Main

China has undergone remarkable social and economic development over the past two decades to become the world’s second largest economy. Over the same period, this successful development has led to a large increase in demand for food, especially for livestock products1,2. The import value of agricultural products has increased by 78% in constant US$ (ref. 3) while domestic agricultural value increased by 36% from 2010 to 2018. For soybean products in particular, the reliance on imports increased from 46% to 83%; for ruminant meat from 2% to 17% and for dairy products from 11% to 24% (ref. 2). The increasing demand also presents a great challenge to achieving the Sustainable Development Goals (SDGs)4 in China and worldwide as the agricultural sector is a key contributor to greenhouse gas (GHG) emissions (SDG 13), air and water pollution (SDGs 3 and 6) and biodiversity loss (SDG 15).

China’s domestic crop production increased by 44% between 2000 and 2018. Cropland expansion (4.9 Mha) (ref. 5) contributed 7% of production increase, with the remaining 93% from intensification. As a result, the use of nitrogen fertilizer in China today accounts for 32% of global fertilizer use. Similarly, livestock production also intensified, with increased reliance on concentrate feeds2. China’s agricultural production is now responsible for 13% of global GHG emissions2. Air and water pollution have reached 4.2-fold and 2.7-fold, respectively, of sustainability thresholds6,7 defined by fine particulate matter (PM2.5) and nitrogen discharge, largely due to the agriculture intensification. In addition, irrigation water use in China represents 13% of global water withdrawals, and the efficiency (48%) has substantial room for improvement compared with the levels in Europe and in North America (55–71%) (refs. 8,9).

Expanding imports are contributing to environmental pressure in exporting countries. Recent studies showed that displacement of resource use and environmental damage through international trade in the recent past represented a substantial share of the environmental impacts of domestic food production10,11,12. The contribution of China’s food demand to the challenge of achieving sustainable development of China’s trading partners has also been highlighted. For example, 43% of deforestation emissions due to soybean cultivation in Brazil can be attributed to China’s soybean imports in 201713. In addition, GHG emissions embodied in ruminant products exported to China accounted for 17% of total New Zealand livestock emissions in 201014.

China’s food demand is projected to keep increasing in the coming decades with further increase in the reliance on food and feed imports15. It is therefore necessary to assess the impacts of such growing demand on China’s domestic environment as well as the environment of its trading partners to inform sustainable development policies. However, current forward-looking assessments (Supplementary Methods 1) either focused on local impacts only without considering global market spillovers14,16,17, covered only a part of the agricultural sector (for example, bioenergy demand and afforestation18,19) or assessed only one or two environmental dimensions20,21,22. Assessments of future trade patterns mostly present trade with a world pool market23,39. Careful spatial planning is therefore necessary to exploit the environmental efficiency potentials to facilitate sustainable development. Increasing ruminant productivity is another promising way for reducing environmental pressure since China still has large productivity gaps compared with developed countries (Supplementary Fig. 18). We also find that assumptions about livestock feed efficiency change in the ROW have an important impact on the agricultural land and GHG emissions footprint of Chinese consumption (FEEF in Supplementary Fig. 15b). China could thus reduce its footprint also by promoting productivity improvement in its trading partners.

Sourcing agricultural imports sustainably

Imported environmental impacts vary considerably not only depending on the openness of trade but also depending on the country of origin. For example, milk-related GHG emissions intensity of the European Union is 0.9 kgCO2eq kg–1 product, whereas in New Zealand it is 1.4 kgCO2eq kg–1 (Supplementary Table 2), as shown also by other studies40. Our results show that increasing openness of trade (HD TRADE scenario) without accompanying measures can lead to both positive and negative impacts on the environment. Higher dairy imports from the European Union and bovine meat from the United States would lead to less GHG emissions relative to the BAU scenario; however, this scenario would also lead to increased beef imports from Latin American countries where land footprints are high (Supplementary Discussion 2). In addition, the past ban on soybean imports from the United States raised concerns about potential substitution with imports from Brazil and the related impacts on deforestation in the Amazon41. The environmental considerations need to be taken into account next to economic efficiency and political sensitivities when designing China’s trade policies to avoid unintended environmental consequences.

It is also recognized that even within an exporting country, supply chains may widely differ in their environmental impacts42. The environmental performance of specific supply chains is promoted, among others, by certification schemes such as ‘zero deforestation’ beef43 or ‘fairtrade’ labelling44. However, the effectiveness of these measures is limited if non-certified production still finds abundant markets. China, as one of the biggest importers, can play a key role in promoting adoption of environmentally friendly production systems in exporting countries by favouring imports of products from certified supply chains and, in general, by enforcing respect of ambitious environmental standards by its trading partners.

In summary, our results show that satisfying China’s food demand while achieving environmental sustainability domestically and in exporting regions is probably one of the biggest challenges of the coming decades. Carefully designed policies across the whole of China’s food system, including consumers, producers and international trade, are necessary to ensure that future demand can be satisfied without destroying the environment. Design of such policies will require models with high spatial resolution recognizing the heterogeneity of production conditions as well as environmental impacts in a country the size of China. Although the role of international trade is a buffer to shocks on the domestic market, in addition to satisfying part of food demand as a stable source, potential consequences of global short-term events will need to be considered. These important aspects would, however, go beyond the scope of our study.

Methods

This section presents the integrated modelling approach adopted, model developments for enhanced representation of China and model validation. Then the scenario design and the methodology used for sensitivity analysis are introduced. Virtual trade flows calculation is finally described.

Modelling approach

The quantitative analysis presented in our study relied on the GLOBIOM, a bottom-up partial equilibrium economic model designed to represent the key land-use sectors, including crops, livestock, forestry and bioenergy. GLOBIOM is extensively used for assessment of environmental impacts related to agriculture, such as sustainable water use27, GHG emissions29, land-use change and related biodiversity impacts45. The model is particularly suitable for forward-looking assessment of environmental impacts embodied in trade because of its bilateral trade representation28. Finally, the model is flexible enough to allow for a detailed representation of a region of interest, in this case China, while still kee** it embodied in the global modelling framework46.

The spatial resolution of the supply side relies on simulation units, which are aggregated from 5 to 30 arcmin pixels belonging to the same altitude, slope and soil class and the same country. For the purpose of this study, they were further aggregated to 2°. Commodity markets and international trade are represented for 37 economic regions in this study. Endogenous adjustments in market prices lead to balance among supply, demand and trade for each product and region. The market equilibrium is found through maximization of the sum of consumer and producer surpluses under constraints, such as land- and water-use balances. The model is solved with recursive dynamics in ten-year time steps. Main exogenous drivers of forward-looking scenarios in GLOBIOM are population and economic growth, technological change, dietary preferences and bioenergy demand. Main endogenous variables are market variables, including demand, supply, trade and prices, and environmental variables such as land and water use, GHG emissions and sinks, and nutrient balances.

Data on agricultural regional market variables, including demand and production, are for the base year harmonized with FAOSTAT2. The spatially explicit land-use allocation is initialized for 2000 with GLC200047. The spatially explicit productivity of crops, grasslands, forests and short-rotation tree plantations is estimated together with related environmental parameters (GHG budgets, nutrient and water balance) at the level of the simulation units. For crops, yields under different management systems are calculated with the biophysical Environmental Policy Integrated Climate (EPIC) model48,49. For forest parameters, GLOBIOM relies on the outputs of a dynamic forest management model, the Global Forest Model (G4M)50. Grassland productivity is obtained by combining results from EPIC and CENTURY biogeochemistry model25,51. Livestock production systems are parameterized with the global database developed by Herrero et al52. A detailed overview of data sources for the environmental indicators used in this study is presented in Supplementary Methods 4.

GLOBIOM represents international trade through net bilateral trade flows, which allow only one direction of trade flow between two regions. To simulate trade, GLOBIOM uses the Enke–Samuelson–Takayama–Judge spatial equilibrium approach, assuming homogeneous goods (imported and domestic products are the same)53. Thereby, GLOBIOM represents international trade through net bilateral trade flows, which allows only one direction of trade flow between two regions. In addition, the region will only import if its domestic price is greater than the price in the exporting country plus the cost of trade. In equilibrium, the difference in price between the importer and exporter equals the cost of trade. Compared with other trade assumptions (for example, Armington, trade can occur in both directions and gross trade is represented), this trade specification allows for new trade flow creation (no observation in the base year) in response to future price changes. As China is the largest importer for agricultural products and many countries strengthen cooperation in promoting trade with China, this approach is more appropriate for this study. Data on bilateral trade in the base year are from the BACI database54, and data on tariffs between different countries and commodities are from the MAcMap-HS6 database55. Additional information about the model can be found on www.globiom.org.

GLOBIOM–China

For this study, we modified the core GLOBIOM model to improve representation of China. To better capture the recent and future trends in Chinese agriculture, we included mechanisms mimicking relevant policies in place. One of the key drivers of land use in China is afforestation policies initiated in the 1990s. They already led to afforestation of 53 Mha at the cost of cropland, pasture and other land (unmanaged grass/shrubland, non-/sparse vegetation). Considering Chinese consumers’ preference for monogastric products and important structural changes in the sector, we calibrated the shift from smallholder to industrial systems for pig and poultry production. Fertilizer use efficiency development was calibrated to represent the ‘zero chemical fertilizer growth by 2020’ policy. We also enforced the self-sufficiency in three major cereal crops of 95% under the baseline scenario in line with the current trade policies. Supplementary Methods 2 and Supplementary Table 4 present the model improvements in further detail.

Model calibration and validation

A careful model calibration was performed for the period 2000–2020. FAOSTAT data and Chinese national statistical data until 2019/2020, as well as the OECD–FAO Agricultural Outlook projections for China until 202956, were then used to validate the model behaviour (Supplementary Figs. 2–7). The validation focused on the following key variables: crop yield, crop area, per capita food consumption, total demand, production and trade. The performance of the model for the very recent past has been quantitatively documented in Supplementary Methods 3. We also provide the interpretation of mismatches caused by recent pandemic outbreaks.

Bilateral trade calibration is of vital importance for this study. In GLOBIOM, future trade flows are determined by commodity prices and trade costs. Trade costs include tariffs, transport costs and a nonlinear trade expansion cost that reflects persistency in trade patterns. Tariffs and transport costs are kept the same as the base year. The trade expansion costs are used in GLOBIOM to represent the capacity constraints slowing down expansion of trade flows in the short term. They can be regarded as investments necessary to expand trading infrastructure. GLOBIOM allows for the appearance of new trade flows, which were not observed in the base year. Exponential function represents the trade cost (equation (1)) when trade flows are observed in the base year; for new trade flows, a quadratic trade cost function (equation (2)) is used:

$${\mathrm{Trade}}\,{\mathrm{cost}}_t = \frac{\varepsilon }{{1 + \varepsilon }} \times \frac{{{\mathrm{Tariff}} + {\mathrm{Transport}}\,{\mathrm{cost}}}}{{{\mathrm{Shipment}}_{t - 1}^{1/\varepsilon }}} \times {\mathrm{Shipment}}_t^{\frac{1}{\varepsilon } + 1}$$
(1)
$${\mathrm{Trade}}\,{\mathrm{cost}}_t = {\mathrm{Intercept}} \times {\mathrm{Shipment}}_t + 0.5 \times {\mathrm{slope}} \times {\mathrm{Shipment}}_t^2$$
(2)

Trade costs in period t are calculated with ε and slope reflecting the elasticity of trade costs to traded quantity in the respective equations. The intercept is equal to the tariff plus transport cost. The bilateral trade flows between China and other countries until 2020 were calibrated to match the recent Food and Agriculture Organization trade matrix statistics2 by manipulating the elasticities and slopes in the trade cost equations. The bilateral trade validation of major commodities is shown in Supplementary Fig. 7. Calibration work also benefited from feedback by seven country teams of the FABLE Consortium.

Scenario design

The aim of this study is to provide medium to long-term ex ante assessment of a global business-as-usual scenario aligned with current socioeconomic trends. We complemented this scenario with two variants with contrasted assumptions on future drivers and decompose those drivers to explore the range of results uncertainty. Development of such scenarios at the global level, with consistency across all sectors and regions, is a non-trivial task. Therefore, we decided to rely on the well-established framework of the SSPs, which provide a set of narratives and quantified drivers designed to analyse global trajectories of future development30. These pathways represent the backbone of the climate-related scenario analysis within the Intergovernmental Panel on Climate Change (IPCC)57 and have recently been used also for forward-looking biodiversity assessment in the context of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES)58. We acknowledge that some outbreaks (such as the US–China trade war in 2018 or COVID-19) may cause shocks and obstruct development of trade. However, in general these shocks are short-term disruptions59, and our scenarios can cover these large uncertainties.

A BAU scenario following SSP260 that mostly continues recent trends in consumption and technological developments was used as baseline in this study. The two alternative scenarios are (1) the RD scenario and (2) the HD scenario. The RD scenario follows the SSP3 assumption61 where the population in China increases faster, and growth in the GDP is slower, which leads to lower total food demand, in particular lower demand for livestock products compared with BAU. In this scenario, international trade becomes more restricted and fragmented, reflecting lower international cooperation. The HD scenario follows the SSP5 assumption62 and orients towards high economic growth but limited resource efficiency, leading to inclusive development but at the expense of the environment. International trade expands rapidly in globalized markets in this scenario. All these scenarios make the assumption of a diverse development trajectory of different regions following their economic growth in per capita change63, which are primary drivers for diet shifts and agricultural productivity changes.

As the food-demand patterns have been aggregated at the country level, income per capita drives changes in food diets64. Food prices are also important drivers for food-consumption pattern changes and are determined by demand-price elasticities of food products65. The crop yield trends are estimated on the basis of estimation of correlation between yield and scenario-specific GDP growth assumed in the SSPs66. In addition, re-allocation of cropland and shift of crop systems endogenously modelled also affect crop yield. For livestock systems, technical change is applied through exogenous assumption on feed conversion efficiencies estimated on the basis of historical trends for the BAU scenario and differentiated for the alternative scenarios on the basis of the average projected crop yield growth67,68. Trade assumption is one of the key differences among scenarios. Elasticity or slope of trade costs are varied depending on whether trade flow is observed in the base year. The trade liberalization or restrictiveness28 across scenarios reflecting infrastructure, non-tariff trade barriers and regional factor changes determines whether elasticities (slopes) are multiplied or divided by 10. More information on GLOBIOM trade specification can be found in Janssens et al.28. The values of key scenario drivers for China are provided in Supplementary Table 5, and a detailed description of alternative results can be found in Supplementary Discussion 1.

Considering that our assumptions of future changes (BAU, RD and HD scenarios) are based on a set of drivers (demographic and economic development, dietary preferences, agricultural productivity growth and international trade policies), we conducted a sensitivity analysis in which the impact of individual elements in the RD and HD scenarios is decomposed following the approach by Stehfest et al.69. The decomposition was implemented at the (1) global level, (2) ROW level and (3) China level only. This makes it possible to assess the individual impact of the preceding. Demographic development (POP) affects mainly future demand volumes adjusted by price effects. Economic development (GDP) affects income and associated food demand. Dietary preference (DIET) presents differences in dietary patterns between scenarios. Diet shifts and food waste are both included in this dimension. Crop productivity (YILD) is characterized by a different speed of technological changes. Livestock feed conversion efficiency (FEEF) is another key component on the supply side, determining future livestock productivity. Trade development (TRADE) represents the level of integration among global regions. The detailed results of the sensitivity analysis are presented in Supplementary Discussion 1 and Supplementary Figs. 14–16.

Calculating virtual trade flows in environmental impacts

Virtual trade flows refer to resources or pollution embodied in international trade. We focus our analysis about four environmental aspects (land, GHG, irrigation water and nitrogen) on seven major trading partners of China—Argentina, Australia, Brazil, Canada, New Zealand, the United States and the European Union—which account for more than 80% of the value of China’s agricultural imports (Supplementary Table 6). With respect to China trade flows, we also calculated the export effects (Supplementary Table 7); however, since imports dominate the overall trade pattern of China, we allocated the export impacts into the domestic production side. To calculate trade impact, we assume the same environmental intensity of products for domestic consumption and for export in a country. This is the assumption commonly used in many previous studies on virtual trade in water70, land71, GHG10 and nitrogen72. The environmental intensity in a resource for a specific product P in exporting region R and specific year T is defined as:

$$\begin{array}{rcl}{\mathrm{Virtual}\_{\mathrm{area}_{\mathrm{R,P,T}}}} &=& {\mathrm{BilateralT}_{\mathrm{R,P,T}}} \times {\mathrm{Land}}\_{\mathrm{intensity}_{\mathrm{R,P,T}}}\\ &=& {\mathrm{BilateralT}_{\mathrm{R,P,T}}} \times \frac{{{\mathrm{AREA}_{\mathrm{R,P,T}}}}}{{{\mathrm{PROD}_{\mathrm{R,P,T}}}}}\end{array}$$
(3)
$$\begin{array}{rcl}{\mathrm{Virtual}}\_{\mathrm{N}}_{\mathrm{R,P,T}} & =& {\mathrm{Bilateral}}\,{\mathrm{T}}_{\mathrm{R,P,T}} \times {\mathrm{N}}\_{\mathrm{intensity}}_{\mathrm{R,P,T}}\\ &=& {\mathrm{BilateralT}}_{\mathrm{R,P,T}} \times \frac{{\mathrm{N}_{\mathrm{input}_{\mathrm{R,P,T}}}}}{{\mathrm{PROD}_{\mathrm{R,P,T}}}}\end{array}$$
(4)
$$\begin{array}{rcl}{\mathrm{Virtual}}\_{\mathrm{water}}_{\mathrm{R,P,T}} &=& {\mathrm{Bilateral}}{\mathrm{T}}_{\mathrm{R,P,T}} \times {\mathrm{Water}}\_{\mathrm{intensity}}_{\mathrm{R,P,T}}\\ &=& {\mathrm{BilateralT}}_{\mathrm{R,P,T}} \times \frac{{\mathrm{Water}_{\mathrm{R,P,T}}}}{{\mathrm{PROD}_{\mathrm{R,P,T}}}}\end{array}$$
(5)
$$\begin{array}{rcl}{\mathrm{Virtual}}\_{\mathrm{Agri}\_{\mathrm{GHG}_{\mathrm{R,P,T}}}} &=& {\mathrm{BilateralT}_{\mathrm{{R,P,T}}}} \times {\mathrm{Agri}\_{\mathrm{GHG}\_{\mathrm{intensity}}_{\mathrm{R,P,T}}}}\\ &=& {\mathrm{BilateralT}_{\mathrm{R,P,T}}} \times \frac{{{\mathrm{Agri}\_{\mathrm{GHG}_{\mathrm{R,P,T}}}}}}{{{\mathrm{PROD}_{\mathrm{R,P,T}}}}}\end{array}$$
(6)

where BilateralTR,P,T is the net bilateral trade quantity (Mt) of product P exported to China from region R in year T; PRODR,P,T is total production (Mt) of product P of exporting region R in the year T; AREAR,P,T is total harvested area (Mha) of product P in exporting region R.

Virtual nitrogen (N) and water calculations follow the same logic (equations (4) and (5)), where NinputR,P,T represents synthetic fertilizer use (Mt), and WaterR,P,T represents irrigation water use (km3) for product P of exporting region R in year T. For nitrogen and irrigation water, we used crop-specific resource intensity informed by EPIC model calculations.

Equation (6) was used to calculate virtual agricultural-related GHG emissions (MtCO2eq yr−1). Fertilizer nitrous oxide (N2O) emissions and methane (CH4) from rice paddies were considered as direct crop-related GHG emissions. N2O was calculated on the basis of nitrogen fertilizer consumption and IPCC emission coefficients73 while rice CH4 was based on FAOSTAT average emission factors2. For livestock products, we used emissions intensity parameters for CH4 from enteric fermentation and for CH4 and N2O from manure management, manure dropped on pastures, rangelands and paddocks and from the global livestock production systems database52.

To calculate emissions from deforestation, we rely on a top-down indirect allocation approach74. We first determined forest losses in exporting regions on the basis of the G4M model calculations50 and then determined the deforestation attributable to cropland and pasture expansion on the basis of Curtis et al.75. Then we allocated the cropland deforestation emissions to individual crops on the basis of their contribution to the total cropland area expansion. The pasture-related deforestation was distributed among ruminant products on the basis of the pasture area necessary to cover the grass feed requirements of each livestock production system. Finally, we calculated the share of China’s virtual land import within the total area of each agricultural product. The deforestation emissions related to crop or pasture expansion are then calculated on the basis of the following equations:

$$\begin{array}{l}{{{\rm{Virtual}}\_{\rm{deforemission}}_{{\rm{R}},{\rm{T}}}}} = {{{\rm{Deforemis}}\_{\rm{crop}}_{{\rm{R}},{\rm{T}}}}} \times\\ \frac{{{\Delta}{{{\rm{Crop}}\_{\rm{area}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}}{{\mathop {\sum }\nolimits_{{{P}} = 1}^{{P}} {\Delta}{{{\rm{Crop}}\_{\rm{area}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}} \times \frac{{{{{\rm{Virtual}}\_{\rm{Crop}}\_{\rm{area}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}}{{{{{\rm{Crop}}\_{\rm{area}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}},\forall {\Delta}{{{\rm{Crop}}\_{\rm{area}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}} > 0\end{array}$$
(7)
$$\begin{array}{l}{{{\rm{Virtual}}\_{\rm{deforemission}}_{{\rm{R}},{\rm{T}}}}} = {{{\rm{Deforemis}}\_{\rm{live}}_{{\rm{R}},{\rm{T}}}}} \times\\ \frac{{{\Delta}{{{\rm{Pasture}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}}{{\mathop {\sum }\nolimits_{P = 1}^P {\Delta}{{{\rm{Pasture}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}} \times \frac{{{{{\rm{Virtual}}\,{\rm{Pasture}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}}{{{{{\rm{Pasture}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}}}}, \forall {\Delta}{{{\rm{Pasture}}_{{\rm{R}},{\rm{P}},{\rm{T}}}}} > 0\end{array}$$
(8)

where Deforemis_cropR,T and Deforemis_liveR,T are deforestation emissions (MtCO2eq yr−1) caused by cropland and pasture expansion in region R and year T, respectively; only the expanded area is accounted for in ΔCrop_areaR,P,T; \(\frac{{Virtual\_Crop\_area_{R,P,T}}}{{Crop\_area_{R,P,T}}}\) indicates the virtual crop area embodied in trade, which is presented in equation (3), divided by Crop_areaR,P,T to calculate the share of virtual land import. Similarly, deforestation caused by virtual pasture trade can be derived from equation (8).

Environmental impacts due to feed production are included in the virtual trade flows related to livestock products. For this purpose, we used the specific feed requirements of the regional livestock production specific feed requirements from Herrero et al52. We calculated the total feed use and the related domestic environmental impacts for different livestock products and allocated them proportionally on the basis of the quantities of the bilateral trade to the environmental impacts virtually imported by China. For feed crops embodied in the trade of livestock products, we considered only locally produced feed. This may lead to minor underestimation of the global impact of China’s imports, but this should remain minor as many livestock product exporters to China are not major feed crop importers.