## Main

China is suffering from severe environmental pollution caused by nitrogen emissions from agriculture, industry and traffic. These emissions contribute to high ammonia (NH3), nitrogen oxide (NOx) and particulate matter (PM2.5) concentrations in the air1,2, and to eutrophication of surface waters and groundwater3. A substantial fraction of the nitrogen losses to air and water bodies originates from the livestock production–consumption chain4,5, which is partly due to the increasing decoupling of crop production and livestock production at farm level6,7. However, the geographic segregation of crop and livestock production systems and the concentration of livestock production in few regions have received less attention.

Based on the regional-level ratio of livestock manure nitrogen excretion to maximal manure nitrogen application to cropland, livestock production is unevenly distributed in China6,8. Livestock production is concentrated in a few regions in China and consequently manure nutrient excretion exceeds the nutrient uptake capacity of local cropland in these regions9,10,11. Transportation of manure to distant cropland is costly, due to its high water content, and hence unprocessed manure is rarely transported over large distances and is often discharged to the environment directly12. Hence, an uneven distribution of livestock production, in particular a high livestock density near urban areas, may lead to a high proportion of the population being exposed to multiple pollutants in the air and in watercourses.

The demand for livestock products will increase in China in the coming decades13,14,15 and it is important to know where the additional livestock should be raised. It has been suggested that a geographically explicit recoupling of crop and livestock production systems would minimize the risk of high nutrient losses, create a situation with a better and more uniform environmental quality, and keep livestock farms at safe distances from residential areas6,7,16. This requires spatial planning of livestock production across and within regions. As a co-benefit of livestock spatial planning, the recycling and utilization of phosphorus and other nutrients, and of organic matter, can be made more efficient17,18. However, little quantitative information is currently available about the impacts of an improved spatial distribution of livestock farms. The objectives of our study are: (1) to analyse the driving forces and impacts of spatial distributions of livestock production in China for the period 1990–2012; and (2) to explore the possible impacts of a spatially more even distribution of livestock production on nitrogen use and losses in 2050. We focus on nitrogen because of the severity of nitrogen pollution in China and for reasons of brevity. The NUtrient flows in Food chains, Environment and Resources use (NUFER) model was adjusted and a multicriteria-based livestock spatial planning concept was developed for the purpose of this study.

## Results

### Unevenly distributed livestock production

The livestock population in China increased by 80% between 1990 and 2012, from 240 to 430 million standard livestock units (LU). A standard livestock unit equals a 500 kg dairy cow19. About 98% of the increase occurred to the right of the Huhuanyong Line (Hu Line) (Fig. 1a,c). The main livestock production counties (that is, counties with more than 320,000 LU), contributed 9.2% to the total LU in 1990 and 55% in 2012 (Fig. 1a,c and Extended Data Fig. 1). The geographic connection between crop and livestock production greatly decreased between 1990 and 2012, as reflected by the rapid increase in livestock density and the increasing ratio of annual manure nitrogen excretion to annual crop nitrogen uptake in some counties (Fig. 1b,d and Extended Data Fig. 2). Over 37% of the counties to the right of the Hu Line had a livestock density >4.0 LU ha−1 in 2012 (Fig. 1b,d and Extended Data Fig. 1). For comparison, Denmark regulates maximum livestock density to 2.0 LU ha−1 to limit nutrient pollution by livestock production20. A livestock density of ≤2 LU ha−1 is a rough indication that the produced manure nutrients can be absorbed by the growing crops in that area, depending also on soil and climatic conditions and management.

The concentration of livestock production in regions is not unique to China; it has been observed in many developed and emerging countries, including the United States (Extended Data Fig. 3). The ratio of annual manure nitrogen excretion to annual crop nitrogen uptake was >2.0 in 11% of the 3,100 counties in the United States in 1992 and in 22% of the counties in 2012. The contribution of these counties to the total manure nitrogen production has increased from 21% to 32% between 1992 and 201221 (Extended Data Fig. 3). Although the United States and China have roughly similar areas of land and cropland, the numbers of humans and livestock are much larger in China, and hence the concentrations of humans and livestock in specific regions are much higher in China than in the United States (Extended Data Figs. 2 and 3). The ratio of synthetic fertilizer nitrogen use to crop nitrogen uptake has a different geospatial trend to the ratio of manure nitrogen excretion to crop nitrogen uptake in the United States than in China. This suggests that more manure nitrogen was recycled effectively and substituted for synthetic fertilizer nitrogen in the United States than in China. This has resulted in a relatively low fertilizer nitrogen application rates in livestock-dense regions in the United States compared with those in China in 2012 (Extended Data Fig. 3).

### Driving forces of uneven livestock distribution

The location of livestock farms is influenced by several factors, such as vicinity of markets, traffic infrastructure, availability of animal feed, availability of animal health and epidemic prevention services, and slaughter and milk processing services9,22,23,24. Concentration of livestock production near markets contributes to improving livestock production efficiency, lowering production costs and shortening the delivery distances25,26. Our panel data analysis showed that the livestock production of a county was significantly and positively correlated with human population and gross domestic product, suggesting that the thickness of local demand played an important role in shaping the geospatial patterns of livestock production during the period 1990–2012 (Extended Data Table 1). River intensity had a significant positive correlation with livestock density in the no-fixed-effects model (Extended Data Table 1). Feed availability (measured by the production of maize and soybean of a county) and feed accessibility (measured by the road length of a county) were also not significantly related to livestock density in the fixed-effects model analysis (Extended Data Table 1).

### Nitrogen pollution impacted by unevenly distributed livestock

There were positive correlations between livestock density, NH3 emissions and nitrogen losses to water in counties in 1990 and 2012 (Fig. 2). There was also positive correlation between livestock density and average synthetic nitrogen fertilizer use per hectare of cropland in counties in 1990 and 2012, which indirectly suggests that little or no synthetic fertilizer nitrogen was replaced by animal manure nitrogen (Fig. 2). The slightly higher slope of the linear regression between livestock density and NH3 emission intensity in 1990 than in 2012 is probably related to lower crop–livestock integration and to improved animal feeding in 2012 compared to 1990. The uneven distribution of livestock production at county level led to a summed manure surplus of 1.8 TgN in 1990 and of 4.6 TgN in 2012 (Fig. 3b). Manure nitrogen surplus is defined here as the difference between total manure nitrogen excretion and total crop nitrogen demand, corrected for the supply of nitrogen by soil, crop residues, waste, biological N2 fixation and atmospheric deposition (thus without synthetic fertilizer nitrogen), at county level (for more detail, see equation (3)). Hence, this manure nitrogen surplus could not be recycled in cropland and thus was largely lost to air and water bodies5. At the same time, there was a rapid increase in synthetic nitrogen fertilizer use, which hindered the effective recycling of livestock manure nitrogen further3,5.

Average airborne PM2.5 concentrations in a county were more positively correlated with livestock nitrogen excretion per county, rather than with fertilizer nitrogen input, or with airborne SO2 and NOx concentrations (Extended Data Fig. 4). Regions with high airborne PM2.5 concentrations correlated with regions with high livestock density. High PM2.5 concentrations contribute to respiratory diseases; there were 1.8 million people aged >65 years with premature mortality related to PM2.5 exposure in China in 201227, and over 60% of these deaths occurred in regions in which PM2.5 concentration was substantially correlated with livestock nitrogen excretions (Extended Data Fig. 4).

### Impacts of livestock production in 2050

It is probable that China will continue to suffer from unevenly distributed livestock production in the near future, especially with the predicted increase in the demand for livestock products. Domestic crop production and livestock production are projected to increase by 20% and 40%, respectively, in terms of protein production between 2012 and 2050 according to a business-as-usual (BAU) scenario28. As a result, total NH3 emissions and nitrogen losses to water bodies will increase by 12% between 2012 and 2050 (Fig. 3c,d). The modest increases of nitrogen losses to the environment in the BAU scenario reflect the balance between the expected increased production and hence increased nitrogen losses, and the expected impacts of policies and measures to decrease these losses (for example, Zero Fertilizer Action29,30, improved crop husbandry practices, improved irrigation and nutrient management31,32 and the replacement of fertilizer nitrogen by manure nitrogen33). The policies and measures together will reduce the average nitrogen losses per unit of crop product and especially per unit of animal product produced.

Around 26% of counties will have a livestock density >4.0 LU ha−1 and these counties will accommodate 58% of total LU in the BAU scenario in 2050 (Fig. 4a and Extended Data Fig. 1). This will result in a summed manure nitrogen surplus of 11 Tg; a surplus that cannot be recycled to cropland in an economically feasible and environmentally sound manner (Fig. 3b). This surplus may have to be converted to N2 via nitrification and denitrification process technologies or converted into desired manure nitrogen products, which would allow transport between counties in an economically feasible manner. The mean NH3 emission intensity will exceed the suggested threshold of 31 kgNH3 ha−1 in 56% of counties in the BAU scenario; the NH3 emission intensity threshold is one of the criteria for achieving Sustainable Development Goal 234. Around 60% of China’s population will live in regions with NH3 emission intensities higher than 31 kgNH3 ha−1 in the BAU scenario in 2050 (Fig. 3e,f).

### Impacts of pig relocation and improved technologies

The south-to-north transfer of pigs (SNT) scenario follows the pig relocation policy of the national government as proposed in 2017; it will reduce nitrogen losses to water bodies by 52% when compared to the BAU scenario due to the transfer of pigs from the watercourse-dense south to the semi-arid north. However, NH3 emissions will increase by 13% in the SNT scenario compared to BAU, in part due to a higher manure nitrogen recycling because discharge of manure from intensive livestock farms to watercourses has been banned (Fig. 3 and Extended Data Fig. 5). Much of the increase of the NH3 emissions will occur in northern China, where air pollution is most serious (Fig. 4b,e). Thus, the SNT scenario leads to pollution swapping from south to north: swapping water pollution in the south for air pollution in the north35,36.

In the integrated technology improvement (ITI) scenario, synthetic fertilizer nitrogen use, NH3 emissions and nitrogen losses to water will reduce by 82%, 63% and 77%, respectively, compared with the BAU scenario (Fig. 3a–d). Evidently, the improved recycling of livestock and human excreta, food waste and crop residues contributed to the substitution of synthetic fertilizers, and reduced nitrogen losses greatly6. However, a total of 11 Tg of manure nitrogen could not be effectively recycled due to the geographic separation of crop and livestock production systems in the ITI scenario. In addition, 40% of the population will still live in counties with NH3 emission intensities higher than the threshold value of 31 kgNH3 ha−1 yr−1 (Fig. 3f).

### Impacts of livestock spatial planning

Based on optimized spatial planning in the NH3 emission intensity criteria (SP-NH3) scenario, which was developed on top of the ITI scenario, we estimate that some 10 billion animals need to be relocated to be able to achieve a lower NH3 emission intensity for most of the population of China (Extended Data Fig. 6). This number is equivalent to 344 million LU, and includes 484 million pigs, 10 million dairy cattle, 30 million beef cattle, 9,300 million chickens, 146 million sheep and goats, and 1.5 million mules and donkeys (Extended Data Fig. 6). The number accounts for 56% of the domestic livestock population in the SP-NH3 scenario, and is equal to the current livestock populations of the United States and the Russian Federation combined30. Livestock need to be transferred mainly from southern and eastern China to north and southwest China (Extended Data Fig. 6). About 15% of the livestock need to be relocated within provincial borders, mainly to shorten the transportation distance to ensure the stability of food supply and reduce transportation costs (Extended Data Fig. 7).

In the SP-NH3 scenario, total NH3 emissions to air and nitrogen losses to water bodies will decrease by 60–70% compared with the BAU scenario. However, the SP-NH3 scenario had slightly higher total NH3 emissions to air and total nitrogen losses to water bodies than the ITI scenario (Fig. 3c,d). However, the SP-NH3 scenario decreased the proportion of people in counties with exposure to NH3 emissions higher than the threshold value compared with the BAU and ITI scenarios (Fig. 3e,f). Regions exceeding the NH3 emission intensity threshold value were mainly found in semi-arid intensive cropping areas, that is, mainly in the North China Plain (Fig. 4k). Manure nitrogen surplus and synthetic nitrogen fertilizer use were higher in the SP-NH3 than in the ITI scenario (Fig. 3a,b). This is counter to the objective to decrease manure nitrogen surplus and fertilizer nitrogen use through improved manure nitrogen recycling. Evidently, there is a trade-off here between keeping the NH3 emission intensity below the threshold value and bringing livestock production more closely to cropland.

Spatial planning according to crop–livestock integration criteria (SP-CLI), which was also developed on top of the ITI scenario, indicated the need to relocate 5.0 billion animals or 177 million LU, including 170 million pigs, 6.5 million dairy cattle, 31 million beef cattle, 4,323 million chickens, 146 million sheep and goats, and 2.6 million mules and donkeys (Extended Data Fig. 8). The number of relocated animals will be 50% less than in the SP-NH3 scenario, and 27% of the animals had to be relocated over relatively short distances in the SP-CLI scenario (Extended Data Fig. 7). The SP-CLI scenario resulted in 13% less NH3 emission, 40% lower nitrogen losses to water bodies, 99% less nitrogen fertilizer use and 47% less manure nitrogen wastage compared to the SP-NH3 scenario (Fig. 3a–d). However, 6.6 Tg of manure nitrogen was still not recycled because the local manure nitrogen supply exceeded the local nitrogen demand by the cropland, which could be reduced through importing more livestock products from global markets. The SP-CLI scenario led to a different geo-relocation of livestock production than the SP-NH3 scenario (Extended Data Fig. 9). The SP-CLI scenario showed a stronger performance in reducing total nitrogen losses, but a weaker performance in reducing the percentage of the population exposed to a high NH3 emission intensity compared to the SP-NH3 scenario (Fig. 3f). Evidently, there was a trade-off between keeping the NH3 emission intensity below the threshold value and reducing total nitrogen losses.

## Discussion

### Instruments to handle regional manure surpluses

Concentrations of livestock production are found in several regions of the world, mostly in areas with concentrations of humans, that is, near urban areas37, but there are also concentrations of livestock amidst extensive areas of cropland, that is, in large confined animal feeding operations, for example, in North and Latin America38. Despite all regulatory measures and economic and voluntary incentives, concentrations of livestock remain persistent hotspots of NH3 emissions and of nutrient losses to groundwater and surface waters. The benefits of a possible relocation of livestock from West European countries to Central European countries have been explored through simulation modelling only39,40. The Netherlands government is considering a buy-out of intensive livestock farms near nature conservation areas41. However, the number of animals considered in these options and plans is very modest compared to the south-to-north relocation of pig production in China35,36 and those of the current study (for example, Fig. 4).

China is a country where spatial relocation of large-scale livestock production could probably be successful because of its large livestock concentrations and its strong central governance. However, spatial planning of livestock production requires clear criteria, principles and procedures (Fig. 5). China started a top-down spatial planning of livestock production in 2014, and has relocated millions of pig farms from regions in the southeast with a dense network of watercourses to north China to reduce water pollution35,36,42. As a result, large spatial changes in livestock density have been observed between 2012 and 2017; livestock density was reduced in the south and increased in the north (Extended Data Fig. 10). However, the combination of water protection policies and the pig relocation policy has led to unexpected effects on pork prices which have been enhanced after the outbreak of African swine fever in 2018–2019 and the COVID-19 pandemic in 2020–202142. As a result, the central government has abandoned the pig relocation plan and given higher priority to the Vegetable and Basket Policy42. As a result, major new pig farms were recently built in high NH3 emission intensity regions and high population density regions (Extended Data Fig. 10).

### Minimizing trade-offs of livestock spatial planning

Spatial planning can be successful in protecting ecosystems when sound ecological and social-economic principles are applied together43. The Vegetable and Basket Policy44,45 may be seen as a one-sided spatial planning approach aimed at increasing vegetable and meat production and its fast delivery to urban consumers. The south-to-north relocation of pig production35,36,42 may be seen as a one-sided spatial planning approach aimed at decreasing surface water pollution. It is clear that these two spatial planning policies have large side-effects.

Serious trade-offs are also observed in the two spatial planning scenarios of our study. Spatial relocation of livestock to reduce the exposure of humans to high NH3 emissions is not effective in enhancing manure nitrogen recycling and reducing synthetic nitrogen fertilizer use in China. In contrast, spatial relocation of livestock aimed at enhancing manure nitrogen recycling and reducing synthetic nitrogen fertilizer use is not effective in reducing the exposure of humans to high NH3 emissions. These results are caused in part by the fact that intensive crop production systems in China already have a high NH3 emission intensity due to the liberal use of urea fertilizers (Fig. 3).

Pollution swapping through relocation of livestock production can be minimized through the implementation of strict emission mitigation measures in crop and livestock production systems. For example, urea-based nitrogen fertilizers have to be deeply placed/injected into the soil or have to be replaced by nitrate-based fertilizers to reduce NH3 emissions46,47,48. Similarly, low-protein animal feeding and low-emission manure storage and application to cropland may greatly decrease NH3 emissions associated with manure application. Implementing low-emission technology and balanced fertilization before any spatial planning action greatly reduces the risk of pollution swapping but also decreases the benefits (and costs) of these actions. An optimum balance needs to be found between the cost and benefits of low-emission technology and the costs and benefits of spatial planning actions, without compromising agricultural productivity. People in rural and peri-urban areas will ultimately benefit from cleaner air and water because of the substantial reductions in NH3 emission and related health costs, but we cannot exclude that food production costs will increase48.

### Sensitivity and implication of livestock spatial planning

Spatial planning of livestock production under NH3 emission control criteria was highly sensitive to the selection of the threshold value. The stricter the threshold value, the higher the number of livestock that need to be relocated within China. The threshold value was set at 31 kgNH3 ha−1—the criteria for achieving Sustainable Development Goal 2—for sustainable and resilient food production systems34. The criteria will be stricter in many European countries by 2030 because the National Emission Ceiling Directive has implemented regulations on total NH3 emissions per country49. China will have to relocate more animals when the threshold value decreases to below 31 kgNH3 ha−1, or will have to further cut domestic livestock production either through dietary changes to reduce meat consumption or through import of more livestock products from the global market. Spatial planning of livestock production under crop–livestock integration criteria was sensitive to the bioavailability of nitrogen from different organic resources. Fewer animals will need to be relocated if the bioavailability of nitrogen is considered because there is a lower surplus of bioavailable manure nitrogen than of total manure nitrogen (Fig. 3b and Extended Data Fig. 5b). Urbanization may also play a role in shaping the relocation of animals: urbanization leads to fewer people in the countryside and hence to fewer people exposed to cropland and livestock production related NH3 emissions50.

Other possible pollutants, such as heavy metals, veterinary antibiotics and greenhouse gas emissions, were not considered in this study but may have large impacts as well. This suggests that additional indicators and ecosystem threshold values have to be considered in spatial planning (Fig. 5). Current barriers for proper manure treatment and recycling to cropland need to be fully removed to be able to recouple crop and livestock production systems effectively51,52. The strive for carbon neutrality will also have implications for the optimal locations of production and consumption centres.

Overall, China may have to relocate 5–10 billion animals to tackle nitrogen pollution of air and water caused by concentrations of livestock production. This is an enormous challenge, with huge economic and societal impacts, but will yield improved nitrogen pollution control as final result. Policy-makers need to carefully select the criteria for spatial planning of livestock production because the ultimate effects of spatial planning depend on the criteria prioritized.

## Methods

The NUFER model was used to quantify the environmental impacts of different spatial distributions of crop and livestock production at county level in China13. NUFER includes data for 2,300 counties (out of 2,850 counties), which accounted for 94% of the total crop production and for 98% of the total livestock production in China in 20126,28,53. NUFER calculates the NH3 emissions and nitrogen losses to watercourses from the whole food production and consumption chain. These results allow us to quantify the relationship between livestock density and nitrogen losses to air and watercourses at county level. NUFER also provides information about fertilizer nitrogen use and total nitrogen flows per county; the nitrogen surplus includes all nitrogen losses to the environment and possible changes of soil nitrogen sequestration. For more information, see ref. 13.

### Geographic distribution of livestock production

The geographic distribution of livestock production was characterized by the livestock density per county and the ratio of manure nitrogen excretion and crop nitrogen uptake per county. County-level animal numbers were derived from ref. 13, which originally came from county, city and provincial level statistical data (census data). Numbers of animals in different livestock categories were converted to LU.

Livestock density was calculated as follows:

$${{{\mathrm{livestock}}}}\;{{{\mathrm{density}}}} = {{{\mathrm{LU}}}} \div {{{\mathrm{agricultural}}}}\;{{{\mathrm{land}}}}\;{{{\mathrm{area}}}}$$
(1)

where livestock density is the livestock density per county (LU ha−1), LU is the total number of LUs per county, and agricultural land area is the total agricultural land area per county (ha).

The geographic (dis)connection between crop and livestock production was characterized by combinations of manure nitrogen excretion, synthetic fertilizer nitrogen input and crop nitrogen uptake per county. Two ratios have been used: (1) ratio of manure nitrogen excretion to harvested crop nitrogen uptake; (2) ratio of synthetic fertilizer nitrogen input to harvested crop nitrogen uptake. Data were derived from ref. 13, and include county-level data for 1990 and 2012. For comparison, we used also data from United States for 1990 and 2012; these were reproduced from the Net Anthropogenic Nitrogen Inputs (NANI) toolbox21.

### Estimates of nitrogen requirements for crop production

Total nitrogen requirement for crop production per county was calculated as follows:

$${{{\mathrm{Ic}}}}_{{{{\mathrm{N}}}}\;{{{\mathrm{requirement}}}}} = \mathop {\sum }\limits_{{{i}} = 1}^{{n}} \left[ {\left( {{{{\mathrm{Oc}}}}_{{{{\mathrm{main}}}}\;{{{\mathrm{product}}}},{{i}}} + {{{\mathrm{Oc}}}}_{{{{\mathrm{straw}}}},{{i}}}} \right)} \right] \times {{{\mathrm{Uf}}}}({{i}}) + {{{\mathrm{Oc}}}}_{{{{\mathrm{managed}}}}\;{{{\mathrm{grass}}}}} \times {{{\mathrm{Uf}}}}$$
(2)

where IcN requirement is the nitrogen requirement per county for crop production under balanced fertilization criteria, that is, when nitrogen application is equal to crop nitrogen demand; Ocmain product,i and Ocstraw,i are the amounts of nitrogen in the main crop product and in straw at harvest per county, respectively (kgN); and Ocmanaged grass is the amount of nitrogen in harvested grass from managed grassland per county (kgN). These values were corrected by the uptake factors (Uf(i)) of each crop species under balanced fertilization (Extended Data Table 2).

We assumed that there is transport of manure within counties and no transport of manure between counties because of the high cost of long-distance transport. Further, we assumed substantial improvements of nutrient management in the whole food system by 2050, including a reduction in NH3 emissions from the manure management chain by 60%, a ban on the discharge of manure to watercourses and to landfills, and enhanced collection, sanitation and utilization of nitrogen in food waste and human excreta6.

The nitrogen balance was calculated per county as follows:

$$\begin{array}{ll}{\rm{I}}{{\rm{c}}_{{\rm{manure}}\ {\rm{N}}\ {\rm{surplus}}}} = & {\rm{I}}{{\rm{c}}_{{\rm{N}}\ {\rm{requirement}}}} - {\rm{I}}{{\rm{c}}_{{\rm{deposition}}}} - {\rm{I}}{{\rm{c}}_{{\rm{BNF}}}} - {\rm{I}}{{\rm{c}}_{{\rm{irrigation}}}}\\ & - {\rm{I}}{{\rm{c}}_{{\rm{straw}}\ {\rm{to}}\ {\rm{field}}}} - {\rm{I}}{{\rm{c}}_{{\rm{manure}}\ {\rm{to}}\ {\rm{field}}}} - {\rm{I}}{{\rm{c}}_{{\rm{food}}\ {\rm{by}}{\mbox{-}}{\rm{product}}}}\\ & - {\rm{I}}{{\rm{c}}_{{\rm{human}}\ {\rm{manure}}}} - {\rm{I}}{{\rm{c}}_{{\rm{soil}}\ {\rm{mineralization}}}}\end{array}$$
(3)

where Icmanure N surplus is the difference between nitrogen demand and nitrogen supply per county (excluding synthetic nitrogen fertilizer), defined as manure nitrogen surplus: a negative value of IcN surplus indicates an oversupply of nitrogen; a positive value of IcN surplus indicates a relative shortage of nitrogen. Then, all the negative values were added together, to evaluate the cumulative effects. Icdeposition, IcBNF and Icirrigation are the nitrogen inputs via atmospheric nitrogen deposition, biological N2 fixation (BNF) and irrigation water per county. Icstraw to field, Icmanure to field, Icfood by-product and Ichuman manure are nitrogen inputs via organic resources per county, that is, crop straw, animal manure, food and kitchen waste, and human excreta. Icsoil mineralization is the nitrogen supply from net soil mineralization.

### Scenarios for 2050

#### BAU scenario

This scenario follows the SSP2 storyline towards 205028. China’s total demand for agricultural products, including food, feed and biofuel, is projected to increase by 25% in per-capita calorie demand in 2050 relative to 2010 in the BAU scenario. Per-capita demand for animal-source calories is projected to increase almost twice as fast as for total calories compared with 2010. The total projected food demand and production in China are listed in Extended Data Table 2. The increase in food production by 2050 was evenly allocated to each county according to their shares in the total in 2012.

#### SNT scenario

This scenario specifically deals with pig production, which is by far the main livestock species in China. It follows the BAU scenario but uses a different geographic distribution of pig production and includes a ban on the discharge of manure to watercourses, following the Ten Water Protection Action. The Ministry of Agricultural and Rural Affairs categorized provinces into four zones in 2017 according to their vulnerability to surface water pollution and their potential for pig production: (1) potential development zones, (2) key development zones, (3) moderate development zones and (4) constrained development zones35. We assumed that the annual increase of pig production in potential development zones was two times higher than that in key development zones, and four times higher than that in the moderate development zones. There was no increase of pig production in the constrained development zones. These assumptions allowed us to calculate the total pig production for different pig production zones in 2050 (Supplementary Information).

#### ITI scenario

This scenario was developed to supplement the BAU scenario, but with substantial technological improvements in nutrient management throughout the whole food system. The technologies were specific for three sectors, that is, crop production, livestock manure management chain, and food waste and human excreta. The technologies aimed to reduce NH3 emissions and to increase the recycling of manure and human excreta, and food waste. Details about the key technologies and potential effects on nitrogen recycling and NH3 emission reduction are presented in the Supplementary Information.

#### SP scenarios

These scenarios were developed to supplement the ITI scenario, but with comprehensive spatial planning of livestock production (Fig. 5). Two contrasting variants were developed: (1) to reduce the exposure of most people to below the United Nations threshold for the NH3 emission intensity (SP-NH3); (2) to optimize the coupling of crop–livestock production in terms of nitrogen requirement and supply (SP-CLI). Detailed descriptions of the procedures for livestock spatial planning under these two scenarios are described in the Supplementary Information.

### Statistics and reproducibility

No statistical methods were used to predetermine sample size.

### Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.