UBC Faculty Research and Publications

The challenge of feeding the world while conserving half the planet Mehrabi, Zia; Ellis, Erle C.; Ramankutty, Navin 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52383-Mehrabi_Z_et_al_Challenge_feeding.pdf [ 6.91MB ]
Metadata
JSON: 52383-1.0376069.json
JSON-LD: 52383-1.0376069-ld.json
RDF/XML (Pretty): 52383-1.0376069-rdf.xml
RDF/JSON: 52383-1.0376069-rdf.json
Turtle: 52383-1.0376069-turtle.txt
N-Triples: 52383-1.0376069-rdf-ntriples.txt
Original Record: 52383-1.0376069-source.json
Full Text
52383-1.0376069-fulltext.txt
Citation
52383-1.0376069.ris

Full Text

   1                 Title:  The challenge of feeding the world while conserving half the planet 1                                 Authors: Zia Mehrabi1, Erle C. Ellis2, Navin Ramankutty1 2 Affiliations: 3  4 1The UBC School of Public Policy and Global Affairs, and Institute for Resources, Environment 5 and Sustainability, University of British Columbia, Vancouver, British Columbia V6T 1Z2, 6 Canada 7 2Department of Geography and Environmental Systems, University of Maryland, Baltimore 8 County, Baltimore, MD 21250, USA. 9  10 Amidst widespread concerns about biodiversity loss, a single clear conservation message is 11 engaging leading conservationists: the proposal to give half the surface of the Earth back to 12 nature.  Depending on the landscape conservation strategy, we find that globally 15-31% of 13 cropland, 10-45% of pasture land, 23-25% of non-food calories, and 3-29% of food calories 14 from crops could be lost if half of Earth’s terrestrial ecoregions were given back to nature. 15  16 The Convention on Biological Diversity, signed by 196 parties, is the world’s primary multi-17 lateral, legally binding, treaty for protection and sustainable use of the planet’s biological 18 resources.  Through it, world leaders made a commitment to halt biodiversity loss by 2010, but 19 failed1. This led to the development of the Strategic Plan for Biodiversity 2011-2020, and twenty 20 ambitious Aichi Biodiversity Targets2. With less than two years to go, these complex and 21 ambiguous targets to halt biodiversity loss by 2020 seem out of reach3.  22  23 Amidst these signs of likely failure, prominent conservation leaders are congregating around an 24 even more ambitious goal to give half the surface of the Earth back to nature (http://www.half-25 earthproject.org/; http://natureneedshalf.org/).  This proposal roughly equates with expanding the 26 – yet to be achieved – Aichi Target 11, to conserve 17% of world’s terrestrial and 10% of marine 27 areas, by roughly 3 and 5 times, respectively. In doing so, the project claims the potential to 28 conserve ~85% of existing species 4, by moving towards a system of interconnected high quality 29 habitats of sufficient scale locally, regionally and globally to support the persistence of natural 30 populations5. The added value of Half-Earth is also in its branding: the idea is conceptually 31 simple and visionary, creating a single banner under which other scattered conservation 32 initiatives could operate. As an aspirational goal, it is a powerful message that could motivate 33 and empower the public and local organizations to take positive action to protect the biosphere at 34 the level needed for reducing biodiversity decline.  35  36 Yet, despite these benefits, the practical costs of Half Earth incurred through trade-offs with 37 other land uses, and its impacts on already disadvantaged populations around the world, remain 38 poorly understood 6,7. Possibly the greatest trade-off embedded in the Half Earth proposal is with 39 agriculture: the dominant land use competing for space with other species on this planet and the 40 nexus of multiple Sustainable Development Goals linked to human health and wellbeing, climate 41 change, biodiversity loss and water security8. While relationships between food production and 42 biodiversity conservation have been analyzed previously9,10, the food production costs of Half 43 Earth have yet to be assessed. 44    2  45 In this paper, we offer an initial assessment of the potential global trade-offs between devoting 46 half a planet for nature and agricultural production.  We do this by assessing Half Earth 47 conservation strategies designed to maximize protection of key biodiversity areas and existing 48 wild lands while minimizing crop calorie losses caused by displacing arable agriculture. We 49 conduct analysis at the global, country, and ecoregion scales, giving half of the land within each 50 geographic boundary at each scale back to nature (see Methods for further details of our ranked 51 prioritization used in the analysis). We run this analysis for two drastically different conservation 52 approaches: “Nature only landscapes”, in which conservation displaces all crop production in 53 regional landscapes11 (~8.4km x ~8.4km pixels), and “Shared landscapes”, in which conservation 54 and crop production are allowed to coexist within each landscape (8.4km x ~8.4km pixel),  in 55 any possible spatial configuration (i.e. both intensive and extensive production are possible, see 56 Ref12 for details of the debate surrounding these extremes). This latter, "Shared landscapes” 57 approach works at the subpixel level, and so allows for implementation of fine scale 58 resolution conservation strategies that are not limited by the landscape size. 59  60 Our results demonstrate clear potential for trade-offs between Half Earth and agricultural 61 production that are strongly mediated by the approach taken and the spatial scale of the land 62 units used for conservation planning. Firstly, there is a strong negative effect of the spatial scale 63 of conservation units, with trade-offs increasing across Global (n=1) vs. Country (n=182) or 64 Ecoregion (n=775) scales. At the global level, under a “Nature only landscapes” strategy, we see 65 ~12% of cropland, ~21% of pasture, ~10% of non-food (i.e. feed, biofuel and other) calories, and 66 ~11% of food calories lost – numbers that increase to ~31% of cropland, ~45% of pasture, ~25% 67 of non-food calories, and ~29% of food calories at the Ecoregion level (Fig 1). Second, there are 68 massive differences in trade-offs between conservation approaches. Under “Shared landscapes”, 69 where agricultural production and conservation are allowed to coexist, productivity costs are 70 much lower, demanding ~15% of cropland, ~10% of pasture land, ~23% of non-food calories, 71 and ~3% of food calories at the Ecoregion level (Fig 1). However, even under a “Shared 72 landscapes” approach, if pasture is converted to conservation before cropland, and non-food 73 calories are abandoned before food calories, losses of non-food calories remain high.  Overall, 74 reaching the Half-Earth target without calorie losses could be achieved in only about 14% of 75 Earth’s ecoregions through a “Nature only” approach, while the “Shared” approach could 76 achieve this across more than four times as many ecoregions (65%) (Supplementary Information 77 Fig S1). Simply put, the trade-offs between agriculture and Half Earth will be much lower if 78 landscapes are allowed to remain as mosaics of shared land uses, and will be much higher if 79 large contiguous areas are given back, as may be required for the conservation of some species, 80 such as mega-fauna. 81  82 There are also differences in the spatial distribution of Half Earth’s agricultural trade-offs, 83 depending on the conservation approach (Figure 2; only ‘calorie losses’ shown). Notably, a 84 Shared approach dramatically minimizes the extent of spatial trade-offs. Yet regardless of 85 conservation approach, food calorie losses occur in many locations and countries with marked 86 food insecurity. At the country scale, even a Shared approach results in substantial losses of food 87 calories in both China (12%), and India (22%) – countries with the highest absolute numbers of 88 undernourished people on the planet at 133.8M and 194.6M, respectively 13. Under any strategy 89    3 other than Global, substantial calorific losses also occur locally in Africa and elsewhere in Asia 90 (Figure 2). 91  92 As our analysis highlights, there are major political consequences for national food systems 93 embedded within global strategies for achieving Half Earth conservation targets, if the goal of 94 feeding the human world equitably is to be balanced with sharing the planet with other species. 95 Even with current populations and energy demands, potentially massive calorie losses would 96 need to be offset by massive increases in the intensity of food production (i.e. by closing yield 97 gaps to achieve 45-70% increases in crop production14), reductions in food waste (i.e. by 98 mitigating the ~24% of calories lost in global supply chains15) and/or dietary shifts away from 99 animal products (i.e. by diverting the 36% of crop calories currently fed to animals toward direct 100 human consumption16). However, all of these co-strategies are extremely challenging ambitions 101 in themselves.  102  103 Demands of future human populations and losses of pasture-fed animal calories are not 104 considered here, nor did we conduct an explicit spatial prioritization for maximizing coverage of 105 species ranges or connectivity (which may prioritize more productive lands than given back in 106 our analysis), so our findings should be seen as conservative with respect to Half Earth’s 107 potential negative impacts on agriculture. That said, land use conversion from agriculture to 108 conservation (e.g. either from pasture or cropland), would produce significant co-benefits too, 109 such as gains in woody perennial vegetation cover and carbon sequestration, including up to a 110 40% increase in tropical deciduous forest cover globally, and a 36% increase in temperate 111 deciduous forests, no doubt helping to meet climate targets while providing other socially 112 valuable benefits of nature conservation  (Supplementary Information Fig S2)17. 113  114 While our analysis has focused almost exclusively on agricultural production, the practical 115 challenges of achieving Half Earth are many, and go far beyond the metrics discussed here. For 116 example, a key concern is potential worsening of land tenure insecurity and increasing land 117 prices, which together with loss of food security and sovereignty, may further impoverish or 118 displace rural peoples18. These impacts, if not carefully managed, may lead to losses in 119 conservation effectiveness over the long-term. Another important question concerns possible 120 agricultural intensification in the other half of the planet under future population and economic 121 growth19, and how different species and taxonomic groups might respond to this. Finally, while 122 conservation spending (e.g. by governments, trust funds, donors) has proven effective where 123 allocated20, economic costings for global conservation priority areas are too narrow in scope, and 124 the ecological, political, sociocultural, economic and business models needed to sustain a global 125 conservation project of such scale and ambition over decades and into the deep future have yet to 126 be developed. 127  128 We show that any discussion of Half-Earth needs to explicitly evaluate the local, regional, and 129 global consequences of displacing land from agriculture and its consequences for food security. 130 The trade-offs between nature and agriculture are potentially large, but depend strongly on the 131 strategies used to achieve Half-Earth, highlighting the possible strengths of shared landscape 132 approaches to expanding conservation with lower tradeoffs for agricultural production. However, 133 we find no clear pathway to give half our planet to nature at a scale that maintains ecosystem 134 connectivity and still feed the world, without at least some nations or subpopulations losing out. 135    4 Under this context, protection of existing vulnerable, malnourished, and food insecure 136 populations is a key priority, and prerequisite, for humans and nature to coexist into the future. 137  138  139 Corresponding author 140 Correspondence and request for materials should be addressed to: zia.mehrabi@ubc.ca 141  142 Acknowledgements:   143 This work was supported by a Natural Sciences and Engineering Research Council of Canada 144 Discovery Grant and a Genome Canada/Genome BC grant (N.R). 145  146 Contributions:   147 ZM, EE, and NR designed the analysis. ZM compiled the data and conducted the analysis. ZM 148 wrote the paper with input from EE and NR.  149  150 Competing interests:   151 The authors declare no competing interests. 152  153 Figure legends  (Main text): 154 Figure 1. Feeding the world under a global deal for nature. Scenarios were computed using a simple land allocation 155 algorithm operating at global (green), country (blue), and ecoregion (red) scales, each allocating 50% of the planet 156 back to nature. All agricultural area was given back in order of least agricultural productivity, based on the caloric 157 content of 41 major crop plants. The allocation in each scenario was ordered by: (1) existing protected areas, (2) key 158 biodiversity areas, (3) non-agricultural lands (i.e. forests, wetlands, shrublands, grasslands, ice-covered areas), (4) 159 pasture lands, (5) non-food producing croplands (e.g. feed, biofuel, other) (6), food producing croplands (e.g. 160 calories for food crops), and (6) urban areas. Under a “Nature only landscapes” approach we allocated land back to 161 nature on a pixel-by-pixel basis, representing a sharp division between conservation and agriculture at the landscape 162 scale, whereas under a “Shared landscapes” approach we allocated land back to nature on an areal basis, allowing 163 the area within pixels to be shared between conservation and agricultural land uses. All analysis was run at a pixel 164 size of 8.4km x 8.4km.” 165 Figure 2. Mapping calorie losses under each Half-Earth Scenario.  Pixels show the summed caloric losses (from 166 non-food and food crop calories) in each pixel on the planet under three different scales of analysis (global, country, 167 ecoregion) and two different conservation approaches (Nature only landscapes vs. Shared landscapes). Even under a 168 Shared landscapes approach, giving back half of every country or ecoregion on the planet leads to significant losses 169 in geographic locations of major food insecurity concerns (i.e. India, China, Indonesia, W. Africa, Madagascar). 170 Land allocations were made by minimizing calorie losses to show the lowest possible caloric costs to Half-Earth 171 under current production, but we do not mean to advocate that farmers working on marginal lands should be 172 compromised first, as this would likely be deleterious to food security for populations with limited access to farming 173 technology and inputs, or access to land with good growing conditions.  174    5 Figure legends (Methods): 175 Figure 3. Flow-diagram for the analysis. 176  177 Methods: 178 To assess the trade-offs between Half Earth and agricultural production at three scales (Global, 179 Country, Ecoregion), and with two conservation strategies (i.e. “Nature only landscapes” vs. 180 “Shared landscapes” approaches), we compiled 8 spatial datasets: cropland and pasture area 21;   181 MODIS land cover 22; calorie production for the world’s 41 major crop plants16; the world 182 database of protected areas 23; Key Biodiversity Areas 24; Ecoregion boundaries 5; country 183 boundaries 25, and Potential Natural Vegetation 26.  All datasets were rasterized, projected to 184 Eckert IV’s equal area at a spatial resolution of ~8.4km x ~8.4 km (the most accurate scale for 185 reproduction of global sums for data set Ref 16), and set to a common extent. We focused on 186 Ecoregions because Half Earth targets have already been deemed achievable or attained in ~49% 187 of the Earth’s 846 terrestrial Ecoregions5 by previous authors, and who argue that maintaining 188 these geographic boundaries is critical for maintaining global conservation planning, as opposed 189 to focusing conservation exclusively on global biodiversity importance maps which would 190 disproportionately bias the tropics (e.g. Ref 19).   191  192 Our analysis involved additional processing as follows. First, we created a new land class 193 variable merging the above datasets from Refs21-24. To do this we used Refs 23-24 to define cells 194 containing ‘protected areas’ and ‘key biodiversity areas’. Following this we used Ref21  to define 195 cells containing ‘cropland’, ‘pasture’, or ‘cropland-pasture’, and then for the remaining cells 196 used the numeric identifiers in Ref22 to represent ‘water’ (ID:0), non-agricultural land (IDs: 1:9, 197 11, 15:16), and  ‘urban’ (ID:13). We then ranked all pixels for this new land class variable in the 198 following order: protected areas > key biodiversity areas > non-agricultural lands (forests, 199 wetlands, shrublands, savannahs ,grasslands, snow-ice-barren), agricultural lands 200 (pasture>cropland-pasture>cropland) and urban, and then ordered each pixel within each rank in 201 order of increasing calorific production based on data from Ref16 Finally, we estimated the 202 proportion of each pixel that was occupied by non agricultural land, land producing non-food 203 crop calories (feed, biofuel and other), and land producing food crop calories (i.e. for direct 204 human consumption) by using fractional cropland area coverage of Ref21 and proportional 205 allocation to each calorie class using Ref16. We then adjusted the total calorie ranking, to allow 206 for differences in the rank order of non-food calories and food crop calories within a given pixel. 207 All water pixels were set to NAs so as to focus on only terrestrial Half Earth.  208  209 We simulated agricultural losses (cropland area, pasture area, feed/biofuel/other calories and 210 food calories), under Half Earth with the two different conservation approaches and at the three 211 different geographic scales. We assessed agricultural losses, by identifying the proportion of 212 pasture land, cropland, and calories included in the pixels summing to 50% of the total area 213 (“Nature only landscapes” approach), or sub-pixel areas summing to 50% of the total area 214 (“Shared landscapes” approach) within each of the focal geographic boundaries (Global, 215 Country, Ecoregion) included in the analysis. Under the “Shared landscapes” approach we 216 maintained a priority ranking as: protected areas > key biodiversity areas > non-agricultural 217 area>non-food calorie (feed/biofuel/other) producing land & pasture land> food calorie 218    6 producing land> urban, so that we prioritized giving back land area producing non-food calories 219 and pasture lands before food calorie producing croplands. All analysis was done in R 3.4.2 27.  220 A flow diagram with an overview of this analysis is given below (Fig 3).  221  222  223 Data availability 224 Additional analyses and a full set R code to reproduce the results is supplied in Supplementary 225 Information. The data used in this study are either publically available, or available from third 226 parties on request but not distributable by the authors. Full details of each data set, download 227 links, and points of contact for third parties are given in the Supplementary Information. 228  229  230  231  232  233 References: 234  235 1. Butchart, S.H.M. et al. Science 328, 1164–1168 (2010). 236 2.        CBD. Strategic Plan for Biodiversity 2011 – 2020 and the Aichi Targets. 1–2 (2011).  237 3. Butchart, S.H.M., Di Marco, M. & Watson, J.E.M. Conserv. Lett. 9, 457–468 (2016). 238 4. Wilson, E. Half earth : our planet’s fight for life. (Liveright, 2016).  239 5. Dinerstein, E. et al. Bioscience 67, 534–545 (2017). 240 6. Balmford, A. & Green, R. Nature 552, 175 (2017). 241 7. Watson, J.E.M. & Venter, O. Nature 550, 48–49 (2017). 242 8. Blanchard, J.L. et al. Nat. Ecol. Evol. 1, 1240–1249 (2017). 243 9. Naidoo, R. et al. Proc. Natl. Acad. Sci. U. S. A. 105, 9495–9500 (2008). 244 10. Duran, A.P., Duffy, J.P. & Gaston, K.J. Proc. R. Soc. B Biol. Sci. 281, 20141529–245 20141529 (2014). 246 11. Noss, R.F. Bioscience 33, 700–706 (1983). 247 12. Phalan, B., Onial, M., Balmford, A. & Green, R.E. Science 333, 1289–1291 (2011). 248 13. Food and Agriculture Organization of the United Nations. FAOSTAT. (2016).  249 14. Mueller, N.D. et al. Nature 490, 254–257 (2012). 250 15. Kummu, M. et al. Sci. Total Environ. 438, 477–489 (2012). 251 16. Cassidy, E. S. et al. Environ. Res. Lett. 8, 34015 (2013).  252 17. Díaz, S. et al. Science (80). 359, 270–272 (2018). 253 18. Fairhead, J., Leach, M. & Scoones, I. J. Peasant Stud. 39, 237–261 (2012). 254 19. Laura Kehoe et al. Nat. Ecol. Evol. 1, 1129–1135 (2017). 255    7 20. Waldron, A. et al. Nature 551, 364–367 (2017). 256 21. Ramankutty, N., et al.. Global Biogeochem. Cycles 22, 1–19 (2008).  257 22. Friedl, M.A. et al. Remote Sens. Environ. 114, 168–182 (2010). 258 23. UNEP-WCMC & IUCN. The World Database on Protected Areas (WDPA)/The Global 259 Database on Protected Areas Management Effectiveness (GD-PAME)]. (2017). 260 24. BirdLife. World Database of Key Biodiversity Areas. (2017). 261 25. South, A. rworldmap: A New R package for Mapping Global Data. R J. 3, 35–43 (2011).  262 26. Ramankutty, N. & Foley, J.A. Global Biogeochem. Cycles 13, 997–1027 (1999). 263 27. R Core Development Team (2017).doi:ISBN 3-900051-07-0 264  265 Half Earth  LineGlobal Country Ecoregion12% 21% 31%0% 10% 15%0% 4% 10%0% 11% 23%0% 0.4% 3%11% 22% 29%10% 16% 25%21% 37% 45%Half Earth  Linelobal Country Ecoregion12  21  310  10  150  4  10  0  0.4  311  22  29  21  37  45Crop feed, biofuel, other caloriesPasture landCrop landCrop food  caloriesPercent of earth given back Percent agricultural lossShared landscapesNature only landscapesGlobal Country Ecoregion Shared landscapesNature only landscapesTotal crop  calorie  loss (kcal) 0108106104102The challenge of feeding the world while conserving half the planetZia Mehrabi, Erle C. Ellis and Navin Ramankuttyzia.mehrabi@ubc.caAugust 8, 2018Supplementary Information1 Supplementary Notes 31.1 Reproducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Aims of this document . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Supplementary Methods A 42.1 Data preparation overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 Rasterizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2.1 Ecoregions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2.2 KBA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2.3 Crop calories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.4 WDPA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.5 Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 Additional processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3.1 Read data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3.2 Convert to equal area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.3.3 Back corrections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.3.4 Set extents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.5 Create stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4 Land class variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.1 Cropland and pasture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.2 Antarctica . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.3 WDPA and KBA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.4.4 Urban . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.4.5 Add new LC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.4.6 Rank land classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.4.7 Rank by calories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.4.8 Compute areal coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.4.9 Add in Country ids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Supplementary Methods B 113.1 Analysis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.2 Read data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.3 Run scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.4 Main Text Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.4.1 Figure 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.4.2 Figure 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.5 Country calorie losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193.5.1 Ecoregion coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203.6 Co-benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2013.6.1 Nature only landscapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203.6.2 Shared lanscapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 Supplementary Figures 244.0.1 Supplementary Figure 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.0.2 Supplementary Figure 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264.0.3 Supplementary Figure 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2721 Supplementary Notes1.1 ReproducibilityWe use the R packages knitr and checkpoint [25, 8]. The package knitr facilitates producing a dynamicdocument that contains all the steps required to analyze the data. checkpoint() will install all packagesversions for you that we used in our analysis to avoid result discrepencies that may arise from softwaredi↵erences. Thus the reader is provided with all the code to fully reproduce the analysis, and adapt orrepurpose it for another analyses.library("checkpoint")checkpoint(snapshotDate = "2018-03-01")For the analysis in this document we will be using the raster [13], rgdal [5], ggplot2 [22], plyr [20], sp[14], R.matlab [3], rworldmap [17], car [11], maptools[6] , reshape2[21] ,gridExtra[1],amd ,cowplot[23],and, data.table[10] packages.packages<-c("raster", "rgdal", "ggplot2", "plyr", "sp","R.matlab", "rworldmap", "car", "maptools","knitr", "reshape2","gridExtra","cowplot", "data.table")lapply(packages, require, character=TRUE)gpclibPermit()1.2 Aims of this documentThe aim of this document is to provide the code to reproduce the analysis presented in Mehrabi, Ellis andRamankutty 2018 entitled “The challenge of feeding the world while conserving half the planet”. Half-Earthis a recent proposal to devote half the surface of the Earth to nature. (http://www.half-earthproject.org/;http://natureneedshalf.org/). The underlying rationale for Half-Earth derives from scaling relationshipsbetween habitat area and species numbers which imply that conserving half of Earth’s habitat area wouldsave 85% of existing species on the planet [24]. The practical costs of Half-Earth incurred through trade-o↵swith other land uses, such as agriculture, and its impacts on already disadvantaged populations around theworld still remain poorly understood [2]. This analysis aims to help fill this research gap.The nuts and bolts of this analysis is a simple thought experiment constructed to better understand someof the broad trade-o↵s between agriculture and Half-Earth proposal. The document is structured into twosections, in section 2 we process the data, in section 3 we conduct the thought experiment, create the plotsused in the figure in the main text of the manuscript, and conduct some additional analysis, and finallyin section 4 we plot the Supplementary Figures. For brevity the code in section 2 is not evaluated in thisdocument.32 Supplementary Methods A2.1 Data preparation overviewTo assess the trade-o↵s between Half-Earth and agriculture we compiled 8 spatial datasets. Details of eachand the currently live urls and points of contact for data set acquisition are listed below. In some cases thesedata are not available for us to redistribute, and so those wishing to reproduce this analysis are recommendedto approach the persons responsible, and back-date to the dates of acquisition listed below.1. Global Cropland and Pasture Area [16], retrieved online from www.earthstat.org , on December 13th2017.2. MODIS land cover [12], retrieved online from http://glcf.umd.edu/data/lc/, on December 17th 2017.3. Calorie production for the worlds 41 major crop plants,[7] retrieved directly from the original authors,available from Minnesota Insitute on the Environment (jsgerber@umn.edu)4. The World Database of Protected Areas, [19], retrieved online from https://www.protectedplanet.net/,on December 13th 2017.5. Key Biodiversity Areas 2017 [4] retrieved via FTP link from Gill Bunting at Bird Life International(Gill.Bunting@birdlife.org) on Nov 6th 2017.6. Ecoregion boundaries,[9], retrieved online from https://storage.googleapis.com/teow2016/Ecoregions2017.zip,on April 20177. Country boundaries [17], retrieved internally in R.8. Potential Natural Vegetation, [15], retrieved online from on April 30th 2018, from www.earthstat.org.All datasets were rasterized, projected to Eckert IVs equal area at a spatial resolution of 8.4km x 8.4 km(the most accurate scale for reproduction of global sums for data set [7], and set to a common extent). Thissection explains that processing.2.2 Rasterizing2.2.1 EcoregionsFirst we read in the ecoregion data, rasterize it, and write to file.Dinn<-shapefile("halfearthdata/Ecoregion2017_Erle/Ecoregions2017.shp")raster.Dinn<-raster(resolution = 0.0833282)extent(raster.Dinn) <- extent(Dinn)Dinn$ECO_NAME<-as.factor(Dinn$ECO_NAME)Dinn.raster <- rasterize(as(Dinn, "SpatialPolygons"), raster.Dinn, field = Dinn@data[,"ECO_NAME"],fun="first")getwd()Dinn.rast <- writeRaster(Dinn.raster, "halfearthdata/processed/Dinnrast.tif", format="GTiff", overwrite=TRUE)2.2.2 KBAThen we read in the KBA data, rasterize it, and write to file.KBA<-shapefile("KBAsGlobal_2017_1a/Global_KBA_poly.shp")raster.KBA<-raster(res = c(0.0833282,0.0833282))extent(raster.KBA) <- extent(KBA)KBA.raster <- rasterize(as(KBA, "SpatialPolygons"), raster.KBA, field = KBA@data[,"SITRECID"],fun="first")kbavals<-values(KBA) #set non KBA values to NAkbavals2<-ifelse(kbavals>0,1,NA)KBA2<-KBAvalues(KBA2)<-kbavals2KBA.rast<- writeRaster(KBA2, "halfearthdata/processed/KBArast.tif", format="GTiff", overwrite=TRUE)42.2.3 Crop caloriesThen we read in the global crop calorie data, and write to file. These data are supplied as .mat files.Cass.total<-readMat("halfearthdata/Emily_data/Cassidy Crop Allocation and Delivery/GlobalTotalkcal.mat")rasterD <-raster(ncol=4320, nrow=2160)values(rasterD)<-c(Cass.total$GlobalTotalkcal)kcal.41<-rasterDplot(kcal.41)kcal.rast<- writeRaster(kcal.41, "halfearthdata/processed/kcalrast.tif", format="GTiff", overwrite=TRUE)We also retrieve the calorie delivery fraction data from Cassidy et al. A unit of 1 identifies that all kcalproduced are delivered as food, wheras a unit of <1 identifies that some proportion of calories are fed toanimals from that pixel.Cass.food<-readMat("halfearthdata/Emily_data/Cassidy Crop Allocation and Delivery/GlobalFoodkcal.mat")rasterD <-raster(ncol=4320, nrow=2160)# Cass.food£GlobalFoodkcal[is.na(Cass.food£GlobalFoodkcal)] <- 0 #replace NA with zerovalues(rasterD)<-c(Cass.food$GlobalFoodkcal)kcal.food<-rasterDkcal.food<- writeRaster(kcal.food, "halfearthdata/processed/kcal.food.tif", format="GTiff", overwrite=TRUE)2.2.4 WDPAHere we read in the world protected area data. This data was first imported into QGIS [18], to simplify thegeometry with the default tolerance. It was then rasterized it in QGIS, to reduce processing time in R. Weset non-WDPA values in this raster to NA, and write the result to file.WDPA<-raster("halfearthdata/WDPA_Oct2017/WDPA.tif")str(WDPA)wdpavals<-values(WDPA)#set non WDPA values to NAwdpavals2<-ifelse(wdpavals>0,1,NA)WDPA2<-WDPAvalues(WDPA2)<-wdpavals2WDPA.rast<- writeRaster(WDPA2, "halfearthdata/processed/WDPArast.tif", format="GTiff", overwrite=TRUE)2.2.5 CountriesFinally we make a raster of the world country data from the rworldmap package.world<-getMap(resolution='low')raster.world<-raster(res = c(0.0833282,0.0833282))extent(raster.world) <- extent(world)world.raster <- rasterize(as(world, "SpatialPolygons"), raster.world, field = world@data[,"ADMIN"],fun="first")world.rast<- writeRaster(world.raster, "halfearthdata/processed/worldrast.tif", format="GTiff", overwrite=TRUE)2.3 Additional processing2.3.1 Read dataHere we read in the data for additional processing. We read in the processed data sets outlined above, as wellas two others. These include the MODIS LC data set as an indication of land cover, and potential naturalvegetation.KBA<-raster("halfearthdata/processed/KBArast.tif")CA<-raster("halfearthdata/CroplandPastureArea2000_Geotiff/cropland2000_area.tif")PA<-raster("halfearthdata/CroplandPastureArea2000_Geotiff/pasture2000_area.tif")Dinn<-raster("halfearthdata/processed/Dinnrast.tif")kcal<-raster("halfearthdata/processed/kcalrast.tif")kcal.food<-raster("halfearthdata/processed/kcal.food.tif")WDPA<-raster("halfearthdata/processed/WDPArast.tif")world<-raster("halfearthdata/processed/worldrast.tif")LC<-raster("halfearthdata/MODIS_GLCF/LC_hd_global_2012.tif")PNV<-raster("halfearthdata/potveg_geotiff/potentialvegetation_geotiff/potentialvegetation.tif")52.3.2 Convert to equal areaHere we project everything to be equal area (set to 8439m x 8439m, as this is with the best match for thekcal data, it is within 0.001%). We use two methods of interpolation, bilinear for continuous data and nearestneighbour for categorical data.rast.list.ngb<-list(Dinn,KBA, WDPA, LC, world, PNV)rast.list.bil<-list(kcal,CA,PA, kcal.food)rast.list.ngb.eq<-lapply(rast.list.ngb, function(x)projectRaster(x, res=c(8439, 8439),crs="+proj=eck4 +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0",method="ngb", over = T))rast.list.bil.eq<-lapply(rast.list.bil, function(x)projectRaster(x, res=c(8439, 8439),crs="+proj=eck4 +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0",method="bilinear", over = TRUE))rast.list<-c(rast.list.ngb.eq, rast.list.bil.eq) #combine datanames(rast.list)<-c("Dinn","KBA", "WDPA", "LC", "world", "PNV","kcal","CA", "PA","kcal.food")Now we double check that the sum of the original crop calorie data is equivalent to the equal area represen-tation. We find these are very close to each other, and so the equal area conversion is reasonable in aggregatefor this variable.sum(values(rast.list.bil.eq[[1]]), na.rm = T) #check kcal valuessum(values(kcal), na.rm = T)2.3.3 Back correctionsNo reprojection of geospatial data is perfect. In some cases it can lead to nonsensical outputs. A goodexample of this is shown here with cropland and pasture data, where the sum of the original lat-long data onproportions of cropland and pasture in a cell is never greater than zero, but where for the derived productsthe summed area can exceed one. This is logically impossible in reality, but is an artifact created throughthe reprojection.plot(CA+PA) #original lat longplot(rast.list$PA+rast.list$CA) #eck 4 versionThe details of the net di↵erences between the aggregate sums of equal area projected and lat-long versionare given below. As we can see reprojection reduced the global area of cropland by 2% and increased it forpasture by 1%.cell.area=8439*8439 #area of grid cellsum(values(rast.list$CA)*cell.area, na.rm=T)/sum(values(CA)*cell.area, na.rm=T)sum(values(rast.list$PA)*cell.area, na.rm=T)/sum(values(PA)*cell.area, na.rm=T)It is dicult to deal with these two issues easily in one fail swoop. It is possible to calibrate using the originaldata sources pretty easily. This is done in the chunk of code below for continuous data. But this does notdeal with the nonsensical output of having proportional coverage in pixels greater than one.values(rast.list$CA)<-(values(rast.list$CA)/sum(values(rast.list$CA),na.rm=T))*sum(values(CA), na.rm=T)#CAvalues(rast.list$PA)<-(values(rast.list$PA)/sum(values(rast.list$PA),na.rm=T))*sum(values(PA), na.rm=T) #PAvalues(rast.list$kcal)<-(values(rast.list$kcal)/sum(values(rast.list$kcal),na.rm=T))*sum(values(kcal), na.rm=T) #food kcalvalues(rast.list$kcal.food)<-(values(rast.list$kcal.food)/sum(values(rast.list$kcal.food),na.rm=T))*sum(values(kcal.food), na.rm=T) #kcalOne work around, is to ensure that no sum of proportional areas in pixels exceeds one. We do this below, andpreferentially reduce pasture areas within each pixel. A more exact approach, would be to write a spatial re-allocation algorithm that deals with both of these issues together. However, for the purposes of this thoughtexperiment, and for the specific results we provide, the approach taken here seems reasonable. Indeed, aswe can see from below, not only have we now achieved sensible plot values, but we still have maintained ourpasture aggregate sums to a high degree – with our reprojected pasture areas being only 0.02% less than theoriginal lat long file. By focusing only on pasture in correcting for proportional coverage, the aggregate sumsof all other continuous variables is maintained.6values(rast.list$PA)<-ifelse(values(rast.list$PA)+values(rast.list$CA)>1,1-values(rast.list$CA), values(rast.list$PA))plot(rast.list$PA+rast.list$CA)sum(values(rast.list$PA)*cell.area, na.rm=T)/sum(values(PA)*cell.area, na.rm=T)2.3.4 Set extentsNext we need to trim the extents. We can see from the plots below, and in the README for the MODIS LCdata, that the LC data does not include Antarctica. In fact it only covers southern latitudes to -64 degrees.par(mfrow = c(1,1))lapply(rast.list, function(x) plot(x, main=names(x))) #checkTherefore, we need to expand the LC data to include the area that will house Antarctica. We draw from thecountry data which includes a boundary for Antarctica. As explained by Prof. Rob Hijmans in his manualfor the package (http://rspatial.org/spatial/rst/8-rastermanip.html), the extend function fills NA’s in rowsthat are added. We couple this with the crop function to maintain equal extents across all datasets, thenapply the setExtent function to ensure that rounding is exactly the same across the files.ex1<-extent(rast.list$Dinn)rast.list.ex<-lapply(rast.list, function(x) extend(x, ex1)) #extendrast.list.c<-lapply(rast.list.ex, function(x) crop(x, ex1)) #then croprast.list.et<-lapply(rast.list.c, function(x) setExtent(x, ex1, keepres=TRUE)) #then force extent to be the same2.3.5 Create stackWith the data all with the same resolution, extent and equal area CRS, we can now create a stack of all thelayers.rast.all<-stack(rast.list.et)Now let’s check the plots again.plot(rast.all) #check2.4 Land class variable2.4.1 Cropland and pastureWe now need to make a common land class variable that merges cropland and pasture area data and MODISLC data. We will use the Ramankutty et al. identifiers for cropland and pasture data.We create a numeric code for each cropland, pasture and cropland-pasture mix and add this to the land classvariable. For cropland we assign 18, for the pasture-cropland mix we assign 17 , and for pasture we assign19.newclass<-ifelse(values(rast.all$CA) >0 & values(rast.all$PA) >0,17,ifelse(values(rast.all$CA) > 0 & values(rast.all$PA) %in% c(0,NA) ,18,ifelse(values(rast.all$PA) >0 & values(rast.all$CA) %in% c(0,NA) , 19,values(rast.all$LC))))summary(as.factor(newclass)) #check2.4.2 AntarcticaWe need to specify Antarctica in the LC data .Antarctica is identified under an id of 9 in the country data.We therefore designate Antarctica an id of 15, to correspond to the id for ”Rock and Ice” in the READMEgiven to accompany the LC file. We then check this by plotting, it looks good. We then add this to thenewclass variable. [We add this in here because doing it before defining cropland, mixed and pasture pixels,would lead to it being lost due to NAs in CA and PA data sets].7values(rast.all$LC)<-ifelse(values(rast.all$world)==9, 15, values(rast.all$LC))plot(rast.all$LC) #check Antarctica has been addednewclass<-ifelse(values(rast.all$LC)==15, 15, newclass) #add to newclass2.4.3 WDPA and KBAWe then assign numeric values for the WDPA and KBA areas: 20, and 21 respectively. These may includeagricultural production, but we will give these areas top priority in any Half-Earth Scenario for conservation.We keep the ocean as zero.newclass<-ifelse(is.na(values(rast.all$KBA))|values(rast.all$LC)==0, newclass, 21)newclass<-ifelse(is.na(values(rast.all$WDPA))|values(rast.all$LC)==0, newclass, 20)2.4.4 UrbanWe also maintain urban areas with agricultural production as urban.newclass<-ifelse(values(rast.all$LC)==13, 13, newclass)2.4.5 Add new LCWe then create a new layer in the stack for the new land class variablerast.all$newclass<-rast.all$LCvalues(rast.all$newclass)<-as.numeric(newclass)summary(values(rast.all$newclass)) #checkAnd then join the data into a single data frame.data.all<-data.frame(coordinates(rast.all), values(rast.all))summary(as.factor(data.all$newclass))2.4.6 Rank land classesFinally we recode and rank the land cover variables into a more limited number of classes. The rank representsthe order in which we will give back individual pixels (“nature only landscape approach”), or area withinpixels (“Shared landscape approach”). All agricultural areas are given back in order of least productivity.We define agricultural productivity based on calories gained from 41 major arable crops. The ranking forthis hypothetical deal with nature is given below. The order of this ranking is not supposed to be definitive,but simply to highlight broad scale the trade-o↵s between Half-Earth and agricultural production. We assignwater to be NA.1. World protected areas2. Key Biodiversity Areas3. Non-agricultural lands (Forests, Wetlands, Shrublands and savannas, Grasslands, Snow, Ice, Barren)4. Agricultural lands (Pasture)5. Agricultural lands (Mixed)6. Agricultural lands (Cropland)7. Urban8data.all$rank<-recode(data.all$newclass," c('0')= NA;c('1','2','3','4','5','6', '7','8','9' ,'10', '11','12', '14','15', '16') = '3';c('13')= '7';c('17')= '6';c('18')= '5';c('19')= '4';c('20')= '1';c('21')= '2'")summary(as.factor(data.all$rank)) #check2.4.7 Rank by caloriesWe then need to compute the calories in each pixel. We compute this in a similar step-wise process to theareas, for non crop productive locations (i.e. a value of zero), and for crop areas [split into non-food calories,i.e. feed and bio fuel and other, and food calories].data.all$kcal.noag<-(0) #zero kcal for all non-ag areasdata.all$kcal.nonfood<- data.all$kcal-data.all$kcal.fooddata.all$kcal.prop<-data.all$kcal.food/data.all$kcalhead(subset(data.all, kcal.prop>0))We then join the calories column with the rank column. This orders the agricultural productivity, and othervariables by the calories, but keeps the ranking of the land use classes. We generate three di↵erent calorierankings, one for food, and one for non-food calories, and one for all sources of calories; marking them bydi↵erent orders of magnitude.data.all$rank.all<-(data.all$kcal/max(data.all$kcal, na.rm=T)+data.all$rank)data.all$rank.nonfood<-(data.all$kcal.nonfood/max(data.all$kcal.nonfood, na.rm=T)+data.all$rank)+10data.all$rank.food<-(data.all$kcal.food/max(data.all$kcal.food, na.rm=T)+data.all$rank)+100head(data.all[which(data.all$kcal>0),])2.4.8 Compute areal coverageNext we compute the non-agricultural area in each grid cell, and the area allocated to non food calories (feed,bio fuels, other) vs. food calorie production. First we check to see if there are any cases with calorie datawhere our cropland area product is 0. There are a few cases here, but they only represent a very small amountof global kcal. This is an extremely minor part of the data, and likely results from processing error.sub<-subset(data.all, kcal>0 & (CA==0))#cells with kcal but no crop f. area datasum(sub$kcal, na.rm=T)/sum(data.all$kcal, na.rm=T) #Then we define the variables that we will need for later analysis, and sanity check these against those reportedin Ramankutty et al. 2008. We also compute the non-crop area, feed/bio fuel/other area, and food areawithin each pixel in each ranked class. We disaggregate non-food vs food areas using the calorie fractiondata, on the assumption that areal allocation is equal per unit of calories for a given pixel.area<-cell.area*nrow(subset(data.all, !is.na(rank))) #terrestrial areaice.free<- area- (cell.area*nrow(subset(data.all, LC==15))) #ice -free terrestrial areacrop.area<-data.all$CA*cell.area #crop areapasture.area<-data.all$PA*cell.area #crop areacrop.area[is.na(crop.area)] <- 0 #remove NAspasture.area[is.na(pasture.area)] <- 0 #remove NAsnon.ag<-ifelse(!is.na(data.all$rank), cell.area-crop.area-pasture.area, NA) #non.agfood.area<-crop.area*data.all$kcal.prop #food areafeed.area<-crop.area-food.area #feed/biofuel areafood.area[is.na(food.area)] <- 0 #remove NAsfeed.area[is.na(feed.area)] <- 0 #remove NAsnon.ag[is.na(other.ca.area)] <- 0 #remove NAssum(crop.area, na.rm=T)/ice.free #sanity checks, should be ~12\%,9sum(pasture.area, na.rm=T)/ice.free #sanity checks, should be ~22\% ,sum(crop.area,pasture.area, na.rm=T)/ice.free #sanity checks, should be ~34\%,sum(non.ag,crop.area,pasture.area, na.rm=T)/areaWe then assign the relevant area calculations to the data frame. For all land classes except urban, first wegive back non-agricultural area, then non-food (feed, bio fuel, and other ) crop area alongside pasture area,then finally agricultural areas producing food calories. For urban locations we reverse the grouping – first wegive back non-food (feed, bio fuel, other) agricultural area, then food crop areas and finally, non-agriculturalareas. Notably, we simply embed this reversion for urban pixels in the data, to account for logic of highpriority for avoiding encroaching on urban settlements, but it is not reflected in the variable names.data.all$food.area<- ifelse(data.all$rank==5,food.area+non.ag, food.area)data.all$feed.area<-feed.area+pasture.areadata.all$non.ag<-ifelse(data.all$rank==5,0, non.ag)sum(crop.area, pasture.area, na.rm=T)/ice.free #sanity checksum(data.all$non.ag,data.all$feed.area,data.all$food.area, na.rm=T)/area #sanity checksum(non.ag, crop.area,pasture.area, na.rm=T)/area #sanity checkdata.all$past.area<-pasture.area #add in pasture area for later2.4.9 Add in Country idsDuring processing we lost the country identifiers, so here we add that back in now, both in ADMIN names,and ISO3’s.world<-getMap(resolution='low')lookup<-as.data.frame(cbind(as.character(world@data[,'ISO3']),world@data[,'ADMIN'],as.character(world@data[,'ADMIN'])))matched<-lookup[match(data.all$world, lookup$V2), ]data.all$country<-matched[[1]]data.all$ISO3<-matched[[3]]Finally we save the processed data as an RDS file.saveRDS(data.all, "halfearthdata/processed/data.all.new.rds")103 Supplementary Methods B3.1 Analysis overviewIn this section we conduct the analysis underlying the results presented in the manuscript. To simulatecrop calorie losses under Half-Earth, we gave back pixels (i.e. “nature only landscapes”), or area in pixels(“shared landscapes”), on the planet in sequence of the rank and increasing calorie production, at global (noboundary), country (boundaries set by countries), and ecoregion (boundaries set by Ecoregions) scales. Weassess losses at 50% land re-allocation under each scenario.3.2 Read dataFirst we read in data.data.all<-readRDS("halfearthdata/processed/data.all.new.rds")Let’s double check the number of countries and ecoregions.geocount<-c(length(unique(data.all$Dinn)),length(unique(data.all$ISO3)))geocount[1] 821 231Because we want consistency in the number of cells, and area, given back under each scenario, and becauseeach of the polygon boundaries for the ecoregions, countries and the land cover datasets are not exactly thesame, we ensure NAs across the datasets are consistent. In addition we remove the NAs in the land coverdata which represent water bodies, so we are working only on the terrestrial land surface area from here onin.data.all<-subset(data.all, !is.na(Dinn) & !is.na(country) & !is.na(rank))Now let’s check how many countries and ecoregions we are working with now.subgeocount<-c(length(unique(data.all$Dinn)),length(unique(data.all$ISO3)))subgeocount[1] 775 1823.3 Run scenariosWe then write out a loop to compute the half-earth scenario for each geographical scale of analysis. Allthe processing is included as one chunk of code here, but is hopefully laid out in a logical way that can berefactored if needed in future. It takes the following inputs:1. geog = column name with the geographic identifier (i.e. indicating scale of analysis)2. data= dataset (i.e. data.all)3. scen = character label for the scenario being run11And then creates a list output of length 4. The first and second elements are the output for the “natureonly landscapes” approach to Half-Earth. The third and fourth are the outputs for the “shared landscapes”approach. The first, and third, elements include all original variables in the data.all data.frame, alongsidea range of new ones. As outlined below, these list elements and the new variables in them can be used toidentify the exact pixels given back under any Half-Earth scenario (it includes a variable called below.midwith 1’s to identify these cells), and for summarizing kcal losses by ecoregion and country. The second andfourth elements of this list are dataframes with the following three columns summarizing losses under eachscenario:1. perc.world = cumulative sum of terrestrial land area given back2. kcal = kcal cumulative of the kcal lost in cells given back3. perc.kcal = cumulative sum of the percent of kcal given backEach of the lists for each scale of analysis are saved to their own RDS file. Before we run this analysis wefirst call a function for computing cumulative sums while ignoring NAs. We also re-call the cell area.cum.na <- function(x) { #cumulative sum, ignore NA functionx[which(is.na(x))] <- 0return(cumsum(x))}cell.area <-8439*8439 #define cell areaAnd now we run the scenarios.#############################SETUP#####################################geog.n<-list(1,as.factor(data.all$Dinn), as.factor(data.all$ISO3))#geographic idsscen<-c("global", "ecoregion", "country") #scenariosgeog.n<-list(1)#geographic idsscen<-c("global") #scenariosfor(i in 1:length(geog.n)){data<-data.allgeog<-geog.n[[i]]dat.split<-split(data,geog) #split data#############################NATURE ONLY APPROACH##########################ORDER CELLS BY RANK#dat.split.s<-lapply(dat.split, function(x) x[order(x$rank.all),])#IDENTIFY GRID CELLS TO GIVE BACK#dat.split.s<-dat.split.s[lapply(dat.split.s,nrow)>0] #remove locations with no datadat.split.s<-lapply(dat.split.s, function(x) cbind(x, 1/nrow(x))) #assign numbersdat.split.s<-lapply(dat.split.s, function(x) cbind(x, round(cumsum(x$`1/nrow(x)`), digits=2))) #add in the cumulative percentage for each.names<-c(colnames(dat.split.s[[1]])[1:26], "id", "perc.area") #reset namesdat.split.s<-lapply(dat.split.s, setNames, names)dat.split.s<-lapply(dat.split.s, function(x) cbind(x, ifelse(x$perc.area<0.51, 1,0)))dat.split.all.short<-rbindlist(dat.split.s) #use rbindList, quicker#COMPUTE THE PERCENT LOSSES BY AREA ACROSS GEOG UNITS#dat.all.sum.short<-ddply(dat.split.all.short,"perc.area", summarize,kcal.sum=sum(kcal, na.rm=T),PA.sum=sum(PA*cell.area, na.rm=T),CA.sum=sum(CA*cell.area, na.rm=T),kcal.food.sum=sum(kcal.food, na.rm=T),kcal.feed.sum=sum(kcal.nonfood, na.rm=T)) #summarize the kcal by percent area, take timedat.all.sum.short$perc.kcal<-cumsum(dat.all.sum.short$kcal.sum)/max(cumsum(dat.all.sum.short$kcal.sum))dat.all.sum.short$perc.CA<-cumsum(dat.all.sum.short$CA.sum)/max(cumsum(dat.all.sum.short$CA.sum))dat.all.sum.short$perc.PA<-cumsum(dat.all.sum.short$PA.sum)/max(cumsum(dat.all.sum.short$PA.sum))dat.all.sum.short$perc.kcal.non<-cumsum(dat.all.sum.short$kcal.feed.sum)/max(cumsum(dat.all.sum.short$kcal.feed.sum))dat.all.sum.short$perc.kcal.food<-cumsum(dat.all.sum.short$kcal.food.sum)/max(cumsum(dat.all.sum.short$kcal.food.sum))12#############################SHARED APPROACH################################MAKE DATA LONG#vars<-setdiff(colnames((dat.split[[1]])), c("feed.area", "food.area", "non.ag"))dat.split.l <-lapply(dat.split, function(x) melt(x, id.vars = c(vars), value.name="area", variable.name="Area.type"))head(dat.split.l[[1]])#DEFINE NEW RANK AND OUTCOME VARIABLES#dat.split.l<-lapply(dat.split.l , function(x) cbind(x, ifelse(x$Area.type=="non.ag",x$rank.all,ifelse(x$Area.type=="feed.area", x$rank.nonfood, x$rank.food)), #new rank columnifelse(x$Area.type=="non.ag", 0,ifelse(x$Area.type=="feed.area", x$kcal.nonfood, x$kcal.food)), #new kcalifelse(x$Area.type=="feed.area", x$kcal.nonfood,0), #new kcal.nonfoodifelse(x$Area.type=="food.area", x$kcal.food,0), #new kcal. foodifelse(x$Area.type=="non.ag", 0, x$CA*cell.area), #new CA columnifelse(x$Area.type=="non.ag", 0,x$past.area))) #new PA columnnames<-c(colnames(dat.split.l[[1]])[1:25], "rank.new","kcal.new", "kcal.new.feed", "kcal.new.food","CA.n", "PA.n") #reset namesdat.split.l<-lapply(dat.split.l, setNames, names)#ORDER CELLS BY NEW RANK#dat.split.l<-lapply(dat.split.l, function(x) x[order(x$rank.new),])#IDENTIFY GRID CELLS TO GIVE BACK#dat.split.l<-lapply(dat.split.l , function(x) cbind(x,cum.na(x$area), #cumsum arearep(nrow(x)*cell.area*(1/3), nrow(x)))) #total areadat.split.l<-lapply(dat.split.l , function(x)cbind(x, round(x$`cum.na(x$area)`/(x$`rep(nrow(x) * cell.area * (1/3), nrow(x))`), digits=2),ifelse(x$`cum.na(x$area)`<= (x$`rep(nrow(x) * cell.area * (1/3), nrow(x))`)/2, 1, 0))) #area percentnames<-c(colnames(dat.split.l[[1]])[1:31], "cum.sum.area","tot.area", "perc.area", "below.med") #set namesdat.split.l<-lapply(dat.split.l, setNames, names)dat.split.all<-rbindlist(dat.split.l) #use rbindList, quicker#COMPUTE THE PERCENT LOSSES BY AREA ACROSS GEOG UNITS#dat.all.sum<-ddply(dat.split.all,"perc.area", summarize,kcal.sum=sum(kcal.new, na.rm=T),PA.sum=sum(PA.n, na.rm=T),CA.sum=sum(CA.n, na.rm=T),kcal.food.sum=sum(kcal.new.food, na.rm=T),kcal.feed.sum=sum(kcal.new.feed, na.rm=T)) #summarize the kcal by percent area, take timedat.all.sum$perc.kcal<-cumsum(dat.all.sum$kcal.sum)/max(cumsum(dat.all.sum$kcal.sum))dat.all.sum$perc.CA<-cumsum(dat.all.sum$CA.sum)/max(cumsum(dat.all.sum$CA.sum))dat.all.sum$perc.PA<-cumsum(dat.all.sum$PA.sum)/max(cumsum(dat.all.sum$PA.sum))dat.all.sum$perc.kcal.non<-cumsum(dat.all.sum$kcal.feed.sum)/max(cumsum(dat.all.sum$kcal.feed.sum))dat.all.sum$perc.kcal.food<-cumsum(dat.all.sum$kcal.food.sum)/max(cumsum(dat.all.sum$kcal.food.sum))#############################SAVE RESULT####################################return<- list(dat.split.all.short, dat.all.sum.short, dat.split.all, dat.all.sum)path<-paste("halfearthdata/processed/",scen[i],"processed.rds", sep="")saveRDS(return, path)}133.4 Main Text FiguresFirst we read in the data processed in the previous subsection.global<-readRDS("halfearthdata/processed/globalprocessed.rds")ecoregion<-readRDS("halfearthdata/processed/ecoregionprocessed.rds")country<-readRDS("halfearthdata/processed/countryprocessed.rds")3.4.1 Figure 1We then join all of the data together for plotting. Here we are interested in the results for both nature onlylandscapes and shared landscapes approaches, and comparing the summary results for each of these. To dothis we, we need to call the second, and fourth elements of each of the lists read in above.to.plot<-rbind(global[[2]], country[[2]], ecoregion[[2]],global[[4]], country[[4]], ecoregion[[4]])to.plot$Scenario<-c(rep("Global.NatOnly", nrow(global[[2]])),rep("Country.NatOnly", nrow(country[[2]])),rep("Ecoregion.NatOnly", nrow( ecoregion[[2]])),rep("Global.Shared", nrow(global[[4]])),rep("Country.Shared", nrow(country[[4]])),rep("Ecoregion.Shared", nrow( ecoregion[[4]])))to.plot$Scenario1<-c(rep("Global", nrow(global[[2]])),rep("Country", nrow(country[[2]])),rep("Ecoregion", nrow( ecoregion[[2]])),rep("Global", nrow(global[[4]])), rep("Country", nrow(country[[4]])),rep("Ecoregion", nrow( ecoregion[[4]])))to.plot$Scenario2<-c(rep("Nature Only", nrow(global[[2]])),rep("Nature Only", nrow(country[[2]])),rep("Nature Only", nrow( ecoregion[[2]])),rep("Shared", nrow(global[[4]])),rep("Shared", nrow(country[[4]])),rep("Shared", nrow( ecoregion[[4]])))Before we plot we print the aggregated losses for each scenario below. These are the main results reportedin the manuscript.to.plot.sub<-subset(to.plot,perc.area<0.501)ddply(to.plot.sub, c("Scenario1" ,"Scenario2"), summarize,kcal.loss.feed=max(perc.kcal.non)*100,kcal.loss.food=max(perc.kcal.food)*100,CA.loss=max(perc.CA )*100,PA.loss=max(perc.PA )*100)Scenario1 Scenario2 kcal.loss.feed kcal.loss.food CA.loss PA.loss1 Country Nature Only 15.7 22.09 21 37.42 Country Shared 11.0 0.39 10 4.13 Ecoregion Nature Only 24.5 28.80 31 45.44 Ecoregion Shared 22.7 2.88 15 10.35 Global Nature Only 9.7 11.24 12 21.16 Global Shared 0.0 0.00 0 0.0And then we make this summary data long for plotting, and make the variable names intelligible.vars<-setdiff(colnames(to.plot), c("perc.kcal.food", "perc.kcal.non", "perc.PA","perc.CA"))to.plot.l <- melt(to.plot, id.vars = c(vars), value.name="loss", variable.name="Loss.type") #make data longto.plot.l$Loss.type<-recode(to.plot.l$Loss.type, #make var names intelligable" c('perc.kcal.food')= 'D. Food kcal';c('perc.kcal.non') = 'C. Feed, biofuel, and other kcal';c('perc.PA')= 'B. Pasture land';c('perc.CA')= 'A. Crop land'" )We then plot out the di↵erent scenarios. These plots form the basis for figure 1 given in the main text of ourmanuscript.14ggplot(to.plot.l, aes(perc.area*100,loss*100, color=Scenario1))+geom_vline(xintercept=50, lty=2, colour="black")+geom_line()+scale_colour_manual(values = c("#56B4E9","#E75E00", "#009E73", "gray"))+theme_bw()+theme(plot.background = element_blank(),panel.grid.major = element_blank(),panel.grid.minor = element_blank(),# strip.text.x = element_blank(),strip.background = element_rect(colour = "white", fill = "white"),legend.key = element_rect(fill = "white", colour = "white"))+ylab("Percent agricultural loss")+xlab("Percent of earth given back")+# ylab("")+# xlab("")+facet_wrap(~Loss.type*Scenario2, nrow=4, ncol=2) + theme(legend.position="none")15D. Food kcalNature OnlyD. Food kcalSharedC. Feed, biofuel, and other kcalNature OnlyC. Feed, biofuel, and other kcalSharedB. Pasture landNature OnlyB. Pasture landSharedA. Crop landNature OnlyA. Crop landShared0 25 50 75 100 0 25 50 75 1000255075100025507510002550751000255075100Percent of earth given backPercent agricultural lossFigure 1: Feeding the world under a global deal for nature. The horizontal dashed line shows the Half-Earthline. Left side: Nature only landscapes, Right side: Shared landscapes. green= global, blue= country, red=ecoregion.163.4.2 Figure 2Here we map out the scenarios. First we need to identify which pixels were given back under each scenario,and the calories lost in each. We do this to visualize where the calorie trade-o↵s are locally. We isolate thedata needed for plotting.global.NatOnly<-subset(global[[1]], `ifelse(x$perc.area < 0.51, 1, 0)`==1)country.NatOnly<- subset(country[[1]], `ifelse(x$perc.area < 0.51, 1, 0)`==1)ecoregion.NatOnly<-subset(ecoregion[[1]], `ifelse(x$perc.area < 0.51, 1, 0)`==1)global.Sharing<-subset(global[[3]], below.med==1)country.Sharing<- subset(country[[3]], below.med==1)ecoregion.Sharing<-subset(ecoregion[[3]], below.med==1)kcal.max<-round(max(log10((global.NatOnly$kcal/1000)+1), na.rm=T), digits=1)kcal.min<-round(min(log10((global.NatOnly$kcal/1000)+1), na.rm=T), digits=1)Then we write a function to map the kcal lost for each scenario in each location given back. We choose ablue colour gradient to aid discernment of logged steps in the scale.breaks=c(0:9)map.scenarios<-function(data, kcal){df<-data.frame(data$x, data$y,kcal)dfr<- rasterFromXYZ(df) #Convert first two columns as lon-lat and third as valuecrs(dfr)<-"+proj=eck4 +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0"values(dfr)<-log10((values(dfr)/1000)+1)colfunc <- colorRampPalette(c("gray95", "#edf8b1", "#c7e9b4","#7fcdbb", "#41b6c4","#1d91c0","#225ea8", "#253494","#081d58", "black")) #set the colordata(wrld_simpl) #get world map datawrld <- spTransform(wrld_simpl, crs(dfr))plot(wrld, col='light gray', border="transparent",lwd=0.2)plot(dfr, col=colfunc(10),add=TRUE, legend.width = 1, legend.shrink=0.5,legend.args=list(text=' loss (log10 kcal)' ,side=4,cex=0.8,line=2), breaks=breaks)}And then we plot all of these together. The result forms the basis of figure 2 in the main text of themanuscript.par(mfrow = c(3,2),oma = c(0,0,0,0) ,mar = c(0,0,1,0))map.scenarios(global.NatOnly, global.NatOnly$kcal)map.scenarios(global.Sharing,global.Sharing$kcal.new)map.scenarios(country.NatOnly,country.NatOnly$kcal)map.scenarios(country.Sharing,country.Sharing$kcal.new)map.scenarios(ecoregion.NatOnly,ecoregion.NatOnly$kcal)map.scenarios(ecoregion.Sharing,ecoregion.Sharing$kcal.new)170123456789 loss (log10 kcal)0123456789 loss (log10 kcal)0123456789 loss (log10 kcal)0123456789 loss (log10 kcal)0123456789 loss (log10 kcal)0123456789 loss (log10 kcal)Figure 2: Maps of calorie losses under each Half-Earth scenario. From top to bottom: global, country andecoregion. Legends show kcal losses in log10 units. Left side: Nature only landscapes, Right side: Sharedlandscapes.183.5 Country calorie lossesHere we report the countries with calorie losses under the two di↵erent conservation approaches at theecoregion scale. We report countries with the top losses under the natures only landscapes approach below.Notably China and India show substantial calorie losses.country.loss.hard<-ddply(ecoregion.NatOnly, c("country"), summarize, #globalkcal.loss=sum(kcal, na.rm=T))head(country.loss.hard[order(-country.loss.hard$kcal.loss),])country kcal.loss31 CHN 4.9e+1476 IND 4.5e+14172 USA 3.1e+1475 IDN 1.6e+14115 MYS 7.5e+13119 NGA 6.6e+13We then compute the losses under a shared landscapes approach. While these reduce the losses for Chinaand India by at least a half, the losses still remain high.country.loss.soft<-ddply(ecoregion.Sharing, c("country"), summarize, #globalkcal.loss=sum(kcal.new, na.rm=T))head(country.loss.soft[order(-country.loss.soft$kcal.loss),])country kcal.loss76 IND 2.1e+1431 CHN 2.0e+14172 USA 1.7e+1475 IDN 9.3e+13115 MYS 6.4e+13119 NGA 5.1e+13Next we compute the total calories produced in di↵erent countries. Below we show the top calorie producingnations.country.kcal<-ddply(data.all, c("country"), summarize, #globalkcal=sum(kcal, na.rm=T))head(country.kcal[order(-country.kcal$kcal),])country kcal32 CHN 1.7e+15174 USA 1.4e+1578 IND 1.0e+1577 IDN 5.0e+1424 BRA 4.8e+14117 MYS 3.6e+14And finally we join all of these data, to estimate the percent losses. We are particularly interested in thelosses under the shared approach, which we report in the main text of the paper. We show the top losses inpercent terms below.kcal.count<-country.kcal[match(country.loss.soft$country,country.kcal$country), ]country.loss.soft$perc.loss<-country.loss.soft$kcal.loss/ kcal.count$kcal(country.loss.soft[order(-country.loss.soft$perc.loss),])[1:10,]country kcal.loss perc.loss24 BRN 1.4e+10 0.3061 GHA 8.8e+12 0.29119 NGA 5.1e+13 0.27130 PNG 1.4e+12 0.2514 BEL 8.5e+11 0.2496 LBR 4.3e+11 0.2376 IND 2.1e+14 0.21107 MKD 4.0e+11 0.20154 SVN 3.0e+11 0.2075 IDN 9.3e+13 0.19193.5.1 Ecoregion coverageIn an earlier publication Dinerstein et al. [9] reported that Half-Earth has been already attained, or beendeemed achievable, in 49% of the Earth’s 846 terrestrial Ecoregions. Dinerstein et al. identified 98/846 (or12%) had achieved more than half protected, and that 313/846 Ecoregions (or 37%) to have not achievedhalf protected, but to have enough “unaltered habitat” available to achieve it. They defined this unalteredhabitat (also synonymously labelled as both “natural” in the main text, and “natural/semi-natural” in theSupplementary Methods) by a two-step procedure, first using satellite classified tree cover data to identifyforests, and then defining remaining habitats according to anthropogenic biomes. However, in their analysisthey did not quantitatively identify the trade-o↵s with agricultural production. Here we identify how manyof the world’s Ecoregions could achieve the 50% figure without a loss of calories, under both the natureonly landscapes, and shared landscapes approaches. We find that for a nature only landscapes approach,Half-Earth can be achieved in 101 or the 775, or 14%, of Ecoregions used in this analysis without any loss ofcalories.lossbyeco<-ddply(ecoregion.NatOnly, 'Dinn', summarize,loss=sum(kcal, ra.rm=T))nrow(subset(lossbyeco, loss<10)) #number of ecoregions with no calorie loss[1] 111(nrow(lossbyeco[which(lossbyeco$loss<10),]))/nrow(lossbyeco) #percent of ecoregions[1] 0.14Interestingly, under a shared landscapes approach, we find that Half-Earth can be achieved in nearly fivetimes more ecoregions without calorie losses (or 65%).lossbyeco.s<-ddply(ecoregion.Sharing, 'Dinn', summarize,loss=sum(kcal.new, ra.rm=T))nrow(subset(lossbyeco.s, loss<10)) #number of ecoregions with no calorie loss[1] 501(nrow(lossbyeco.s[which(lossbyeco.s$loss<10),]))/nrow(lossbyeco) #percent of ecoregions.[1] 0.653.6 Co-benefitsTo understand the co-benefits of a Half-Earth scenario, (e.g. such as climate mitigation potential) it isimportant to understand the gross change in vegetation that might result from such a plan. Here we showthe how agriculturally productive land given back under each scale of analysis (country, ecoregion, andglobal), influences change in the coverage of potential vegetation. To do this we use the Potential NaturalVegetation (PNV) data set [15].First we need to calculate PNV coverage in area without agricultural production – this represents a baseline.We will then calculate the PNV area inside agriculturally productive pixels given back under each Half-Earthscenario. The ratio of these two numbers is used to indicate the potential vegetation benefits of giving backagricultural land. It should be noted that the baseline estimates are consistent across scenarios, but are likelyto overestimate PNV in current in non-agricultural lands because these lands are unlikely to be completelyunaltered by human activity (and hence our estimates of PNV gains are likely to be conservative).3.6.1 Nature only landscapesFirst we get the number of cells for the baseline PNV, in non-agriculturally productive and non-urban landglobally.cell.area <-8439*8439 #define cell areadata.all$PNV.base<-ifelse(data.all$rank >=7, NA,data.all$PNV)baseline<-ddply(data.all, 'PNV.base',20summarize,n.cells= length(x),ag.area= sum(CA*cell.area, PA*cell.area, na.rm=T),area=length(x)*cell.area)baseline$PNV.base<-baseline$area-baseline$ag.area #this is the baseline vegetation on the planetThen we obtain the PNV in recovered agricultural land in each scenario.world.no<-ddply(global.NatOnly, 'PNV',summarize,area= sum(CA*cell.area,PA*cell.area, na.rm = T),n.cells= length(x))country.no<-ddply(country.NatOnly, 'PNV',summarize,area= sum(CA*cell.area,PA*cell.area, na.rm = T),n.cells= length(x))ecoregion.no<- ddply(ecoregion.NatOnly, 'PNV',summarize,area= sum(CA*cell.area,PA*cell.area, na.rm = T),n.cells= length(x))Finally we join all data together into a common data.frame. The Gain.prop column represents the propor-tional gain in cells for each PNV class for agriculturally productive land.gain<-c(world.no$area/baseline$PNV.base,country.no$area/baseline$PNV.base,ecoregion.no$area/baseline$PNV.base)scenario<-c(rep("Global", 16), rep("Country", 16), rep("Ecoregion", 16))class<-rep(world.no$PNV, 3)gain.df.no<-data.frame(gain, scenario, class)colnames(gain.df.no)<-c("Gain.prop", "Scenario", "PNV.class")gain.df.no<-(subset(gain.df.no, !is.na(PNV.class))) # remove NAsWe then write the code to plot the results.NO<-ggplot(gain.df.no, aes( PNV.class,Gain.prop*100, colour=Scenario))+# geom_point()+geom_hline(yintercept=c(25,50,75,100), lty=2, colour="black", lwd=0.1)+geom_line()+scale_colour_manual(values = c("#56B4E9","#E75E00", "#009E73"))+theme_bw()+theme(plot.background = element_blank(),panel.grid.major = element_blank(),panel.grid.minor = element_blank(),strip.background = element_rect(colour = "white", fill = "white"))+theme(legend.position="none") +scale_x_continuous(breaks= 1:15, labels= c("1" ="Tropical Forest (E)", "2" ="Tropical Forest (D)","3" ="Temp.Forest (BE)","4" ="Temp.Forest (NE)","5" ="Temp.Forest (D)","6" ="Boreal.Forest (E)","7" ="Boreal.Forest (D)","8" ="Mixed Forest (E/D)","9" ="Savanna","10" ="Grassland/Steppe","11" = "Dense Shrubland","12" = "Open Shrubland", "13" ="Tundra", "14" ="Desert","15" ="Polar"))+ylab("Change in area (%)")+xlab("")+coord_flip()3.6.2 Shared lanscapesNow we repeat the analysis except for the sharing landscapes approach.world.s<-ddply(global.Sharing, 'PNV',summarize,n.cells= length(x),area= sum(CA.n,PA.n, na.rm = T)21)country.s<-ddply(country.Sharing, 'PNV',summarize,n.cells= length(x),area= sum(CA.n,PA.n, na.rm = T))ecoregion.s<- ddply(ecoregion.Sharing, 'PNV',summarize,n.cells= length(x),area= sum(CA.n,PA.n, na.rm = T))And join all data together into a common data.frame.gain.s<-c(world.s$area/baseline$PNV.base,country.s$area/baseline$PNV.base,ecoregion.s$area/baseline$PNV.base)scenario<-c(rep("Global", 16), rep("Country", 16), rep("Ecoregion", 16))class<-rep(world.s$PNV, 3)gain.df<-data.frame(gain.s, scenario, class)colnames(gain.df)<-c("Gain.prop", "Scenario", "PNV.class")gain.df.s<-(subset(gain.df, !is.na(PNV.class))) # remove NAsThen we create the plot.s<-ggplot(gain.df.s, aes( PNV.class,Gain.prop*100, colour=Scenario))+geom_hline(yintercept=c(25,50,75,100), lty=2, colour="black", lwd=0.1)+geom_line()+scale_colour_manual(values = c("#56B4E9","#E75E00", "#009E73"))+theme_bw()+theme(plot.background = element_blank(),panel.grid.major = element_blank(),panel.grid.minor = element_blank(),strip.background = element_rect(colour = "white", fill = "white"))+theme(legend.position="none")+scale_x_continuous(breaks= 1:15, labels= c("1" ="", "2" ="","3" ="","4" ="","5" ="","6" ="","7" ="","8" ="","9" ="","10" ="","11" = "","12" = "", "13" ="", "14" ="","15" =""))+ylab("Change in area (%)")+xlab("")+coord_flip()We also print a table of the gains below which show the percent gain figures under each scenario.all.gain<-data.frame(gain.df.no$Scenario, gain.df.no$PNV.class, gain.df.no$Gain.prop, gain.df.s$Gain.prop)colnames(all.gain)<-c("Scenario", "PNV.class", "Nat.Only", "Shared")all.gain$PNV.class<-recode(all.gain$PNV.class," c('1')= 'Tropical Forest (E)';c('2') = 'Tropical Forest (D)';c('3')= 'Temp.Forest (BE)';c('4')= 'Temp.Forest (NE)';c('5')= 'Temp.Forest (D)';c('6')= 'Boreal.Forest (E)';c('7')= 'Boreal.Forest (D)';c('8')= 'Mixed Forest (E/D)';c('9')= 'Savanna';c('10')= 'Grassland/Steppe';c('11')= 'Dense Shrubland';c('12')= 'Open Shrubland';c('13')= 'Tundra';c('14')= 'Desert';c('15')= 'Polar'")22all.gainScenario PNV.class Nat.Only Shared1 Global Tropical Forest (E) 0.0398 0.000002 Global Tropical Forest (D) 0.1214 0.000003 Global Temp.Forest (BE) 0.0717 0.000004 Global Temp.Forest (NE) 0.0444 0.000005 Global Temp.Forest (D) 0.1272 0.000006 Global Boreal.Forest (E) 0.0068 0.000007 Global Boreal.Forest (D) 0.0069 0.000008 Global Mixed Forest (E/D) 0.0173 0.000009 Global Savanna 0.1360 0.0000010 Global Grassland/Steppe 0.4150 0.0000011 Global Dense Shrubland 0.1262 0.0000012 Global Open Shrubland 0.2300 0.0000013 Global Tundra 0.0601 0.0000014 Global Desert 0.0295 0.0000015 Global Polar 0.0111 0.0000016 Country Tropical Forest (E) 0.0779 0.0710617 Country Tropical Forest (D) 0.2601 0.3931718 Country Temp.Forest (BE) 0.1016 0.0111719 Country Temp.Forest (NE) 0.0994 0.0219120 Country Temp.Forest (D) 0.2256 0.0529821 Country Boreal.Forest (E) 0.0098 0.0003822 Country Boreal.Forest (D) 0.0104 0.0021723 Country Mixed Forest (E/D) 0.0321 0.0062524 Country Savanna 0.2298 0.0850425 Country Grassland/Steppe 0.7275 0.2264926 Country Dense Shrubland 0.1989 0.0639427 Country Open Shrubland 0.4318 0.0832028 Country Tundra 0.1172 0.0037429 Country Desert 0.0388 0.0072730 Country Polar 0.0184 0.0031931 Ecoregion Tropical Forest (E) 0.0909 0.0758532 Ecoregion Tropical Forest (D) 0.4022 0.3777133 Ecoregion Temp.Forest (BE) 0.2689 0.1188334 Ecoregion Temp.Forest (NE) 0.1533 0.0664535 Ecoregion Temp.Forest (D) 0.3630 0.1835836 Ecoregion Boreal.Forest (E) 0.0103 0.0029837 Ecoregion Boreal.Forest (D) 0.0110 0.0042038 Ecoregion Mixed Forest (E/D) 0.0525 0.0282039 Ecoregion Savanna 0.3498 0.1424440 Ecoregion Grassland/Steppe 0.9416 0.6567041 Ecoregion Dense Shrubland 0.3280 0.1967142 Ecoregion Open Shrubland 0.4264 0.2230043 Ecoregion Tundra 0.0730 0.0319544 Ecoregion Desert 0.0359 0.0081145 Ecoregion Polar 0.0131 0.00297234 Supplementary FiguresIn this section we plot the supplementary figures that we do not present in the main text of the manuscript.4.0.1 Supplementary Figure 1First we plot the maps of binarized calorie losses by country under Half-Earth. Each plot shows Ecoregionsin which Half-Earth can be achieved without loss of calories. Note that Antarctica and Greenland are listedas locations under a “Nature only landscapes” approach where loss of calories occurs. This is due to thefact these locations are listed under a common “Rock and ice” Ecoregion, along with other locations on theEarth,such as a few pixels in Northern India and Nepal, which are subject to calorie loss. Removing this“Rock and ice” Ecoregion from this analysis makes no di↵erence to the percentage of losses for this approach(e.g. 101/775 vs 100/775).map.ecoregion.loss<-function(data, lossby){data.all$Dinn.no<-ifelse(data.all$Dinn %in% c(subset(lossby, loss<10)$Dinn), 1,0 )dfloss<-data.frame(data.all$x, data.all$y,data.all$Dinn.no)dfrloss<- rasterFromXYZ(dfloss) #Convert first two columns as lon-lat and third as valuecrs(dfrloss)<-"+proj=eck4 +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0"colfunc <- colorRampPalette(c("red", "light gray")) #set the color rampdata(wrld_simpl) #get world map datawrld <- spTransform(wrld_simpl, crs(dfrloss))plot(wrld, col='light gray', border="transparent",lwd=0.2)plot(dfrloss, col=colfunc(10), add=TRUE, legend=F)}par(mfrow = c(2,1),oma = c(0,0,0,0) ,mar = c(0,0,1,0))map.ecoregion.loss(data.all, lossbyeco)map.ecoregion.loss(data.all, lossbyeco.s)24Figure 3: Ecoregions where Haf-Earth be achieved without a loss of calories. Gray= Ecoregions with noloss of calories, Red= Ecoregions with loss of calories. Top = Nature only landscsapes. Bottom = Sharedlandscapes.254.0.2 Supplementary Figure 2Then we plot out the co-benefits of Half-Earth for recover of natural vegetation. As we see from below, theecoregion based Half-Earth plan leads to large gains in grasslands and modest gains in tropical and temperatedeciduous forests, and in savanna and shrublands.plot_grid( NO,s, align =c("v", "h"), ncol = 2)Tropical Forest (E)Tropical Forest (D)Temp.Forest (BE)Temp.Forest (NE)Temp.Forest (D)Boreal.Forest (E)Boreal.Forest (D)Mixed Forest (E/D)SavannaGrassland/SteppeDense ShrublandOpen ShrublandTundraDesertPolar0 25 50 75 100Change in area (%)0 25 50 75 100Change in area (%)Figure 4: Recovery of natural vegetation from agriculture under each Half-Earth Scenario. The x axis showsthe percent gains in areal coverage in potential natural vegetation attained by giving back least agriculturallyproductive lands. green= global, blue= country, red= ecoregion. E=evergreen, D=deciduous, N=needle,B=broadleaf. Left = Nature only landscsapes. Right = Shared landscapes.264.0.3 Supplementary Figure 3And finally we plot out a flow diagram for the analysis, which we reference in the methods section of ourpaper.Rasterize data, set to common extent, and re-project to equal area Create common land use class variable Rank pixels by land use class Rank pixels in each land class by crop calorie production   Algorithm to allocate pixels to nature within each geographic boundary in order of rank, until 50% areal coverage is achieved.Rank area within each pixel based on non-agricultural area, pasture area, non-food calories cropland and food calorie cropland.Algorithm to allocate area to nature within each geographic boundary in order of rank until 50% areal coverage is achieved. Estimate percent losses to cropland area, pasture area, non-food calories, and food caloriesRun algorithms at 3 scales —by setting the geographic boundary to global, country or ecoregion."Nature only”Scale of analysisTrade-offsData processingLand use prioritizationConservation strategy“Shared”27Session informationsessionInfo()R version 3.4.2 (2017-09-28)Platform: x86_64-apple-darwin15.6.0 (64-bit)Running under: macOS High Sierra 10.13.5Matrix products: defaultBLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylibLAPACK: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libLAPACK.dyliblocale:[1] en_CA.UTF-8/en_CA.UTF-8/en_CA.UTF-8/C/en_CA.UTF-8/en_CA.UTF-8attached base packages:[1] stats graphics grDevices utils datasets methods baseother attached packages:[1] data.table_1.10.4-2 cowplot_0.8.0 gridExtra_2.3[4] reshape2_1.4.3 maptools_0.9-2 car_2.1-6[7] rworldmap_1.3-6 R.matlab_3.6.1 plyr_1.8.4[10] ggplot2_2.2.1 rgdal_1.1-10 raster_2.6-7[13] sp_1.2-7 checkpoint_0.4.1 knitr_1.17[16] RevoUtils_10.0.6loaded via a namespace (and not attached):[1] Rcpp_0.12.13 nloptr_1.0.4 compiler_3.4.2[4] highr_0.6 R.methodsS3_1.7.1 R.utils_2.5.0[7] tools_3.4.2 lme4_1.1-14 dotCall64_0.9-04[10] nlme_3.1-131 evaluate_0.10.1 tibble_1.3.4[13] gtable_0.2.0 lattice_0.20-35 mgcv_1.8-22[16] rlang_0.2.1 Matrix_1.2-11 parallel_3.4.2[19] SparseM_1.77 spam_2.1-1 stringr_1.2.0[22] MatrixModels_0.4-1 fields_9.0 maps_3.2.0[25] grid_3.4.2 nnet_7.3-12 foreign_0.8-69[28] minqa_1.2.4 magrittr_1.5 scales_0.5.0[31] MASS_7.3-47 splines_3.4.2 pbkrtest_0.4-7[34] colorspace_1.3-2 labeling_0.3 quantreg_5.33[37] stringi_1.1.5 lazyeval_0.2.0 munsell_0.4.3[40] R.oo_1.21.0Supplementary References[1] Baptiste Auguie. gridExtra: Miscellaneous Functions for ”Grid” Graphics. R package version 2.3. 2017.url: https://CRAN.R-project.org/package=gridExtra.[2] Andrew Balmford and Rhys Green. “How to spare half a planet”. In: Nature 552 (2017), p. 175. issn:14764687. doi: 10.1038/d41586-017-08578-7.[3] Henrik Bengtsson. R.matlab: Read and Write MAT Files and Call MATLAB from Within R. R packageversion 3.6.1. 2016. url: https://CRAN.R-project.org/package=R.matlab.[4] BirdLife. World Database of Key Biodiversity Areas. 2017. url: www.keybiodiversityareas.org(visited on 11/01/2017).[5] Roger Bivand, Tim Keitt, and Barry Rowlingson. rgdal: Bindings for the Geospatial Data AbstractionLibrary. R package version 1.1-10. 2016. url: https://CRAN.R-project.org/package=rgdal.[6] Roger Bivand and Nicholas Lewin-Koh. maptools: Tools for Reading and Handling Spatial Objects. Rpackage version 0.9-2. 2017. url: https://CRAN.R-project.org/package=maptools.[7] Emily S Cassidy et al. “Redefining agricultural yields: from tonnes to people nourished per hectare”.In: Environmental Research Letters 8 (2013), p. 034015. issn: 1748-9326. doi: 10.1088/1748-9326/8/3/034015. url: http://iopscience.iop.org/1748-9326/8/3/034015/article/.[8] Microsoft Corporation. checkpoint: Install Packages from Snapshots on the Checkpoint Server for Repro-ducibility. R package version 0.4.1. 2017. url: https://CRAN.R-project.org/package=checkpoint.28[9] Eric Dinerstein et al. An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm. 2017. doi:10.1093/biosci/bix014. arXiv: 1611.06654.[10] Matt Dowle and Arun Srinivasan. data.table: Extension of ‘data.frame‘. R package version 1.10.4-2.2017. url: https://CRAN.R-project.org/package=data.table.[11] John Fox and Sanford Weisberg. car: Companion to Applied Regression. R package version 2.1-6. 2017.url: https://CRAN.R-project.org/package=car.[12] Mark A. Friedl et al. “MODIS Collection 5 global land cover: Algorithm refinements and characteriza-tion of new datasets”. In: Remote Sensing of Environment 114.1 (2010), pp. 168–182. issn: 00344257.doi: 10.1016/j.rse.2009.08.016. url: http://dx.doi.org/10.1016/j.rse.2009.08.016.[13] Robert J. Hijmans. raster: Geographic Data Analysis and Modeling. R package version 2.6-7. 2017. url:https://CRAN.R-project.org/package=raster.[14] Edzer Pebesma and Roger Bivand. sp: Classes and Methods for Spatial Data. R package version 1.2-7.2018. url: https://CRAN.R-project.org/package=sp.[15] Navin Ramankutty and Jonathan A Foley. “Estimating historical changes in global land cover: crop-lands from 1700 to 1992”. In: Global Biogeochemical Cycles 13 (1999), pp. 997–1027.[16] Navin Ramankutty et al. “Farming the planet : 1 . Geographic distribution of global agricultural landsin the year 2000”. In: 22.August 2007 (2008), pp. 1–19. doi: 10.1029/2007GB002952.[17] Andy South. rworldmap: Mapping Global Data. R package version 1.3-6. 2016. url: https://CRAN.R-project.org/package=rworldmap.[18] QGIS Development Team. QGIS Geographic Information System.Open Source Geospatial FoundationProject. 2018. url: http://qgis.osgeo.org.[19] UNEP-WCMC and IUCN. The World Database on Protected Areas (WDPA)/The Global Database onProtected Areas Management E↵ectiveness (GD-PAME)]. 2017. url: https://www.protectedplanet.net/ (visited on 11/01/2017).[20] Hadley Wickham. plyr: Tools for Splitting, Applying and Combining Data. R package version 1.8.4.2016. url: https://CRAN.R-project.org/package=plyr.[21] Hadley Wickham. reshape2: Flexibly Reshape Data: A Reboot of the Reshape Package. R package version1.4.3. 2017. url: https://CRAN.R-project.org/package=reshape2.[22] Hadley Wickham and Winston Chang. ggplot2: Create Elegant Data Visualisations Using the Grammarof Graphics. R package version 2.2.1. 2016. url: https://CRAN.R-project.org/package=ggplot2.[23] Claus O. Wilke. cowplot: Streamlined Plot Theme and Plot Annotations for ’ggplot2’. R package version0.8.0. 2017. url: https://CRAN.R-project.org/package=cowplot.[24] EO Wilson. Half earth : our planet’s fight for life. New York: Liveright, 2016.[25] Yihui Xie. knitr: A General-Purpose Package for Dynamic Report Generation in R. R package version1.17. 2017. url: https://CRAN.R-project.org/package=knitr.29

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.52383.1-0376069/manifest

Comment

Related Items