This is the accepted version of the article: Folberth, Christian; Khabarov, Nikolay; Balkovic, Juraj; [et al.]. «The global cropland-sparing potential of high-yield farming». Nature sustainability, Vol. 3, Issue 4 (April 2020), p. 281–289. DOI 10.1038/s41893-020-0505-x This version is avaible at https://ddd.uab.cat/record/224224 under the terms of the license 1 The global cropland sparing potential of high-yield farming 2 Christian Folberth, Nikolay Khabarov, Juraj Balkovič, Rastislav Skalský, Piero Visconti, Philippe Ciais, Ivan 3 A. Janssens, Josep Peñuelas, Michael Obersteiner 4 Summary 5 The global expansion of cropland exerts substantial pressure on natural ecosystems and is expected to 6 continue with population growth and affluent demand. Yet, earlier studies indicated that crop 7 production could be more than doubled if attainable crop yields were achieved on present cropland. 8 Here we show based on crop modelling that closing current yield gaps by spatially optimizing fertilizer 9 inputs and allocation of 16 major crops across global cropland would allow to reduce the cropland area 10 required to maintain present production volumes by nearly 50% of its current extent. Enforcing a 11 scenario abandoning cropland in biodiversity hotspots and uniformly releasing 20% of cropland area for 12 other landscape elements, still enabled reducing the cropland requirement by almost 40%. As a co- 13 benefit, greenhouse gas emissions from fertilizer and paddy rice, as well as irrigation water 14 requirements are likely to decrease with reduced area of cultivated land, while global fertilizer input 15 requirements remain unchanged. Spared cropland would provide space for substantial carbon 16 sequestration in restored natural vegetation. Only targeted sparing of biodiversity hotspots supports 17 species with small-range habitats, while biodiversity would hardly profit from a maximum land sparing 18 approach. 19 Introduction 20 Globally, agricultural activity and the continuous expansion of croplands impose wide-ranging 21 environmental burdens on natural ecosystems. Intensively managed cropland is characterized by 22 excessive and imbalanced applications of N and P, whereas low-input agricultural systems result in 23 nutrient-poor soils and low yields1,2. Globally, freshwater use in agricultural irrigation consumes about 24 70% of total water withdrawals3, and cropland farming contributes about 5% of global GHG emissions, 25 mainly through emissions of paddy rice methane (CH4) and soil nitrous oxide (N2O) from added mineral 26 N fertilizer and manure4. Biodiversity loss is challenging to quantify, but estimated to exceed safe 27 boundaries, primarily due to habitat loss5. Most recently, the land sparing debate6–9 has gained new 28 momentum from the Half Earth project10 that aims to return half the area of land under anthropogenic 29 management to natural land cover to restrict biodiversity losses and abate other externalities of 30 anthropogenic land use6. The need for this type of strategy is even more urgent, given the increasing 31 global demand for agricultural products11,12. Yet, biophysical benchmarks for ambitious land sparing 32 targets and associated externalities remain virtually unknown. 33 Earlier studies have suggested that cropland will likely further expand in the future due to population 34 growth and climate change13, while effective cropland sparing would need to involve measures such as 35 dietary change to reduce crop demand14–16. In contrast, global nutrient input intensification, crop 36 switching, and expansion of irrigated land may increase global crop production volumes by up to 150% 37 for major crops17–21 depending on whether and how these strategies are combined. Intensification has 38 also been identified in conceptual and semi-quantitative studies as a promising strategy for the 39 abatement of land conversion, expansion of natural land cover7,8,22, and reduction of environmental 40 impacts, depending on management specifics23. However, while average yields for major crops have 1 41 been increasing globally during the past decades, they have stagnated or decreased in various parts of 42 the world and the present pace in yield gains is considered insufficient to meet future crop demand24. 43 Persisting global yield gaps in major crops have been attributed foremost to nutrient deficits and to a 44 lesser extent to insufficient water supply19. 45 Estimating global cropland requirement 46 In this study, we quantified the potential of land sparing through intensification of nutrient inputs to 47 meet plant requirements and optimal spatial allocation of 16 major crops to estimate a lower boundary 48 of cropland requirement for meeting present crop demands (Figure 1). We used the established global 49 gridded crop model EPIC-IIASA17 to estimate non-nutrient limited crop yields, with and without sufficient 50 irrigation water supply, depending on land use information to avoid expansion of irrigated land. EPIC- 51 IIASA combines the process-based agronomic model Environmental Policy Integrated Climate25,26 (EPIC) 52 with a global data infrastructure gridded at 5’ x 5’ resolution. The 5 arcmin grid cells with identical soil 53 texture and topography classes and located within the same 30’ x 30’ climate grid and administrative 54 region were aggregated to simulations units. The resulting 120000 simulation units thus vary in size from 55 5’ x 5’ to 30’ x 30’ or total corresponding surface areas from 69 to 2500 km2 near the equator depending 56 on input data heterogeneity (Supplementary Figure S19; Supplementary Text S2). Maps of present 57 cropland were aggregated from 5’ x 5’ source data to the same spatial scale of simulation units to 58 provide consistent input data on area and crop yields for the cropland allocation model. 59 Crop distributions were spatially allocated using a linear optimization algorithm under three simple 60 criteria that comprised minimizing the extent of current global cropland; maintaining 2011-2015 global 61 production volumes for each crop; and avoiding novel expansion of cropland locally. This was done (I) 62 allowing the full use of the current cropland in each simulation unit to create a global “maximum land 63 sparing” (MLS) scenario or (II) with a complete release of annual cropland in biodiversity hotspots and a 64 forced release of at least 20% of cropland area in each simulation unit to create a “targeted land 65 sparing” (TLS) scenario. The first serves for providing a benchmark of what extent of land sparing is 66 technically feasible given present agricultural technologies. The latter provides a benchmark for a global 67 scenario focused on habitat restoration for threatened species in hotspots combined with the 68 establishment of uniformly distributed landscape compartments as wildlife habitats27 or buffers for 69 adverse impacts of high-input agriculture28. Two supplementary scenarios serve for assessing how 70 constraining crop distributions to their present growing regions (scenario MLSncs) or allowing crops to 71 cover a maximum of 34% cropland in a simulation unit – indirectly increasing crop diversity locally - 72 (scenario MLSwcd) affect results in the MLS scenario. As such, these scenarios are hypothetical, leaving 73 aside policy- and socio-economic implications, but they can nonetheless inform decision-makers about 74 the bio-physical feasibility of ambitious land sparing targets. While the focus of our analysis is on 75 cropland sparing potential, we also quantified, based on model results directly or auxiliary datasets, 76 changes in requirements for N and P fertilizer and irrigation water; selected GHG emissions; carbon (C) 77 storage in resultant, expanded areas of natural vegetation; and potential increase in natural habitats for 78 wildlife. Further details are provided in the Methods section. 2 79 80 Figure 1: Schematic of the study design. Attainable crop yields (A) from the EPIC-IIASA global gridded crop model (or statistically 81 derived yield datasets) are combined with present cropland data for these crops (B) from SPAM2005 (or other suitable land use 82 datasets) into a linear optimization model (C). This model has the objective to minimize cropland extent via cropland allocation 83 based on land use efficiency while maintaining present production volumes (D) for two main scenarios (I) with the only 84 constraint of presently available cropland (MLS scenario) or (II) with imposing a release of cropland in biodiversity hotspots and 85 a uniform global release of 20% cropland (TLS scenario). Further constraints (E) are introduced for two supplementary scenarios 86 (see Methods). The optimization results in crop-specific land use datasets (F), which are aggregated to total remaining cropland 87 including the crops not considered in the optimization (G). Externalities (H) are quantified based on outputs of the crop model 88 itself (nutrient input and irrigation water requirement) or based on external data and models (carbon sequestration potential, 89 change in area of habitat, and greenhouse gas emissions). Crop model simulations and cropland allocation were performed at 90 the level of globally 120000 simulation units aggregated from 5’ x 5’ pixels (about 8.3 km x 8.3 km near the equator) to a 91 maximum size of 30’ x 30’ (about 50 km x 50 km near the equator) based on physical heterogeneity and administrative borders. 92 The cropland area in each 5’ x 5’ pixel was subsequently scaled according to the relative change in cropland extent in the 93 overlying simulation unit (see Supplementary Figure S19) for the estimation of externalities and visualization. The central 94 cropland allocation scheme is shown for exemplary simulation units in Supplementary Figure S18. 95 Global cropland sparing potential and spatial patterns 96 Intensification and optimal crop reallocation under the MLS scenario decreased the cropland 97 requirement to nearly 50% of the baseline for all crops and to 46% for the 16 selected crops (Figure 2). 98 The greatest sparing potential was for typical smallholder crops, such as sorghum and pulses, with >80% 99 of land released (Supplementary Table S1). Lower land gains (<50%) were estimated on the other hand 100 for crops for which production tends to be highly intensified, such as maize, rice, soybean, wheat, and 101 sugar crops. The TLS scenario also reduced cropland area to remaining 62% of the baseline, indicating 102 that radical reductions in cropland area are not restricted to a narrow set of solutions, and high yields 103 may be sustained across large regions for most crops. Results were highly comparable for a wider range 104 of land use and attainable crop yield datasets, showing that our estimates are robust within the limits of 105 available data (Supplementary Text S1 and S2). 106 3 107 108 Figure 2: Global extent of annual cropland in the reference period 2011-2015 (column one), and estimated area of cropland of 109 16 major crops optimized for maximum sparing potential (column two) and sparing of at least 20% of cropland in each 110 simulation unit and completely abandoning biodiversity hotspots (column three). Crops not considered in the optimization are 111 aggregated into three groups at the base of each bar. Percent values refer to area of total annual cropland (upper) and 112 simulated crops (lower, in parentheses). Globally, annual crops plus sugarcane extended to about 1100 Mha of cropland during 113 the reference period29, of which about 950 Mha were planted with the crops considered in the optimization (major cereals, 114 grains, pulses, and sugar). The remaining 150 Mha encompassed “Other row crops”, “Fruits and Vegetables”, and “Non- 115 food/feed crops” shown at the base of each bar (Supplementary Table S1). 116 Contiguous regions of cropland release in the MLS scenario are primarily located in agro-climatically 117 unfavourable regions, such as the Western USA, Central Asia, and Sahel, but also in productive regions 118 such as large parts of South Asia and in southern Russia (Figure 3b). Despite local concentrations of 119 cropland in most productive areas, patterns of total fresh matter production volumes per continent 120 remained comparable to the baseline with substantial and moderate gains in Africa and Asia at the cost 121 of Europe and especially America (Supplementary Figure S4). The TLS scenario resulted in a wider 122 distribution of cropland (Figure 3c), which is mostly driven by implicit cropland release in this scenario 123 (Supplementary Figure S5). About 20% global annual cropland were released in biodiversity hotspots 124 and globally uniform a minimum of 20% in the remainder of the cropland area (corresponding to 17% of 125 global annual cropland). This left only a minor fraction of areas released subject to land use efficiency 126 gains. These were again mostly located in agro-climatically adverse regions such as desert borders. 127 128 129 130 4 131 132 Figure 3: Proportion of each 5’ x 5’ pixel covered by cropland cultivated c. 2005 (SPAM dataset) (a) and fraction released after 133 optimization of cropland requirement for maximum land sparing (b) and targeted land sparing with complete release of 134 cropland in biodiversity hotspots and uniformly ≥20% of cropland (see Supplementary Figure S16) (c). Data in (b) and (c) 135 correspond to bars two and three in Figure 2. 136 Drawbacks of and barriers to cropland sparing and concentration 137 The release of cropland over large contiguous regions in both scenarios may entail substantial socio- 138 economic implications with respect to livelihoods, as shown also in recent research on global 139 conservation targets30, and may affect regional food self-sufficiency. Yet, the fact that patterns of 140 cropland release are largely contrasting among the two scenarios indicates that a mixed approach, 141 including the sparing of cropland in biodiversity hotspots only to the degree necessary for maintaining 5 142 wildlife habitats, could be implemented to balance socio-economic trade-offs with land sparing benefits 143 among regions. Comprehensive global research on social acceptance for land sparing is lacking and 144 certainly context-dependent. Conceptual studies suggest a range of policy measures from financial 145 compensation for abandoned cropland to payments for restored vegetation management and further 146 knowledge transfer and infrastructure for improved crop management to steer policy implementations 147 of intensification for cropland sparing8. Notwithstanding, the reconciliation of global targets with local 148 and regional stakeholder demands will require holistic approaches bridging these scales, which likely 149 poses the greatest challenge in achieving effective global land sparing31,32. And any further 150 concentration of crop production will increase the already extensive reliance of large parts of the world 151 on food imports, amplifying the requirement for resilient global trade systems33. 152 Spatial shifts in crop cultivation areas are a constant process34 and have been subject to disruptive 153 regime shifts for specific crops and regions in the past35. Both do not necessarily follow patterns of 154 domestic demand but serve often for income generation and diversification36. However, the adoption of 155 new crops or farming practices in general requires more information and policy interventions in regions 156 in which they are not practiced so far. Analysing the spatial occurrence of crops in both scenarios herein 157 reveals that <20% of resulting cropland area are occupied by crops in simulation units in which they are 158 presently not grown, and <1% in major Koeppen-Geiger climate regions and countries in which the 159 respective crops are presently not cultivated (not shown). Constraining the cropland allocation model to 160 only assign crops in the MLS scenario to simulation units in which they are presently cultivated while 161 allowing their local acreage to change (supplementary scenario MLSncs) results in 1% lower land sparing 162 potential (Supplementary Figure S6). This indicates that the free shifting of crops is not a key mechanism 163 behind our findings and that crops are already cultivated in regions in which they are or can be most 164 productive. Yet, areas of crops presently cultivated for cultural and historic reasons may be given up in 165 the model. In this context, it needs to be stressed that our study aims to provide information on the 166 cropland that is essentially required to meet present demand and should not suggest to abandon 167 agriculture in places in which it provides important local cultural and social services. 168 Furthermore, optimizing cropland distribution based on land use efficiency may result in wide-spread 169 monocropping systems with higher vulnerability to biotic and abiotic stressors, high requirement for 170 pest control agents, and little provision of on-farm biodiversity. To address the impact of enforcing crop 171 diversity on cropland sparing potential, we evaluated a supplementary scenario allowing only up to 34% 172 of each simulation unit to be covered by a specific crop in the MLS scenario (supplementary scenario 173 MLSwcd). This reduces the cropland sparing potential by 5% relative to the present extent 174 (Supplementary Figure S6) while resulting in the co-occurrence of 2-3 crops in most simulation units 175 (Supplementary Figure S8). The concurrence of several crops translates into the feasibility of inter- 176 annual crop rotations, which are a key measure for integrated crop protection37. Due to the capped area 177 share of single crops, simulation units with only one or two crops attributed can similarly implement 178 rotating inter-annual fallows. This supplementary scenario also results in higher crop diversity at the 179 continental scale, especially in Europe, compared to the other land sparing scenarios (c.f. 180 Supplementary Figure S7 and Supplementary Figure S4). 181 Associated changes in externalities 182 Reductions in cropland area, combined with optimal N and P fertilization, may reduce or at least not 183 exacerbate major agricultural input requirements and externalities globally (Figure 4). We found that 6 184 total N and P application would increase by only 6% in the MLS scenario, and decrease by 1-4% in the 185 TLS scenario. This includes the presently inevitable prevalence of substantial nutrient losses such as 186 leaching and erosion. Our results confirm that current excessive and imbalanced nutrient supply 187 outweigh soil nutrient mining2 and that the reduction in area of nutrient-mined soils, which can be 188 expected to increase the exogenous nutrient demands for closing yield gaps in our scenarios, can be 189 compensated by reduced applications of N and P tailored to meet crop demands in areas of presently 190 excessive fertilization (Supplementary Text S3). Yet, locally, foremost N and partly P surpluses may well 191 exceed those reported for around the year 2000, depending on which input sources are considered 192 (Supplementary Figure S11). Especially the MLS scenario results in a shift towards higher local N 193 surpluses per area whereas the TLS scenario closely resembles past patterns of conservative estimates 194 neglecting inputs form manure, deposition, and biological fixation. For the TLS scenario, low P surpluses 195 occur more frequently, in part due to the larger extent of remaining cropland compared to the MLS 196 scenario, which again exhibits a more frequent occurrence of moderate to high surpluses. Notably, the 197 latter is also caused by a larger fraction of cropland remaining in tropic regions in which weathered soils 198 with high P fixation occur more frequently38. 199 200 Figure 4: Changes in key agricultural externalities following optimization of area of cropland for (a) maximum land sparing (bar 201 2 in Figure 2) or (b) targeted land sparing (bar 3 in Figure 2), compared with the baseline scenario (100% circle; see bar 1 in 202 Figure 2; status in 2011-2015) for 16 major crops (dark colors) and the remaining annual crops (light colors; not estimated for 203 biodiversity potential). Proportions of nitrogen (N) and phosphorus (P) fertilizer, and irrigation water applied to crops during the 204 reference period were extrapolated linearly from crops in c. 2000 reported by Mueller et al.19, or for irrigation water from Siebert 205 and Doell3. Greenhouse gas emissions comprise methane from rice and nitrous oxide from fertilizer and assume the other major 206 sources (manure and crop residue) remain unchanged. Carbon (C) lost from potential natural vegetation is the amount of C 207 stored in potential natural vegetation after cropland sparing relative to that during the baseline (100% of C in natural vegetation 208 lost in cropland), using data from West et al39. Habitat for small-range species is the average change in habitat area for 209 terrestrial mammals intolerant to cropland in the lower quartile of range size distributions of terrestrial mammals in the IUCN 210 Red List of Threatened Species40 after recovery of natural vegetation on abandoned cropland. Details on the quantification of 211 each externality are provided in the Methods. 212 Crop water requirement from irrigation decreased under the MLS scenario by 380 km3 to 65% of the 213 baseline (approx. 1100 km3), and under the TLS scenario by 218 km3 to 78% of the baseline, precluding 7 214 losses within the irrigation system that exceed the actual global crop water requirement41. Water 215 requirements vary with crop3, climate, and land surface extent42; hence the reduction in cropland area is 216 a main driver of reduced irrigation volume. Thus, cropland sparing does not necessarily entail expansion 217 of irrigation infrastructures if yields in rainfed regions are maximized by optimal fertilization and crop 218 choice. This is consistent with earlier global and regional studies finding nutrient limitations to be a 219 substantially more important driver for current yield gaps than irrigation19,43. 220 Greenhouse gas emissions from paddy rice and fertilized soils decreased to 87% and 82% (-0.15 and - 221 0.21 Pg CO2 equiv.) of the baseline in the MLS and TLS scenario, respectively. As N application remains 222 fairly constant, this is mostly due to the decrease of CH4 emissions from the reduced cultivation area for 223 rice. Carbon (C) lost from potential natural vegetation is used as a proxy for C sequestration potential, if 224 natural vegetation on spared cropland fully recovers. The largest C storage capacity occurs in tropical 225 ecosystems, the lowest in arid climates39. Accordingly, the proportion of cropland remaining in the 226 tropics under the MLS scenario (Figure 3b) resulted with 29% avoided loss of C from natural vegetation 227 on present cropland in a proportionally low sequestration potential. However, this sequestration 228 potential is equivalent to 20.5 Pg C, underpinning that land sparing for vegetation restoration may halt 229 further deforestation that is a major contributor to global CO2 emissions. The amount of C sequestration 230 potential is higher in the TLS scenario, as major biodiversity hotspots are located in the tropics 231 (Supplementary Figure S16). This increases the C sequestration potential to 24.2 Pg C. 232 The habitat suited mammal species with restricted ranges and intolerant to cropland (n=716) in 233 presently cultivated regions increases substantially in the TLS scenario (+12.8%) but only marginally in 234 the MLS scenario (+2.6%). When considering all species of terrestrial mammals occurring in present 235 cropland regions (n=3922), the average gains decrease to 7.6% under the TLS scenario and increase to 236 4.9% in the MLS scenario (see Supplementary Figure S17 for results on various species groups). The 237 effect in the TLS scenario is partly due to the fact that cropland is spared therein specifically for small- 238 range species. The modest gain in average habitat for all terrestrial mammals in the MLS scenario in turn 239 reflects that cropland presently covers about 10% of the global ice-free land surface and therefore only 240 a comparably small fraction of actual and potential natural vegetation. Thus, our results underpin that 241 land sparing is most effective if pursued in a targeted way and focused on species strongly affected by 242 conversion of natural vegetation to cropland. 243 Our assessment of potential biodiversity impacts quantifies changes in suitable habitat area for species 244 intolerant to cropland, a time-independent indicator free of assumptions on population dynamics and 245 applicable for a wide range of species44,45. Yet, this neglects potential impacts of intensification on 246 biodiversity in situ on cropland. The bulk of empirical studies on species density-crop yield relationships 247 found that these follow a negatively convex functional form for species sensitive to cropland with 248 rapidly decreasing species density already at low yields46. This favours land sparing as a conservation 249 strategy opposed to land sharing or wildlife-friendly farming. The abundance of species tolerant to 250 cropland in turn may depend on multiple factors such as crop diversity and field configuration, nutrient 251 inputs, pesticide applications, small-scale landscape configuration, and species’ sensitivities to these 252 aspects47. Due to lack of data and granular spatial resolutions, these aspects cannot be addressed herein 253 and hardly in global studies at present. Indications that substantial land sparing can be achieved with 254 sustainable intensification in some regions but less so in others is provided in the evaluation of crop 255 diversity and nutrient budgets above. Yet, local assessments employing detailed species- and 256 ecosystem-specific knowledge will be required to explicitly quantify such effects. 8 257 In summary, both land sparing scenarios entail various co-benefits along agro-environmental 258 dimensions. Thereby, the targeted land sparing approach not only allows for the implicitly higher habitat 259 restoration potential, but also lower nutrient requirements and higher C sequestration potential, 260 although differences between scenarios are often marginal. As all modelling studies, our findings are 261 subject to a range of uncertainties and limitations, which we consider to render our results conservative 262 rather than overly optimistic (Supplementary Text S3 and S4). 263 Conclusions and wider implications of extensive land sparing 264 The potential for cropland sparing quantified herein contrasts with earlier agro-economic studies 265 indicating that further cropland expansion is likely to occur in future decades due to increasing crop 266 demands and slow diffusion rates of agricultural technologies13,48,49. Noteworthy, these forward-looking 267 studies account for changes in climate and atmospheric CO2 concentration as well as socio-economic 268 drivers and constraints, including diffusion rates for improved agricultural technologies, national 269 agricultural policies, international trade relations, and future increases in demands, which limit their 270 comparability to ours. Earlier studies exploring combinations of biophysical and socio-economic options 271 for abating increasing land pressure of agricultural production already identified agro-technologic 272 change as an important element15,16 but presented compound scenarios that do not allow for 273 quantifying the land sparing potential of optimal crop production and associated externalities directly. 274 Quantifications of production potentials17–21 in turn do not consider actual crop demands and none of 275 the mentioned studies covered targeted land sparing for wildlife habitats and other landscape elements. 276 In this context, our results provide a benchmark of the present potential for cropland sparing if high land 277 use efficiency was realized and if specific targets are defined for restoring wildlife habitats. 278 The gap between the present extent of global cropland and the actual cropland requirement quantified 279 herein indicates that at the global scale land management and associated policies, rather than 280 biophysical limitations, are the major production-side drivers of adverse environmental change 281 mediated by the expansion of cropland48. Thus, achieving ambitious land sparing targets in the near 282 term will require radical acceleration in the dissemination of available agro-technologies as well as 283 integration across society31 to avoid cropland expansion often caused by sole incentives for 284 intensification7 while maintaining livelihoods of populations potentially affected by agricultural change. 285 Globally coordinated efforts50 will be required to balance national interests concerning food security and 286 agricultural revenues with global environmental targets. 9 287 Methods and Data 288 Study design and land sparing scenarios 289 The study investigated global cropland sparing potential based on crop modelling of attainable yields for 290 16 major crops, crop-specific land use datasets, and spatial optimization of cropland allocation (Figure 1) 291 to minimize global cropland extent via maximizing land use efficiency, i.e. assigning the most productive 292 crops to cropland locally. The considered crops represent 85% of global cropland cultivated with annual 293 crops and sum up to more than 75% of total cropland area, vegetal calorie supply, and fertilizer 294 consumption29. With the exceptions of cassava and sugarcane, we excluded perennial crops from our 295 analyses, due to their low flexibility for crop switching and specific trajectories of yield improvement. 296 Within the optimization algorithm, current crop-specific area may expand or shrink with the goal of 297 minimizing global cropland extent, while maintaining defined crop-specific production volumes reported 298 by FAO for 2011-201529 and without expanding total cropland extent locally. We opted for the most 299 recent period for which data are available to account for contemporary increases in crop production. 300 The five-year mean is a compromise between avoiding bias from selecting a single year and 301 underestimating present production volumes when using a longer historical period. The study design is 302 further detailed in Supplementary Methods S1 and visualized in Figure 1. 303 We evaluated cropland sparing potential for two distinct main scenarios: (i) the “maximum land sparing” 304 (MLS) potential allowing the entire present cropland in each simulation unit or pixel to remain occupied 305 after crop reallocation if it is a solution of the optimization, and (ii) a “targeted land sparing” (TLS) 306 scenario. The latter forces the release of all cropland covered by the considered crops in biodiversity 307 hotspots and a uniform release of at least 20% of present cropland cover by 16 major crops in each 308 simulation unit or pixel. The latter fraction is considered to spare a compartment of the landscape for 309 other, i.e. regenerative, uses. Herein, it is assumed to be covered by natural vegetation in the 310 quantification of externalities (carbon sequestration and area of habitat), but may in principle also serve 311 for buffer strips, windbreaks, or other landscape elements. 312 Two supplementary scenarios based on the MLS scenario (Figure 1E) provide additional information (I) 313 whether the cultivation of crops in regions in which their cultivation is presently not recorded plays a 314 major role in the land sparing potential found herein, which is termed “MLS without crop switching” 315 (MLSncs); and (II) if substantial cropland sparing is still feasible if single crops are allowed to only cover 316 ≤34% of cropland in each simulation unit, indirectly enforcing the occurrence of several crops in most 317 simulation units and hence fostering crop diversity, which allows for crop rotations. This scenario is 318 termed “MLS with crop diversity” (MLSwcd). 319 Land use optimization approaches similar to the main scenarios have been studied earlier, but 320 addressed global production potentials employing input intensification only19, crop switching only20,21, or 321 both18, or investigated production potentials for single crops under climate change17. Land sparing 322 potential of optimized cropland allocation has been addressed by Müller et al.51 among other aspects of 323 crop production and consumption. Yet, constraints on available land for cropping per pixel were not 324 considered below the physical pixel area, crop demands were partly computed, and intensification was 325 not accounted for. 326 Crop modelling framework 10 327 Crop simulations were performed for 16 major crops (Figure 2) with the well-established global gridded 328 crop model (GGCM) EPIC-IIASA17, which is based on the field-scale process-based agronomic 329 Environmental Policy Integrated Climate (EPIC) model25,26 (formerly known as Erosion Productivity 330 Impact Calculator). EPIC-IIASA has been applied extensively in global impact studies and across regions, 331 and has been evaluated positively for reproducing both historic absolute yields under business-as-usual 332 management and inter-annual yield variability52–54. Simulated attainable crop yields were capped at the 333 95th percentile globally to avoid bias towards extremely high yields in the crop-to-cropland allocation. 334 Key processes of the core model EPIC are summarized in Folberth et al.55 and briefly described in 335 Supplementary Methods S3. 336 EPIC-IIASA is based on a 5 x 5’ grid (equivalent to about 8.3 km x 8.3 km near the equator) for soil 337 characteristics56 and topography57 that are aggregated, based on classification of key characteristics, to 338 homogenous response units. These are further intersected using a 30 x 30’ climate grid (about 50 km x 339 50 km near the equator) and national administrative boundaries to define final simulation units58. 340 Accordingly, simulation units vary in size from 5’ x 5’ to 30’ x 30’ depending on local heterogeneity. 341 More detail on the definition of simulation units is provided in Supplementary Methods S2. The EPIC 342 model was run for each simulation unit, crop, and water management system (rainfed or with sufficient 343 irrigation) separately, treating it as a representative homogenous field. Climate data were based on the 344 daily climate database AgMERRA59, specifically developed for agricultural applications, at a spatial 345 resolution of 30’ x 30’. Crop-specific growing seasons were derived from Sacks et al.60. Supplementary 346 Methods S2 provide further details on the EPIC-IIASA model. 347 Data on multi-cropping are lacking at the global scale and are only reflected in reported harvest areas 348 that partly exceed the physical area of cropland. As our focus was on physical cropland sparing, we 349 focused our optimization on single cropping of physical cropland, disregarding potential multi-cropping 350 and rotations. The exception was for rice cultivation: according to SPAM 2005 v3.261, total cropping 351 intensity is about 115%, with single cropping dominant in most crops, but an intensity of 150% for rice. 352 Therefore, rice was simulated for two seasons where suggested by calendar data, to minimize 353 underestimation of rice double cropping. Yields for the two seasons were summed to treat double- 354 cropped rice as a single crop in the estimation of physical area requirements. For the land use datasets 355 referring to harvested area (see below), separate rice simulations were performed for a single season. A 356 brief discussion of the potential impacts of multi-cropping is provided in Supplementary Text S4. 357 Evaluation of attainable crop yields 358 We evaluated simulated attainable yields against two widely used spatially explicit datasets, based on 359 reported yields and extrapolation of (a) high-input rainfed and irrigated crop yields from SPAM 2005 360 v3.261 and (b) attainable yields19 based on the M3 dataset19. Evaluations are presented in Supplementary 361 Text S2 and Supplementary Figures S12 and S14. All three datasets (including the estimates from 362 biophysical crop modelling) were derived using different methodologies; this limits comparability of 363 yield distributions. It may be assumed, however, that our comparison allows for the evaluation of crop 364 model overestimation of yield potentials. It should be noted that the yield category closest to attainable 365 yields in SPAM reports rainfed, high-input yields, based on moderate to sufficient levels of nutrient 366 input. Irrigated yields are a single category that may typically be assumed to receive high (unknown) 367 levels of nutrient inputs. The attainable yields from M3 are based on spatially explicit reported yields c. 368 2000 from administrative level censuses and climate bins based on temperature and precipitation. For 11 369 each of these climate bins, the upper 95th percentile of reported yields is assumed to represent the 370 attainable yield. 371 Cropland allocation model 372 Spatially explicit cropland optimization was performed at the level of simulation units with the objective 373 of minimizing global cropland requirement, but maintaining 2011-2015 production volumes for each 374 crop29 (Figure 1). Reported production as a target accounts for any dietary and other use preferences as 375 opposed to more aggregated approaches based on recommended supply levels or requirements. 376 The main cropland dataset selected for the analysis was SPAM 2005 v3.2, because it provides crop- 377 specific physical areas. In contrast to other datasets that typically report either crop-specific harvested 378 areas or total physical cropland, this dataset allows for the assessment of physical cropland sparing 379 potential only for cropland cultivated with the crops included in this analysis. Robustness of our results 380 was evaluated from the optimization of two additional crop-specific harvested area datasets (see 381 below). 382 The land use optimization model was programmed in GAMS software (https://gams.com/), where input 383 data are yield potentials from either the EPIC crop model or inventory data (see below) and current 384 crop-specific areas at the simulation unit level. Thresholds for uniform cropland release in the TLS 385 scenario were defined by finding a minimal feasible solution in steps of 85%, 80%, 67%, and 50% for 386 each attainable crop yield x cropland dataset combination. For the SPAM 2005 physical area dataset, 387 this threshold was found to be 80% (or 20% of uniformly released land). 388 The optimization problem is formulated as: 389 σ௜ǡ௝ǡ௞ܽ௜௝ ‹‹‹œ‡  ‫ݏ‬௜௝௞ (Eq. 1) ௦ ೔ೕೖ 390 • Ǥ–Ǥ  σ௜ǡ௝ܽ௜௝ ‫ݏ‬ ‫ݕ‬௜௝௞ ௜௝௞ ൒‫݌‬௞ǡ (Eq. 2) 391 σ௞‫ݏ‬ ൑ ௜௝௞ ߙǡ‫ݏ‬ ൒ͲǤ ௜௝௞ (Eq. 3) 392 393 where ܽ௜௝ is current area of cropland [ha] occupied by the considered crops in simulation unit ݅under 394 water supply type ݆ ;‫ݏ‬ ௜௝௞is the respective share allocated to crop ݇to be optimized; ‫ݕ‬௜௝௞ is the 395 simulation unit-, irrigation type-, and crop-specific yield [t ha ]; ‫݌‬௞is current production of crop ݇[t]; -1 23 396 ߙis the maximum allowed optimized cropland share within the considered simulation unit area, ߙൌͳ 397 for maximum land sparing; and ߙൌͲǤ ͺͲ for SPAM for the TLS scenario. For the complementary 398 datasets (see below), ߙൌͲǤ ͸͹ for MIRCA, and ߙൌͲǤ 85 for the M3 dataset. 399 We performed optimizations for additional datasets and combinations thereof to account for 400 uncertainties in cropland distribution62 and attainable yields. Crop model estimated attainable yields 401 were combined with cropland distributions from SPAM 2005 v3.261 or MIRCA200063 that provide 402 spatially explicit harvested areas for the considered crops, for rainfed and irrigated cultivation systems 403 separately. We performed the same complementary optimization using a set of statistically derived 404 attainable yields and corresponding areas from M319,64; this dataset does not distinguish between 405 rainfed and irrigated systems, so yields were not combined with the other land use datasets. As none of 406 the spatial datasets provides the same crop-specific areas as FAOSTAT for the reference period 2011- 12 407 201523, crop areas from FAOSTAT were used as a basis from which to derive relative cropland area 408 reduction, after an absolute number of cropland requirement had been obtained in the optimization 409 routine. Accordingly, cropland areas in all spatial datasets underestimate present cropland, which 410 increased for the considered crops by about 14% since 2000 (M3 and MIRCA2000 reference), and by 7% 411 since 2005 (SPAM 2005 reference). Further limitations and uncertainties of the land sparing modelling 412 and estimation of attainable yields are addressed in Supplementary Text S4. 413 Definition of biodiversity hotspots 414 Biodiversity hotspots were defined based on rarity-weighted richness as the sum of number of species 415 present in a grid cell weighted by their range size (1/Area of Habitat (AOH))65. Higher values occur in grid 416 cells rich in species with small ranges. These cells have a large global responsibility for species 417 conservation. Rarity-weighted richness was quantified in absolute terms and in addition normalized per 418 WWF ecoregion66 and continent to account for regions of (a) high absolute importance for biodiversity, 419 which are typically concentrated in the tropics, and (b) regional importance for biodiversity within 420 specific ecoregions44. From both resulting datasets, the 90th percentile was selected to be abandoned for 421 targeted land sparing in the TLS scenario (Supplementary Figure S16). 422 Quantification of agricultural externalities 423 Crop nutrient requirements 424 N, P, and irrigation water were applied by the EPIC model based on deficits compared with optimal 425 supply and relative crop stress thresholds (see Supplementary Methods S2). The model considered 426 losses (leaching, runoff, erosion, immobilization, and gaseous emissions) and limited the number of crop 427 management operations to a level common to current management practices (annual application of P, 428 and restricted number of applications for N and water within a given time period) to represent an 429 optimal management strategy that balances realistic overheads for plant nutrient inputs. Fertilizer 430 requirements for crops that were not considered in the optimization were derived from the proportions 431 of crop-specific fertilizer application rates around 200019 to total fertilizer application volumes during 432 the 2011-2015 reference period, as reported in FAOSTAT29. 433 Besides exogenous inputs, nutrients used by crop plants are also sourced from soil stocks and 434 mineralization of organic matter as well in the field as in the crop model. While these represent a 435 substantial short-term source of nutrients, depletion occurs over time that may lead to the 436 underestimation of fertilizer requirement. Amounts of N and P required for sustainable nutrient 437 replenishment in such cases were estimated from a fertilizer requirement of 120% of crop uptake. For 438 soils with high or very high P immobilization potential67, we ensured the fertilizer requirement was twice 439 the crop uptake38. For leguminous crops (groundnuts, pulses, and soybean), we assumed that at yields 440 >2.5 t ha-1, only 80% of N demand is met through fixation68, and added 20% of crop uptake as 441 supplementary fertilizer. More details on the ex-post accounting for potentially higher nutrient 442 requirements than estimated by the crop model are provided in Supplementary Methods S3. 443 Nutrients in plant residues and manure 444 N and P embodied in removed crop residues (straw, stalks, stover) or burning of crop residues in the 445 field were not modelled explicitly. To account for removal of N and P from the field as post-harvest 446 residues in supplementary evaluations, we estimated crop residue dry matter from reference period 13 447 crop production volumes23 and crop harvest indices in the EPIC model, and then calculated volumes of N 448 and P based on the USDA crop nutrient tool69. National crop-specific residue removal and burning rates 449 were obtained from a recent global report70 that covers all crops included in this study, with the 450 exception of sugar beet, groundnut, pulses, millet, and rice. For the first four of these crops, we 451 approximated values using coefficients of potato for sugar beet, soybean for groundnuts and pulses, and 452 sorghum for millet. For countries lacking data, we applied a mean based on major UN regions. Data for 453 rice were obtained from a recent literature review4. For burned residue, we assumed that 80% of N and 454 40% of P are lost as emissions. Total removal from the field amounted to 19.6 Tg N and 2.2 Tg P, 455 respectively. Fertilizer requirements were scaled according to a fertilizer:uptake ratio in crop yield, to 456 account for additional losses due to increased fertilizer application. Present amounts of N and P 457 contributed by manure cycling to cropland were estimated from the literature as 17.3 Tg and 4.2 Tg, 458 respectively71,72. The additional or reduced requirements for N and P replenishment with present rates 459 of residue removal and manure application, as well as uncertainties in the nutrient budgets, are 460 discussed in Supplementary Text S3. 461 Irrigation water requirement 462 Irrigation water requirements estimated by the EPIC model to meet plant water demand do not 463 consider inefficiencies due to losses during the extraction to field application process. These may be 464 more than twice the actual plant demand, depending on the irrigation system in place41. For the relative 465 change in irrigation water requirement for the crops considered, we compare the irrigation requirement 466 on the total cropland to that in each land sparing scenario. To account for the crops not considered in 467 the simulations, we scaled crop-specific irrigation water requirements from a study based on the Global 468 Crop Water Model (GCWM) model that considers all major crops or crop groups3. 469 Expansion of irrigated land would also provide a means for increasing crop yields73 and accordingly 470 decreasing land requirement. We do not consider this option here due to its lower flexibility compared 471 to nutrient input intensification as (a) it requires upfront investment in infrastructure, (b) it is subject to 472 policy and governance decisions on water resources, (c) it is subject to competition among sectors, and 473 (d) inter-annual variations in water availability for irrigation affect crops differently in-situ based on 474 economic considerations among others74. 475 Greenhouse gas emissions 476 Greenhouse gas emissions in CO2 equivalents were calculated following the tier 1 methodology of FAO75 477 for the major cropland emission contributors of paddy rice fields (CH4) and nitrogen fertilizer (N2O), 478 based on fixed N2O emissions per unit of applied fertilizer and national coefficients of CH4 emissions ha-1 479 of harvested paddy rice. Other emissions, for example from manure and crop residues, were assumed to 480 remain constant. Estimates of emissions of N2O for crops not considered in the optimization were based 481 on N fertilizer requirements, as calculated above. 482 Carbon in potential natural vegetation 483 The potential loss of C from natural vegetation expected to develop on spared cropland has been 484 investigated by West et al.39 to quantify C losses in food production. Using the publically available 485 dataset of C stored in potential natural vegetation [t ha-1], we quantified reductions in C loss following 486 minimization of cropland area compared with the baseline cropland area in the SPAM 2005 v3.2 14 487 database for crops considered in the optimization and for other crops, separately. The exact calculation 488 is provided in Supplementary Methods S4. 489 Area of habitat 490 We modeled the Area of Habitat (AOH) for each terrestrial mammal species with range data and habitat 491 preferences available from the IUCN Red List database (accessed April 2018). The AOH is defined as the 492 area characterized by abiotic and biotic properties that is habitable by a particular species. Specifically, 493 we modelled the AOH as the areas that (i) fall within the mapped range and (ii) map to the known 494 habitat preferences of the species. The species ranges of terrestrial mammals were downloaded from 495 the IUCN database. We considered only habitat types coded as ‘suitable’ by taxonomic experts within 496 the IUCN database. In absence of a map of IUCN habitat classes, and similarly to all previous work 497 modelling of AOH44,45, we cross-walked the IUCN habitat classes into an existing land-use product to 498 translate habitat preferences into land-cover and land-use preferences. Accordingly, our assessment 499 only accounts for biogeographic distributions of species habitats but not for impacts of land use 500 intensification on wild species on cropland in situ. 501 As land-cover base layer we used the European Space Agency CCI (ESA-CCI) land-cover map for the year 502 201576 and re-allocated cropland areas as calculated from the SPAM2005 baseline or the land sparing 503 scenarios, including annual and perennial crops not considered in the land use model to account for all 504 cropland. When cropland area was higher than estimated in the ESA-CCI map, the additional area was 505 allocated to all natural land-cover types (except water and ice) in proportion to their extent in the grid 506 cell. Similarly, when cropland area was lower than estimated in the ESA-CCI map, the excess cropland 507 was allocated to all natural land-cover types (except water and ice) in proportion to their extent in the 508 grid cell. We then summarized the results as distribution of AOH changes in optimized versus baseline 509 scenarios across all species, species sensitive to cropland areas (those for which cropland is considered 510 unsuitable according to IUCN habitat preferences), species in the lower quartile of range-size 511 distribution, and species in the lower quartile of range-size distribution sensitive to cropland areas. The 512 latter was selected as the main results, outcomes for the other species sub-selections are presented in 513 Supplementary Figure S17. 514 Data processing and visualization 515 Evaluations were performed in R77, and plots were produced using ggplot278 and rasterVis79. The 516 visualization of simulation units in Supplementary Figure S19 was produced with ESRI ArcGIS 10.7. 517 518 Correspondence and requests for materials should be addressed to CF 519 Acknowledgements 520 CF, NK, JB, RS, PC, IAJ, JP, and MO were supported by European Research Council Synergy grant ERC- 521 2013-SynG-610028 Imbalance-P. We gratefully acknowledge the provision of threatened species data by 522 IUCN and the provision of land cover data by the ESA CCI Land Cover project. 523 Author contributions 15 524 CF, NK, and MO designed the study; CF and NK performed central analyses; JB, RS, and PV contributed 525 models and data; CF wrote an initial draft; all authors contributed substantially to interpretation of the 526 results and revisions of the manuscript. 527 Data availability 528 The datasets generated during the current study are available from an FTP server for review purposes: 529 http://user.iiasa.ac.at/~folberth/land_sparing/. They will be made available in a public repository upon 530 publication. 531 16 532 References (including Methods) 533 1. van der Velde, M.et al. African crop yield reductions due to increasingly unbalanced 534 Global Change Biology20, 1278–1288 (2014). Nitrogen and Phosphorus consumption. 535 2. MacDonald, G. K., Bennett, E. M., Potter, P. A. & Ramankutty, N. Agronomic phosphorus 536 imbalances across the wor ld’s croplands. Proceedings of the National Academy of Sciences 537 108, 3086–3091 (2011). 538 3. Siebert, S. & Döll, P. Quantifying blue and green virtual water contents in global crop 539 Journal of Hydrology384, production as well as potential production losses without irrigation. 540 198–217 (2010). 541 4. Carlson, K. M.et al. Greenhouse gas emissions intensity of global croplands. Nature Climate 542 Change7, 63–68 (2017). 543 5. Steffen, W.et al. Planetary boundaries: Guiding human development on a changing planet. 544 Science347, 1259855 (2015). 545 6. Balmford, A. & Green, R. How to spare half a planet. Nature 546 http://www.nature.com/articles/d41586 -017-08579-6 (2017) doi:10.1038/d41586 -017-08579- 547 6. 548 7. Ewers, R. M., Scharlemann, J. P. W., Balmford, A. & Green, R. E. Do increases in 549 Global Change Biology15, 1716–1726 (2009). agricultural yield spare land for nature? 550 8. Phalan, B.et al. How can higher Science351, 450–451 -yield farming help to spare nature? 551 (2016). 552 9. Salles, J. -M., Teillard, F., Tichit, M. & Zanella, M. Land sparing versus land sharing: an 553 17, 1455–1465 (2017). economist’s perspective. Reg Environ Change 554 10. Wilson, E. O.Half-earth: our planet’s fight for life. (WW Norton & Company, 201 6). 17 555 11. Bodirsky, B. L.et al. Global Food Demand Scenarios for the 21st Century. PLOS ONE 556 10, e0139201 (2015). 557 12. Popp, A.et al. Land-use futures in the shared socio -economic pathways. Global 558 42, 331–345 (2017). Environmental Change 559 13. Nelson, G. C.et al. Climate change effects on agriculture: Economic responses to 560 PNAS111, 3274–3279 (2014). biophysical shocks. 561 14. Mehrabi, Z., Ellis, E. C. & Ramankutty, N. The challenge of feeding the world while 562 Nature Sustainability1, 409–412 (2018). conserving half the planet. 563 15. Erb, K.-H. et al. Exploring the biophysical option space for feeding the world without 564 7, 11382 (2016). deforestation.Nature Communications 565 16. Springmann, M.et al. Options for keeping the food system within environmental limits. 566 Nature 562, 519–525 (2018). 567 17. Balkovič, J. et al. Global wheat production potentials and management flexibility under 568 the representative concentration pathways. 122, 107–121 Global and Planetary Change 569 (2014). 570 18. Mauser, W.et al. Global biomass productionotentials p exceed expected future demand 571 without the need for cropland expansion. 6, 8946 (2015). Nature Communications 572 19. Mueller, N. D.et al. Closing yield gaps through nutrient and water management. Nature 573 490, 254–257 (2012). 574 20. Koh, L. P., Koellner,T. & Ghazoul, J. Transformative optimisation of agricultural land 575 PeerJ1, e188 (2013). use to meet future food demands. 18 576 21. Davis, K. F., Rulli, M. C., Seveso, A. & D’Odorico, P. Increased food production and 577 reduced water use through optimized cropribution. dist 10, 919–924 Nature Geoscience 578 (2017). 579 22. Daedalus144, Balmford, A., Green, R. & Phalan, B. Land for Food & Land for Nature? 580 57–75 (2015). 581 23. Balmford, A. et al. The environmental costs and benefits of high -yield farming.Nature 582 Sustainability1, 477–485 (2018). 583 24. Ray, D. K., Ramankutty, N., Mueller, N. D., West, P. C. & Foley, J. A. Recent patterns of 584 crop yield growth and stagnation. 3, 1293 (2012). Nature Communications 585 25. Williams, J. R. The erosion -productivity impact calculator (EPIC) model: a case history. 586 Phil. Trans. R. Soc. Lond. 329, B 421–428 (1990). 587 26. Izaurralde, R. C., Williams, J. R., McGill, W. B., Rosenberg, N. J. & Jakas, M. C. Q. 588 Simulating soil C dynamics with EPIC: Model description and testing against long -term data. 589 Ecological Modelling192, 362–384 (2006). 590 27. Feniuk, C., Balmford, A. & Green, R. E. Land sparing to make space for species 591 dependent on natural habitats and high nature value farmland. Proceedings of the Royal 592 286, 20191483 (2019). Society B: Biological Sciences 593 28. Schulte, L. A.et al. Prairie strips improve biodiversity and the delivery of multiple 594 ecosystem services from corn PNAS114, 11247 (2017). –soybean croplands. 595 29. FAO. FAOSTAT st atistical database . (2016). 596 30. Schleicher, J.et al. Protecting half of the planet could directly affect over one billion 597 people.Nat Sustain(2019) doi:10.1038/s41893 -019-0423-y. 598 31. Science364, 1226–1228 (2019). Ellis, E. C. Sharing the land between nature and people. 19 599 32. Verburg, P. H., Mertz, O., Erb, -H., K. Haberl, H. & Wu, W. Land system change and 600 food security: towards multi -scale land system solutions. Current Opinion in Environmental 601 Sustainability5, 494–502 (2013). 602 33. Puma, M. J., Bose, S.,hon, C S. Y. & Cook, B. I. Assessing the evolving fragility of the 603 Environ. Res. Lett.10, 024007 (2015). global food system. 604 34. Alston, J. M., Babcock, B. A. & Pardey, P. The G. shifting patterns of agricultural 605 production and productivity worldwide . (Midwest Agribusiness Trade Research and 606 Information Center, 2010). 607 35. Müller, D. et al. Regime shifts limit the predictability of land -system change. Global 608 28, 75–83 (2014). Environmental Change 609 36. Kastner, T., Erb, K. -H. & Haberl, H. Rapid growth in agricu ltural trade: effects on global 610 Environ. Res. Lett.9, 034015 (2014). area efficiency and the role of management. 611 37. Barzman, M.et al. Eight principles of integrated pest management. Agron. Sustain. Dev. 612 35, 1199–1215 (2015). 613 38. Roy, E. D.et al. The phosphorus cost of agricultural intensification in the tropics. Nature 614 Plants2, 16043 (2016). 615 39. West, P. C.et al. Trading carbon for food: Global comparison of carbon stocks vs. crop 616 yields on agricultural land. Sci 107, 19645– Proceedings of the National Academy ofences 617 19648 (2010). 618 40. IUCN. The IUCN Red List of Threatened Species. https://www.iucnredlist.org. (2018). 619 41. Jägermeyr, J. et al. Water savings potentials of irrigation systems: global simulation of 620 Hydrol. Earth Syst.Sci. 19, 3073–3091 (2015). processes and linkages. 20 621 42. Sterling, S. M., Ducharne, A. & Polcher, J. The impact of global-cover land change on 622 the terrestrial water cycle. 3, 385–390 (2013). Nature Climate Change 623 43. Folberth, C., Yang, H., Gaiser, T., Abbaspour, K. C. & Schulin, R. Modeling maize yield 624 responses to improvement in nutrient, water and cultivar inputs -Saharan in sub Africa. 625 119, 22–34 (2013). Agricultural Systems 626 44. Leclere, D.et al. Towards pathways bending the curve of terrestrial biodiversity trends 627 within the 21st century. http://pure.iiasa.ac.at/id/eprint/15241/ (2018). 628 45. Visconti, P.et al. Projecting Global Biodiversity Indicators under Future Development 629 9, 5–13 (2016). Scenarios.Conservation Letters 630 46. Phalan, B. T. What Have We Learned from the Land aring-sharing Sp Model? 631 Sustainability10, 1760 (2018). 632 47. ǦDewenter, I. & Thies, C. Landscape Tscharntke, T., Klein, A. M., Kruess, A., Steffan 633 perspectives on agricultural intensification and biodiversity – ecosystem service management. 634 Ecology Letters8, 857–874 (2005). 635 48. Stehfest, E.et al. Key determinants of global land -use projections.Nature 636 Communications10, 2166 (2019). 637 49. Schmitz, C.et al. Land-use change trajectories up to 2050: insights from a global- agro 638 Agricultural Economics45, 69–84 (2014). economic model comparison. 639 50. Schmidt-Traub, G., Obersteiner, M. & Mosnier, A. Fix the broken food system in three 640 steps.Nature569, 181–183 (2019). 641 51. ǦCampen, H., Cramer, W. & Lucht, W. Comparative Müller, C., Bondeau, A., Lotze 642 impact of climaticand nonclimatic factors on global terrestrial carbon and water cycles. 643 20, (2006). Global Biogeochemical Cycles 21 644 52. Müller, C. et al. Global gridded crop model evaluation: benchmarking, skills, deficiencies 645 10, 1403–1422 (2017). and implications.Geosci. Model Dev. 646 53. Balkovič, J. et al. Impacts and Uncertainties of +2°C of Climate Change and Soil 647 Earth’s Future 6, 373–395 (2018). Degradation on European Crop Calorie Supply. 648 54. Balkovič, J. et al. Pan-European crop modelling with EPIC: Implementation,-scaling up 649 and regional crop yield validation. 120, 61–75 (2013). Agricultural Systems 650 55. Folberth, C.et al. Uncertainty in soil data can outweigh climate impact signals in global 651 7, 11872 (2016). crop yield simulations. Nature Communications 652 56. FAO, IIASA, ISRIC, ISSCAS & JRC. Harmonized World Soil Database (version 1.2). 653 (2012). 654 57. US Geological Survey. GTOPO30 - 30 arc seconds digital elevation model from US 655 Geological Survey. (2002) doi:https:/ /doi.org/10.5066/F7DF6PQS. 656 58. Skalský, R.et al. GEO-BENE global database for bio -physical modeling.GEOBENE 657 project (2008). 658 59. Ruane, A. C., Goldberg, R. & Chryssanthacopoulos, J. Climate forcing datasets for 659 agricultural modeling: Merged products for ap-filling g and historical climate series 660 200, 233–248 (2015). estimation.Agricultural and Forest Meteorology 661 60. Sacks, W. J., Deryng, D., Foley, J. A. & Ramankutty, N. Crop planting dates: an analysis 662 19, 607–620 (2010). of global patterns.Global Ecology and Biogeography 663 61. You, L., Wood, S., Wood -Sichra, U. & Wu, W. Generating global crop distribution 664 maps: From census to grid. 127, 53–60 (2014). Agricultural Systems 665 62. Porwollik, V. et al. Spatial and temporal uncertainty of crop yield aggregations. 666 88, 10–21 (2017). European Journal of Agronomy 22 667 63. Portmann, F. T., Siebert, S. & Döll, P. MIRCA2000 —Global monthly irrigated and 668 rainfed crop areas around the year 2000: A new-resolution high data set for agricultural and 669 24, (2010). hydrological modeling.Global Biogeochemical Cycles 670 64. Monfreda, C., Ramankutty, N. & Foley, J. A. Farming the planet: 2. Geographic 671 distribution of crop areas, yields, physiological types, and net primary production in the year 672 22, 1–19 (2008). 2000.Global Biogeochemical Cycles 673 65. Albuquerque, F. S. de & Gregory, A. The geography of hotspots of-weighted rarity 674 PLOS ONE12, e0174179 (2017). richness of birds and their coverage by Natura 2000. 675 66. Olson, D. M.et al. Terrestrial Ecoregions of the World: A New Map of Life Earth. on 676 BioScience51, 933–938 (2001). 677 67. Batjes, N. H.Global distribution of soil phosphorus retention potential . (2011). 678 68. Cafaro La Menza, N., Monzon, J. P., Specht, J. E. & Grassini, P. Is soybean yield limited 679 ch 213, 204–212 (2017). by nitrogen supply?Field Crops Resear 680 69. Crop Nutrient Tool | USDA PLANTS. https://plants.usda.gov/npk/main. 681 70. R Köble. The global nitrous oxide calculator – GNOC – Online tool manual v1.2.4. 682 (2014). 683 71. Liu, J. et al. A high-resolution assessment on global nitrogen flows in cropland. 684 107, 8035–8040 (2010). Proceedings of the National Academy of Sciences 685 72. Bouwman, L.et al. Exploring global changes in nitrogen and phosphorus cycles in 686 agriculture induced by livestock production over the 1900 –2050 period.Proc Natl Acad Sci 687 USA110, 20882 (2013). 688 73. Jägermeyr, J. et al. Integrated crop water management might sustainably halve the global 689 food gap.Environ. Res. Lett.11, 025002 (2016). 23 690 74. Nat Sustain2, 515–523 (2019). Qin, Y. et al. Flexibility and intensity of global water use. 691 75. Tubiello, F. N.et al. The FAOSTAT database of greenhouse gas emissions from 692 8, 015009 (2013). agriculture.Environmental Research Letters 693 76. Bontemps, S.et al. Consistent global land cover maps for climate modelling 694 communities: curre nt achievements of the ESA’s land cover CCI. in Proceedings of the ESA 695 Living Planet Symposium, Edinburgh 9–13 (2013). 696 77. RDevelopment Core Team. R: A language and environment for statistical computing . (R 697 foundation for statistical computing Vienna, Aust ria, 2008). 698 78. Wickham, H.ggplot2: elegant graphics for data analysis . (Springer, 2016). 699 79. Perpinan, O. & Hijmans, R. rasterVis. R package version 0.41 (2016). 700 24