Abstract
Agriculture is a major driver of global biodiversity loss1,2, accounts for one quarter of greenhouse gas emissions3, and is responsible for 70% of freshwater use4,5. How can land be used for agriculture in a way that minimises the impact on the world’s natural resources while maintaining current production levels? Here, we solved this more than 10 million dimensional optimisation problem and find that moving current croplands and pastures to optimal locations, while allowing then-abandoned areas to regenerate, could simultaneously decrease the current carbon, biodiversity and water footprint of global agriculture by up to 71%, 91% and 100%, respectively. This would offset current net CO2 emissions for half a century, massively alleviate pressure on global biodiversity and greatly reduce freshwater shortages. Whilst these achievements would require global coordination of agricultural policies, reductions of up to 59%, 78% and close to 100% are achievable by relocating production within national borders, with the greatest potential for carbon footprint reduction held by the world’s top three CO2 emitting countries.
Main Text
The conversion of almost half of the world’s ice-free land area6 to cropland and pasture has contributed to three of humanity’s most pressing environmental challenges7,8: (1) agriculture accounts for a quarter of anthropogenic greenhouse gas emissions3, largely from the release of carbon stored in vegetation and soils9,10; (2) agriculture is the predominant driver of habitat loss, the greatest threat to global biodiversity1,2; and (3) agriculture is responsible for 70% of global freshwater usage for irrigation, leading to shortages of potable water in many arid areas of the world4,5. A rising demand for animal products11 thwarts hopes that the potential for dietary shifts to decrease the environmental footprints of food production7,12,13,8 can be fully realised in the near future. Yield increases through more resource-efficient practices, technological advancements and genetically enhanced crop varieties are promising7,14,8, however a growing human population and increasing per-capita consumption15,16 threaten to offset the potential of these advancements without complementary measures.
Optimising the spatial distribution of production could help to minimise the impact of agriculture17. Empirical evidence shows that biodiversity and carbon stocks previously lost through land conversion can rapidly reach pre-disturbance levels if these lands are allowed to regenerate, often without active human intervention (Supplementary Information). Relocating croplands and pastures that are currently situated in areas with high potential biodiversity and carbon stocks, and subsequently allowing these areas to regenerate, may therefore lead to net carbon and biodiversity benefits. If, in addition, new agricultural areas were established where sufficient rainfall obviates the need for irrigation, the water footprint of global agriculture could be reduced significantly at the same time.
We used global maps of the current distribution of pasture and harvested areas of 43 major crops (Extended Data Table 1), which between them account for over 95% of global agricultural land (Methods), to assess the current carbon and biodiversity footprints of agriculture. The carbon impact in a specific area was calculated as the difference between local potential natural carbon stocks in vegetation and soils, and carbon stocks under the type of agricultural land use present there10 (Methods). Similarly, the local biodiversity impact of agriculture was estimated as the difference between local biodiversity under natural vegetation, and under cropland or pasture18 (Methods). Biodiversity is measured in terms of range rarity, in which local bird, mammal and amphibian species richness of weighted by the inverse of the species’ ranges. Range rarity has been advocated as a particularly meaningful biodiversity metric for conservation planning19,20.
By the same methods, we predicted potential carbon and biodiversity impacts in areas that are currently not cultivated but are suitable for agricultural use (Methods). We used estimates of agro-climatically attainable crop and grass production on potential agricultural areas that assume only rain-fed water supply21, so as to identify land use configurations that require no irrigation. We considered three different management levels, representing the range from traditional, subsistence-based organic farming systems to advanced, fully mechanised production that uses high-yielding crop varieties and optimum fertiliser and pesticide application21.
Using these realised and potential yield and impact estimates, we identified the global distribution of agricultural areas that provides the same total production of the 43 crops and grass as the current one, while minimising the total environmental footprint. On a 30 arc-minute (0.5°) grid, this requires solving a more than 106-dimensional linear optimisation problem (Methods). We estimated that for the optimal configuration of agricultural areas and advanced management farming, current carbon and biodiversity impacts of global agriculture could be simultaneously reduced by up to 71% and 91%, respectively (Fig. 1A). This would offset the current annual increase of atmospheric CO2 of 4.7 Pg C y−1(22) for 49 years, while drastically alleviating the pressure on terrestrial biodiversity. As per the data used, no irrigation is required to supplement rainfall water supply. The total worldwide area used for agriculture in this scenario is less than half of its current extent. The trade-off between reducing carbon and biodiversity impacts is minimal; optimising land use for each impact measure independently yields only marginally higher reduction potentials of 74% and 98%, respectively (Fig. 1A). Under traditional farming, simultaneous carbon and biodiversity impact reductions of up to 43% and 84%, respectively, are feasible (Fig. 1C). Whilst this confirms that increasing crop yields is important for reducing the environmental footprint of agriculture23,14,7,16,24,12,8, it demonstrates that a substantial impact reduction could be achieved by land reallocation alone.
Thus far, we have assumed that the entire area of each grid cell is available for agricultural use. How does the potential for reducing impact change if only a proportion of each grid cell can be cultivated, while the remainder is retained as natural ecosystem or used for other purposes? In this scenario, total impacts are necessarily higher, because less optimal areas, in which environmental impacts are higher in relation to yield, also need to be cultivated to meet a given production level. This is disproportionally the case for low intensity farming, which inherently requires more land. We found that when only half of the local land area can be used for agriculture, carbon and biodiversity impacts could be simultaneously reduced by 63% and 90%, respectively, under advanced management (Fig. 1A), but only by 30% and 80%, respectively, for traditional farming (Fig. 1C). Allocating as much land as possible in optimal areas therefore becomes more important the less advanced the farming system is.
Moving agricultural production, and thus labour and capital, across national borders poses numerous political and socio-economic challenges that will be difficult to resolve in the near future. We therefore repeated our analyses, allowing croplands and pastures to be relocated only within countries, while requiring current national production levels to remain unchanged (Methods). We estimated that if each country independently optimised its distribution of agricultural areas, the current global carbon and biodiversity impacts of agriculture could be simultaneously reduced by up to 59% and 78%, respectively (Fig. 1A). In this scenario, the vast majority of production can be relocated so that rainfall provides sufficient water supply; however, some countries produce crops for which national natural agro-climatic conditions are not suitable, and thus some irrigation continues to be needed (Methods). Fig. 3 lists the ten countries with the highest absolute carbon and biodiversity reduction potentials, showing that the world’s three largest CO2 emitters – China, India and the United States26 – are also the countries that can reduce their agricultural carbon footprint the most.
Agricultural areas optimally sited to minimise environmental impacts coincide only to a limited extent with their current distribution (Fig. 2). The world’s most produced crop, maize, for example, is currently planted predominantly in the United States and China, but would ideally be grown in parts of Sub-Saharan Africa (Extended Data Fig. 1). In the scenario of optimal within-country land reallocation, the optimal land use coincides with the current one on 30% of optimal areas, while 42% of optimal areas are located in regions already under some type of agricultural use, and 45% are located in either currently active or abandoned agricultural areas (Extended Data Fig. 2). This overlap is significantly lower in the scenario of across-country relocation (Extended Data Fig. 2). Whilst the expansion of agriculture into degraded areas has been advocated as a way to minimise future biodiversity and carbon losses7,14, our results suggest that potential biodiversity and carbon stocks on currently cultivated and abandoned agricultural areas are often so high that, in principle, their restoration would be preferable to the protection of natural habitat in the identified optimal growing areas.
For computational reasons, we did not explore the possibility of crop rotations and other diversification types, which can have benefits, e.g. for pest and disease suppression27. Our analyses also do not account for possible changes in yields as the result of climate change28. Both aspects may affect the precise location of optimal areas, and decrease or increase the reduction potentials identified here, however, we do not expect them to qualitatively change our overall conclusions.
Whilst our estimates of achievable impact reductions assume a fully optimised distribution of agricultural areas, we stress that even relocating only a small share of production would already generate a substantial portion of these benefits. 50% of the current total carbon impacts of individual crops are caused by areas that account for only 26±5% of the total production, while a mere 8±4% of production are responsible for 50% of biodiversity impacts (Extended Data Fig. 3). Prioritising the relocation of these areas, where the ratio of environmental impact to yield is largest, would have disproportionately large carbon and biodiversity benefits, and represents a particularly ‘low-hanging’ opportunity for countries to reduce impact.
Spatial reallocation of agricultural production has tremendous potential to reduce its environmental footprint, but the implementation of such changes would require careful management of the process. Relocating cultivated areas can only lead to a reduction of impact if abandoned areas with high potential biodiversity and carbon stocks are protected and their regeneration is ensured. This requires effective institutional, legal, and policy frameworks, and financial incentives for landowners29,30,31. Although biodiversity and carbon stocks often regenerate most effectively without human intervention32, active restoration efforts can be necessary to ensure the return of rare species in other cases33.
A range of policy mechanisms have proven effective in steering agricultural production to desirable areas34. Their implementation at the national and international level will be crucial for realising the environmental potential of moving agricultural areas, providing gains that are badly needed if we are to reverse the ongoing degradation of global climate, biodiversity and water under an ever increasing demand for food.
Methods
In the following, we define the mathematical optimisation problem whose solutions represent minimum impact configurations of agricultural land, and specify the datasets that were used to solve it. We use the following notation:
x : index of an arbitrary cell on a global 30 arc-minute (0.5°) grid
A(x) : physical area of grid cell x (ha)
Yi(x) : current yield of crop i in grid cell x (Mg C ha−1 y−1)
Hi(x) : current harvested area fraction of crop i in grid cell x
Ci(x) : carbon impact of crop i in grid cell x (Mg ha −1)
Bi(x) : biodiversity impact of crop i in grid cell x (local range rarity loss)
: agro-climatically attainable yield of crop i in grid cell x (Mg ha−1 y−1)
V (x) : fraction of area available for agriculture in grid cell x
On pastures, yield is assessed in terms of the annual production of forage per hectare. The current total annual production of crop i is given by and the current global carbon and biodiversity impacts of agriculture are given by respectively.
For each crop i and each grid cell x, we determined the harvested area fraction such that the total production of each crop i equals the current production Pi, while the environmental impact is minimised. Any solution must satisfy the equality constraints requiring the total production on new agricultural areas to be equal to the current one, and the inequality constraints which ensure that the local sum of agricultural lands is not larger than the locally available area.
Subject to these constraints, we can identify the configuration that minimises the total carbon or biodiversity impact by minimising the objective function respectively. More generally, we consider the linearly weighted objective function where λ ranges between 0 and 1, thus allowing us to minimise both impacts simultaneously and examine potential trade-offs.
The above framework is identical when examining the potential for impact reduction by means of relocating croplands within national borders rather than globally. In this case, the sum over x in the calculation of national production (Eq. (1)), in the optimisation constraints (Eqs. (2a)–(2b)) and in the objective function (Eq. (3)) is taken over grid cells that correspond to specific countries rather than the whole world, and the optimisation problem is solved independently for each country. Some countries produce small quantities of crops that, according to the data used here, would not grow anywhere within their borders under natural climatic conditions, i.e. these crops likely require irrigation or greenhouses cultivation. Our analysis shows that these crops account for a fraction of 0.12% of current global agricultural areas that can not be relocated within national borders to areas where rain-fed cultivation is possible. These crops were excluded from Eq. (3) for the respective countries; we added the environmental impacts associated with the current growing areas of these crops to the minimum national impacts found by Eq. (3).
Although all data required to compute the relevant variables, A(x), Yi(x), Hi(x), Ci(x), Bi(x), and V (x) (see below), are available at a 5 arc-minute (0.083°) grid resolution, for computational reasons, we upscaled the final data to a 30 arc-minute (0.5°) grid. For pasture and 43 crops, this implies a more than 10 million dimensional linear optimisation problem. We solved Eq. (3) using the dual-simplex algorithm in the function linprog of the Matlab R2018a Optimization Toolbox35.
Current and potential agricultural areas and yields: Hi(x), Yi(x),
We used global maps of harvested areas, Hi(x), and fresh weight yields, Yi(x), of 43 crops36, and a global map for pasture25 (Extended Data Table 1). These areas cover 95.2% of the combined area of pasture and harvested areas of 175 crops36, for which data is available. We used global maps of potential growing areas and agro-climatically attainable dry weight yields, , for baseline climate, rain-fed water supply and three different management levels for the same 43 crops and pasture grass21. Management levels represent the range from traditional, labour-intensive farming systems without synthetic chemicals, to advanced, market-oriented production that is fully mechanised, uses high-yielding crop varieties, and optimum applications of nutrients and pest, disease and weed control21. Potential yields were converted from dry weight to fresh weight using crop-specific conversion factors36. We are not aware of a global dataset of forage production on current pastures, and therefore used potential pasture grass yields for rain-fed water supply and intermediate input management as an estimate on these areas.
Carbon impact: Ci(x)
Following ref.10, the local carbon impact of agriculture, Ci(x), was estimated as the difference between potential natural vegetation and soil carbon stocks, and carbon stocks under agricultural land cover.
The change of carbon stocks in vegetation resulting from land conversion is given by the difference of carbon stored in potential natural vegetation10 and carbon stored in grass or crops, which was calculated as in ref.10, based on the data compiled by ref.36.
Due to the technical difficulties of acquiring empirical data across large spatial scales, spatially-explicit global estimates of soil organic carbon (SOC) dynamics under varying land use types are currently not available. We therefore chose a simple approach, consistent with average estimates across large spatial scales, rather than a complex spatially-explicit model for which, given the limited empirical data, robust predictions on and beyond currently cultivated areas would not be possible. Following ref.10, and supported by empirical meta-analyses37,38,39,40,41, we assumed a 25% reduction of potential natural SOC (see below) from the conversion to cropland. Meta-analyses of the change of SOC stocks when natural habitat is converted to pasture suggest, on average, no significant change39, a slight increase38,41 or slight decrease40. Here, we assumed no change in carbon stocks when natural habitat is converted to pasture. Absolute local SOC loss from the conversion of potential natural vegetation to cropland or pasture was estimated by applying the appropriate loss percentages to a global map of pre-agricultural SOC stocks9. The total local carbon impact of agriculture (Mg C ha−1) is thus given by where Cpotential vegetation(x), Cpotential SOC(x) and Ccrop(i) denote the carbon stocks (Mg C ha−1) of potential natural vegetation, potential natural SOC stocks, and carbon stocks of crop i, respectively, in the grid cell x, and where γ is equal to 0.25 or 0 if land is converted to cropland or pasture, respectively.
We did not consider greenhouse gas emissions from sources other other than from land use change. This includes nitrous emissions from fertilised soils and methane emissions from livestock and rice paddies42. In contrast to the one-off land use change emissions, these are ongoing emissions that are tied to production and incur continually. We do not consider available data sufficient to allow a robust extrapolation of these emission types to currently uncultivated land. We argue, though, that the magnitude of these emissions in a scenario of land reallocation in which total production is constant, is likely similar to that associated with the current distribution of agricultural areas. We also did not consider emissions associated with transport; however, these have been argued to be small compared to other food chain emissions43 and poorly correlated with the actual distance travelled by agricultural products44.
Biodiversity impact: Bi(x)
We assessed the local biodiversity impact of agriculture in terms of range rarity loss. Range rarity has been advocated as a metric for biodiversity that is more relevant to conservation planning than alternative measures, such as species richness45,46,20,19,47. Bi(x) is calculated as the difference between range rarity under natural vegetation and under agricultural land cover as follows: Using a similar approach to that of ref.18, we considered a bird, mammal or amphibian species to be potentially present in a cell of a 5 arc-minute grid if the species’ spatial extent of occurrcence48,49 overlays the grid cell, and if its habitat preferences48,49 include the local potential natural vegetation type50. Each species’ potential natural range (ha) is then given by the total area of all grid cells identified as containing the species. Next, potential natural range rarity of each grid cell was obtained as the sum of the inverse ranges of all species present in the grid cell under potential natural vegetation. Finally, global maps of range rarity loss resulting from the conversion of natural vegetation to cropland or pasture were derived by subtracting, in each grid cell, the sum of the inverse ranges of potentially present species whose habitat preferences also include cropland or pasture, respectively, from the potential natural range rarity. As with Ci(x) (see above), this approach allowed us to estimate biodiversity impact for both currently cultivated and uncultivated areas.
Land available for agriculture: V(x)
We assumed that the maximum area available for agriculture in a grid cell is given by the proportion not occupied by any crop other than the 43 considered here36, or by water bodies, infrastructure or settlements21. Areas where soil and terrain-slope conditions are not suitable for agriculture are already excluded in the potential yield data21.
As specified in the main text, we also examined the scenario in which only a certain fraction of this maximum available area is available as potential agricultural land.
Acknowledgments
We thank Günther Fischer and Paul Donald for their advice and comments during the preparation of this manuscript. This work benefited from conversations with Fiona Sanderson and Catherine Tayleur (Royal Society for the Protection of Birds), Paul Donald (BirdLife International), Sharon Brooks (UN Environment World Conservation Monitoring Centre), and David Coomes, América P. Dur#x00E1;n, Philip Martin and Ben Phalan (University of Cambridge) during a separate research project that was supported by a grant from the Cambridge Conservation Initiative Collaborative Fund (CCI-06-16-008).
References
- [1].↵
- [2].↵
- [3].↵
- [4].↵
- [5].↵
- [6].↵
- [7].↵
- [8].↵
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].↵
- [43].↵
- [44].↵
- [45].↵
- [46].↵
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
- [60].↵
- [61].↵
- [62].↵
- [63].↵
- [64].↵
- [65].↵
- [66].↵
- [67].↵
- [68].↵
- [69].↵
- [70].↵
- [71].↵
- [72].↵
- [73].↵
- [74].↵