighted land use CF reflecting the global land use distribution for livestock production: 86.5% pasture/grazing land and 13.5% cropland for feed production (CF_mixed = 0.865 × CF_pasture + 0.135 × CF_cropland), consistent with the global livestock land use pattern reported by Poore and Nemecek¹²⁷ where approximately 3.1 billion hectares of pasture and 0.5 billion hectares of cropland are dedicated to animal agriculture. Case 2 (r × g) substitutes our regionalized German consumption inventory (Steps 1–2) while maintaining the same global-average CFs used in Case 1, thereby isolating the inventory effect, the change attributable solely to spatially differentiated production data. Case 3 (r × r) applies our full regionalized approach (Steps 1–3), combining regionalized inventory with regionalized, spatially explicit CFs, representing the complete regionalization scenario. The CF effect is then quantified as the difference between Case 3 and Case 2, capturing the influence of spatially differentiated impact assessment independently of inventory choices. Formally, the decomposition is expressed as: Inventory effect = Case 2 − Case 1; CF effect = Case 3 − Case 2; Total regionalization effect = Case 3 − Case 1. This additive structure enables attribution of observed changes in biodiversity impact estimates to either improved spatial resolution in the life cycle inventory (foreground system) or in the impact pathway characterization (LCIA), thereby clarifying whether shifts in commodity rankings are driven by knowing where production occurs or by knowing how sensitive that location is to species loss.

Several methodological differences constrain direct comparability. First, Poore and Nemecek¹²⁷ report total water consumption, whereas our FABIO-based inventory tracks only blue water (surface and groundwater withdrawals). Second, they provide only phosphorus emissions. For freshwater eutrophication from P we calculated an average based on all country values provided. Our comparison is thus limited to P-driven freshwater eutrophication, excluding nitrogen contributions. Third, system boundaries differ substantially as Poore and Nemecek¹²⁷ provide cradle-to-retail inventories (production, processing, packaging, transport, retail), whereas our regionalized inventory captures only the agricultural production stage (farm gate). Consequently, the inventory effect (Case 2 − Case 1) conflates agricultural production differences with the exclusion of post-farm-gate stages, representing a lower-bound estimate of regional production efficiency. The CF effect (Case 3 − Case 2) remains unaffected, as system boundaries are identical between these cases.