Exploring the biophysical option space for feeding the world without deforestation

Safeguarding the world's remaining forests is a high-priority goal. We assess the biophysical option space for feeding the world in 2050 in a hypothetical zero-deforestation world. We systematically combine realistic assumptions on future yields, agricultural areas, livestock feed and human diets. For each scenario, we determine whether the supply of crop products meets the demand and whether the grazing intensity stays within plausible limits. We find that many options exist to meet the global food supply in 2050 without deforestation, even at low crop-yield levels. Within the option space, individual scenarios differ greatly in terms of biomass harvest, cropland demand and grazing intensity, depending primarily on the quantitative and qualitative aspects of human diets. Grazing constraints strongly limit the option space. Without the option to encroach into natural or semi-natural land, trade volumes will rise in scenarios with globally converging diets, thereby decreasing the food self-sufficiency of many developing regions.


Modelling framework
BioBaM is a biophysical modelling framework that assesses the feasibility of scenarios, based on a unique combination of variants of the following parameters: a) human diets, b) origin of livestock products, c) composition of livestock feedstuff composition, d) yields, e) cropland expansion. All these variants are based on published forecasts or, in the case of human diets, recommendation of the USDA and HHS 4 . A scenario is each combination of these variants. All 500 scenarios are tested if supply covers demand and is, in negative cases, classified as cropland limited, grazing land limited or cropland and grazing land limited. Cases where supply of biomass is within 95% to 105% of demand are labelled as 'probably feasible', and as 'feasible' if it is larger than 105%. The model is applied to 11 world regions, and regional deficits in crops and roughage is assumed to be compensated by interregional trade. A schematic representation of the modelling framework is shown in Supplementary Figure 1.

Parameters and description of variants 2.1. Household food demand
We model per-capita food demand as food demand by households, i.e. including food wastes emerging during consumption at the household level. Supplementary Table 1 shows household food demand for the year 2000, which was used as baseline for the definition of the different variants. In 2050, we consider one business as usual variant (BAU) with household food demand differing between regions and four variants in which household food demand converges between regions (RICH, MEAT, VEGETARIAN and VEGAN). All variants for household food demand in 2050 are shown in Supplementary Table 2.
In the variant business as usual (BAU), per capita household food demand in 2050 is derived from FAO projections 3 and adjusted/completed according to projections given by IAASTD 6 in cases where FAO data was lacking or incomplete. Projections for fish consumption were taken from Wijkstrom 7 , however we did not consider potential crop demand by aquaculture. In the variant RICH, per capita food demand of all 11 world regions converge until 2050 to per capita household food demand of North America in 2000 according to FAOSTAT (2014). Currently, North America has the highest overall food demand and the highest demand for livestock products of all regions (Supplementary Table 1).
The variants MEAT, VEGETARIAN and VEGAN are based on recommended diets according to USDA and HHS 8 , and illustrate a diet including meat (MEAT), an ovo-lacto vegetarian diet (VEGETARIAN), and a vegan diet excluding all livestock products (VEGAN). MEAT, VEGETARIAN, and VEGAN start from an average per capita intake of 2,205 kcal cap -1 day -1 , based on the recommended total calorific intake for different age cohorts according to USDA and HHS 8 , in combination with population age cohorts in 2050 according to UN 9 . In order to ensure a sufficient food supply for all persons and in all years, taking inequality in food access and yearly variations in food supply account, we add 10% to the resulting food intake, resulting in an average per capita intake of 2,425 kcal cap -1 day -1 . Based on this calorific intake, we derive intakes for food product groups from USDA and HHS 8 and regroup food categories according to those applied in this study. Household food demand is calculated by adding an identical share of household food waste for all regions, amounting to global average household food waste in 2000, based on values given by Gustavsson et al. 4 . While this assumes that the share of household food waste remains equal on a global level, it implies future reductions of food waste in developed countries but an increase of food waste in developing countries, consistent with a variant of converging diets.

Origin of livestock products in human diets
For each household food demand variant (see above), we additionally modulate the share of livestock products derived from monogastric and ruminant livestock, respectively (Supplementary Table 4).
In the variant BAU, the share of ruminant and monogastric products in each region is in accordance with the FAO main projections for 2050 3 . In the variant MONOGASTRIC, ruminant livestock products (meat and milk) are substituted completely by pigs, poultry and eggs, while the variant RUMINANT assumes that all livestock products from monogastrics (pigs, poultry and eggs) are substituted by ruminant meat and milk. As the variant VEGETARIAN does not allow a substitution of milk by meat from monogastrics in the variant MONOGASTRIC, milk is substituted by eggs in this case. All substitutions refer to food demand in terms of kcal cap -1 day -1 .

Livestock feedstuff composition
In terms of input-output ratios between feed intake and output of livestock products, we differentiate two variants: ROUGH, and GRAIN. For each of the livestock product categories covered in our study, a variant defines the quantity of a) feed from cropland (e.g. concentrates) and b) roughage feed required to produce one unit of livestock product, measured in tons of feed per ton of output (both in dry matter).
The projections for feed intake and product output for different world regions and livestock systems in 2050 are based on Bouwman et al. 10 . The more disaggregated world regions in Bouwman et al. (2005) are aggregated into the 11 world regions used in this study. Additionally, we aggregate the items grass, residues and fodder, and scavenging into the roughage feed category and project the values to 2050, assuming a continuation of the linear development from 1995 to 2030 according to Bouwman et al. 10 . We then modulate livestock feed composition in the variants ROUGH and GRAIN as follows.
In the variant GRAIN, we assume an acceleration of the trend to an increased share of concentrates in livestock feedstuff composition, operationalized as an increase of crop feed demand by 30% in all regions and an accordingly lower roughage demand, assuming a substitution weight of 0.5, i.e. 2 units of roughage are replaced by 1 unit of crops (measured in dry matter). To avoid extreme and rather unrealistic assumptions, we additionally set upper and lower boundary for crop and roughage intake see 11 for details . The variant ROUGH is a counter-trend variant, assuming that crops in ruminant diets are reduced by 50% in all regions compared to the trend scenario based on Bouwman et al. 10 and accordingly replaced by roughage, i.e. grazing (substitution weight as above). As Bouwman et al. (2005) assume increasing feed conversion ratios, the feed demand per kg of output in 2050 is lower than in the year 2000.

Grazing intensity
Due to the zero deforestation assumption, cropland is allowed to expand only into grazing land characterized by a productivity of >200gC m2 -1 yr -1 (Supplementary Figure 2). As a consequence, grazing land is assumed to shrink accordingly. In order to cover an eventual demand for roughage from grazing land by ruminants, the grazing intensity on the remaining grazing land is allowed to increase up to a global average grazing intensity of 33% (defined as total grazed or harvested roughage per total aboveground productivity, Supplementary Table 7).

Yields
Variants for crop yields per area for each region are shown in Supplementary In global average, there are only small differences between the variants FAO and YIELDGAP, which is reflected by the quite similar global biophysical option space shown in Figure 2 of main text. However, there are pronounced differences in terms of crop yields in 2050 between regions (Supplementary Table 5): While especially in Africa, Central Asia and Russia, and Eastern Europe, the variant YIELDGAP assumes considerably higher crop yields than the variant FAO, lower future increases in crop yields are assumed especially for Western Europe and North America, where yield gaps are already closed to a large degree according to the analysis by Mueller et al. 13 .
Crop yields of the variant ORGANIC are derived as follows: In a first step, we assume that crop yields per harvest event (e.g. cereals or oil crops) in organic agriculture are 20% lower compared to conventional agriculture, as supported by two recent meta-studies on crop yields in organic agriculture 14,15 . In addition, it is assumed that 25% of the crop rotation in organic agriculture is required for legumes such as alfalfa 16 in order to replenish soil nitrogen, which we consider as an additional crop yield deduction. Considering the whole crop rotation including legumes, which cannot be used as food crops, this results in an assumed yield reduction by 40% of organic compared to industrial agriculture. However, it is important to consider that such a comparison refers only to organic compared to highly intensified agriculture with modern inputs such as high-yielding seed varieties, synthetic fertilizers and pesticides. It has been shown that in regions with high shares of traditional agriculture, crop yields can be increased without the use of industrial technologies (Pretty et al., 2011). Therefore, the yield deduction of 40% is only accounted for the share of highly intensive industrial agriculture in each world region, resulting in yield deductions of between 4% for Sub-Saharan Africa and 40% in North America compared to the variant FAO (Supplementary Table 5).

Cropland expansion
We consider five variants for the expansion of croplands, +0%, +11%, +22%, +40% and +70%, with the percentage denoting the global expansion of cropland until 2050 compared to global cropland areas in 2000. Region specific cropland areas for all variants are shown in Supplementary Table 6.
The variant +0% shows that croplands of the year 2000 still can support some scenarios in 2050. The variant +11% is in accordance with the FAO projections from Alexandratos and Bruinsma (2012) and assumes a global expansion of croplands by 11%. Global studies of land suitable for cropland (IIASA and FAO, 2000;Ramankutty et al., 2002) suggest that cropland potentials are considerably larger than those assumed by the FAO. In the variant +22%, we thus double the global cropland expansion until 2050 assumed by the FAO.
The variants +40% and +70% are more extreme and assume that half (+40%) or all (+70%) grazing land of the highest productivity (grazing class I in Erb et al., 2007) will be converted into cropland until 2050. Additional land that would be necessary for infrastructure is considered in all variants, based on a extrapolation of current trends 17 . In the NAWA region, variants +40% and +70% were not possible and set equal to variant +11% (Supplementary Table 6). The zero-cropland expansion variant (0%) is identical to the year 2000.