Trade-offs in the externalities of pig production are not inevitable

Livestock farming provides substantial nutrition but imposes large external costs, occupying 75% of agricultural land, contributing 14–17% of anthropogenic GHG emissions, and using more antimicrobials than human medicine. Pig production, in particular, has quadrupled in 50 years and is projected to continue growing. Externalities are often assumed to trade off—systems performing well in one domain are presumed to perform poorly in others—yet empirical, systematic quantification across multiple externalities and livestock systems is rare. This study examines four critical externalities—land use (also a proxy for biodiversity loss), greenhouse gas (GHG) emissions, antimicrobial use (AMU; proxy for antimicrobial resistance), and animal welfare—across diverse commercial pig production systems in the UK and Brazil. The research aims to quantify these externalities, test whether they trade off, and identify best- and worst-performing systems and system types.

Prior work typically assessed externalities in isolation, with few studies quantifying co-variation across systems and outcomes. Mixed associations have been reported: negative associations between freshwater use and GHGs and between freshwater use and eutrophication in tomato systems, but positive associations between GHGs and eutrophication; positive associations among GHG and soil organic carbon in Chinese grain; and among GHGs, acidification and eutrophication in Iranian wheat and barley. In livestock, positive associations have been shown among land-use, GHG, nitrogen, phosphorus and soil impacts in European dairy; between land-use and GHG in Brazilian beef; and among GHGs, acidification, eutrophication, non-renewable energy and land use in European beef, though with some trade-offs involving metal depletion and acidification. One study on Italian dairy found no association between animal welfare and AMU. Notably, no prior work has jointly examined environmental costs and animal welfare across livestock systems. This gap complicates guidance on mitigation, as assumed relations risk counterproductive decisions.

Design and sampling: The study covered 74 UK and 17 Brazilian breed-to-finish pig production systems (each comprising one to three farms), representing most commercial types globally. Farmers were recruited via industry contacts, searches and social media, with targeted recruitment to minimize bias. Some systems shared breeding/rearing herds; statistical analyses used a subset (UK n=43; Brazil n=8) with one randomly selected datapoint per shared upstream unit. UK systems were classified by label type (Red Tractor, RSPCA Assured, Organic; plus free range and woodland) and by husbandry type at breeding and finishing. Brazilian systems were classified by husbandry type (including globally relevant practices such as gestation crates and growth promoters) as formal labels are not established. Data collection: Between Sept 2017 and Dec 2020, farm visits included detailed welfare assessments and questionnaire-based interviews to gather annual data on inputs and outputs needed to estimate land-use, GHG, AMU and welfare costs per kg deadweight (kg DW). When upstream farms supplied multiple finishers, externality costs were allocated proportionally by pig flow. System boundaries and functional unit: Boundaries included feed production, pig production (breeding to finishing), slaughter and processing. All externalities were expressed per kg DW (including meat from finishing pigs and culled sows, with economic allocation based on UK price data; same allocation applied to Brazil due to data limitations). Land-use cost: Measured as total m²·yr per kg DW, summing land to rear pigs and to grow feed. Land under woodland used to rear pigs was assumed to have biodiversity value equivalent to native habitat and was excluded from land-use cost. Feed land use was calculated using farm- and stage-specific formulations and annual quantities, yields from FeedPrint (country-specific where possible), and gap-filling for unknowns. Organic feed yields were adjusted using literature-based conventional vs. organic yield differences. Co-products used economic allocation. GHG cost: Total kg CO₂e per kg DW included emissions from feed (production, milling, processing via FeedPrint), enteric methane, manure management (methane and nitrous oxide), fuel and electricity, transport (farm-specific distances where available; references for emission factors by country), slaughter and processing (assumed from prior literature), plus forgone sequestration (carbon opportunity cost of land use). Methane and N₂O were converted using GWP100. Forgone sequestration used aboveground biomass accrual and soil carbon change values with habitat- and region-specific parameters, assuming 20-year transitions. Sensitivity analyses indicated broad result robustness to assumptions on displaced fertilizer and sequestration rates. AMU cost: Total milligrams of antimicrobials used per kg DW, derived from farm medicine records for the most recent year. In addition to total AMU, critically important antimicrobials for human health (EMA Category B) were quantified and reported (main analyses focus on total AMU for comparability). Animal-welfare cost: Following a previously developed LCA-compatible welfare metric, the number of life years required per kg DW was weighted by quality-of-life scores from Welfare Quality (WQ)-based assessments. WQ-certified assessment covered sows, pre-weaning piglets and fattening pigs, yielding four principle scores (good health, feeding, housing, appropriate behaviour) combined with weights (0.35, 0.25, 0.15, 0.25). Crated sows received the worst score for social behaviour during crating periods. Time with WQ principle scores ≥80 was treated as a welfare benefit (negative cost). Sow/piglet and finisher components were combined by multiplying life years by WQ scores and summing. Statistical analyses: Due to non-independence, analyses were performed on subsets with one datapoint per shared upstream unit (UK n=43; Brazil n=8). Associations used two-sided Spearman rank correlations. Group comparisons used Wilcoxon rank-sum and Kruskal–Wallis tests with Holm-adjusted Dunn’s post hoc tests. Analyses were conducted in RStudio 4.1.1.

Externality ranges: Across UK and Brazil, land-use costs varied 12-fold (3.0–35.8 m²·yr·kgDW⁻¹) and GHG costs 9-fold (6.2–55.9 kgCO₂e·kgDW⁻¹; 1.3–12.2 kgCO₂e·kgDW⁻¹ excluding forgone sequestration). Total AMU ranged 0–606 mg·kgDW⁻¹; critically important AMU 0–65.7 mg·kgDW⁻¹. Animal-welfare costs ranged from harmful to beneficial (negative costs). Country differences: Land-use and GHG costs did not differ significantly between UK and Brazil (Wilcoxon P > 0.2). Brazilian systems had significantly higher total and critically important AMU costs and higher welfare costs than UK systems (all P < 0.01). UK label-type differences: Land-use costs differed among label types (Kruskal–Wallis χ² = 23.3, d.f. = 5, P < 0.01); Organic > RSPCA Assured, Red Tractor, None (post hoc Dunn’s all P < 0.01). GHG costs also differed (χ² = 21.9, d.f. = 5, P < 0.01); Organic > RSPCA Assured (P < 0.01), Organic > Red Tractor (P > 0.01 reported; specific pairwise P = 0.02 vs None). Total AMU differed overall (χ² = 11.7, d.f. = 5, P = 0.04) but no pairwise differences (all P > 0.2). Critically important AMU did not differ among label types. Welfare costs differed (χ² = 34.5, d.f. = 5, P < 0.01), with higher costs in None and Red Tractor than in Woodland and Organic (P = 0.01, P < 0.01; both P < 0.01, respectively) and higher in Red Tractor than Free range (P = 0.01). Associations in UK systems: Strong positive association between land-use and GHG costs (Spearman r_s = 0.97, P < 0.01; r_s = 0.84 excluding forgone sequestration). Moderate positive association between AMU and welfare costs (r_s = 0.54, P < 0.01). No significant association between welfare cost and critically important AMU. Negative associations indicated trade-offs: land use vs AMU (r_s = −0.40, P < 0.01), land use vs welfare (r_s = −0.55, P < 0.01), GHG vs AMU (r_s = −0.41, P < 0.01), and GHG vs welfare (r_s = −0.38, P < 0.01). Best-performing systems (UK): Despite broad trade-offs, several systems combined low costs across pairs of negatively associated externalities. Five systems ranked in the best-performing 50% for all four externalities: three RSPCA Assured (outdoor-bred, straw-yard finished), one fully outdoor Woodland system, and one Red Tractor system (hybrid indoor–outdoor breeding, slatted finishing). One system (RSPCA Assured, outdoor-bred, straw-yard finished) was in the best-performing 25% for all four costs. A further 10 systems were in the best-performing 50% for three externalities. No Organic or Free-range systems were in the best-performing 50% for three or more domains because none was in the best 50% for land-use or GHG; however, 100% of Organic (6/6), 61% of Free-range (11/18), and all Woodland (3/3) systems were in the best 50% for both welfare and AMU. All label and husbandry types also had poorer performers; four systems (three Red Tractor, one RSPCA Assured) were in the bottom 50% across all four externalities, and one Red Tractor system was in the bottom 25% for all. Husbandry-type contrasts (selected): Indoor-bred systems had lower GHG costs than outdoor-bred; outdoor-finished systems had lower GHG than straw yard and slatted. Indoor-bred systems had higher welfare costs than outdoor-bred; slatted-finished systems had higher welfare costs than outdoor-finished. Critically important AMU was higher in indoor-bred compared with outdoor-bred systems. Brazilian systems: Positive association between land use and GHG (r = 0.98, P < 0.01) and suggestive positive association between AMU and welfare (r = 0.69, P = 0.07); other pairwise associations were not significant, potentially reflecting smaller sample size. Three systems were in the top 50% for all four externalities and five more for three externalities. Scenario analysis (UK): If 2021 UK pig production were entirely Organic (using study medians), AMU would fall by 88–98% (8–56 tonnes less active ingredient per year) and welfare would improve relative to production entirely via Red Tractor, RSPCA Assured or unlabelled systems. However, total climate impact would roughly triple (+25 million t CO₂e per year; +2.6 million t excluding forgone sequestration) and land use would quadruple (+1.7 million ha per year; ~10% of UK agricultural land).

Findings confirm that while broad trade-offs exist—systems with low land use typically have low GHG but higher AMU and poorer welfare—these trade-offs are not inevitable. Several real-world systems achieved above-average performance across all four externalities, demonstrating the potential for co-benefits. No label, husbandry, or farming type reliably identified the best or worst performers, indicating that current classification and labelling schemes do not effectively guide consumers or regulators toward low-externality systems. Given the substantial within-type variation (e.g., land-use costs varied more than fivefold within Woodland systems), mitigation strategies should prioritize measurable improvements within system types rather than relying on wholesale shifts between types. The scenario analysis illustrates the systemic consequences of scaling a single system: while Organic could substantially reduce AMU and improve welfare, it would markedly increase land and climate impacts. Stakeholder priorities across externalities will shape preferred pathways, underscoring the need for multi-criteria optimization and policy instruments that incentivize holistic performance.

This study provides the first systematic, empirical comparison of land-use, GHG, AMU and animal-welfare externalities across diverse commercial pig systems in two countries, showing that trade-offs are common but not unavoidable. Some systems perform well across all domains, but no label or husbandry category consistently identifies them. Improving outcomes within system types and developing outcome-based assessment and labelling could more effectively drive sustainability gains. Future research should incorporate additional environmental externalities (eutrophication, acidification), refine biodiversity impact accounting beyond land-use proxies, and broaden scope to include social outcomes (public health, financial viability, scalability, farmer well-being). Extending similar multi-externality analyses to other livestock and agricultural sectors is essential for identifying and promoting practices that minimize total societal costs.

The dataset, while large and diverse, includes few examples of some system types (e.g., three Woodland farms) and a limited Brazilian sample, constraining statistical power and generalizability. Some data required imputation (feed formulations/ingredients, origins, transport), and Brazilian antimicrobial records varied in quality, though excluding lower-quality records did not alter conclusions. Welfare assessments provide snapshots and require assumptions in aggregating WQ scores. GHG and land-use estimates depend on choices of emission factors, feed footprint databases, allocation methods, and assumptions about displaced fertilizer and forgone sequestration; sensitivity analyses suggest broad robustness but residual uncertainty remains. Land-use cost was used as a proxy for biodiversity impact and did not capture regional biodiversity differences beyond woodland assumptions. Only four externalities were assessed; other important environmental and social outcomes were outside scope.