Impacts of GHG emissions abatement measures on agricultural market and 1 food security


 Agriculture, Forestry and Other Land-use (AFOLU) are thought to play a vital role in long-term GHG emissions reduction, especially for their importance in non-CO2 emissions, bioenergy supply and carbon sequestration realized by afforestation. Several studies have noted potential adverse impacts of land-related emissions mitigation on food security, due to food price increases, but these studies have not disaggregated the individual aspects of land-related emissions mitigation that impact food security. Here, we show the extent to which three factors—non-CO2 emissions reduction, bioenergy production, and afforestation—change the food security and agricultural market conditions under 2 °C climate stabilization scenarios, using six global agro-economic models. The results show that afforestation, often implemented in the models by imposing carbon prices on land carbon stocks, causes the largest impacts on food security, followed by non-CO2 emissions policies, generally implemented as emissions taxes. Respectively, these measures put an additional 41.9 and 26.7 million people at risk of hunger in 2050. This study highlights the need for better coordination of emissions reduction and agricultural market management policy.


Introduction 19
In meeting near-and long-term climate change mitigation goals (e.g., the Paris Agreement), the 20 energy sector accounts for the majority of greenhouse gas (GHG) emissions in most nations, and is 21 thus the target of most present-day emissions mitigation policies. However, Agriculture, Forestry and 22 Other Land Use (AFOLU) account for 20-25% of global GHG emissions in 2010 1 and cannot be 23 ignored in the context of meeting ambitious long-term climate change mitigation targets. In addition 24 to the baseline emissions quantities involved, the non-point-source nature of the emissions, combined 25 with the relative lack of available technologies to eliminate emissions, make AFOLU emissions 26 abatement especially difficult. This is in contrast to the energy sector, whose emissions can become 27 net-zero or even net-negative if carbon removal technologies are used 2 . 28 The future emissions reduction potential in the AFOLU sector has been characterized in the 29 literature as having relatively large emission reductions available at low cost, compared with other 30 sectors 3, 4, 5 . However, the emissions reduction potentials are understood to be limited, with full (100%) 31 removal not possible regardless of effort in many cases 6 . Moreover, Hasegawa et al. (2018) highlighted 32 significant food security concerns associated with including AFOLU in climate change mitigation 33 actions 7 . The present study contributes to this discussion, by starting with the observation that there 34 are three major channels by which AFOLU-focused climate change mitigation policy may exacerbate 35 food security. One is promotion of large-scale bioenergy crop expansion; low-emissions scenarios in general agricultural producer price indices are projected to be almost constant over this timeframe, 75 with a range of 0.95 to 1.26 in 2050 (Figure 1c). Price projection diversity across models has been 76 observed in the earlier studies as well 29 . Agricultural technological improvement and demand 77 increases are the main negative and positive drivers of prices, respectively, which tend to offset each 78 other. 79 Under the climate change mitigation scenario to attain well below 2°C global mean temperature, 80 there would be carbon (or GHG) pricing implicitly or explicitly which is fed into the models (see For example, afforestation effect on additional risk of hunger ranges from 10.9 to 90.3 million. These 95 model variations would depend on the representation of the mitigation measures and model structure 96 which are discussed in detail later. Given that there is model uncertainty, we carried out a sensitivity 97 analysis to test whether a specific "extreme" model would lead this conclusion or not. This sensitivity 98 analysis is conducted by withdrawing one model, and iterating for all models. The conclusion is that 99 our results are not dependent on a specific model (Supplementary Figure 3). Also, analysis based on 100 the four models with complete sets of scenarios show similar patterns (Supplementary Figure 5). 101 Note that models which have explicit energy and economy components within the model show 102 non-agricultural and non-land-use related effects to some extent (e.g. income-loss associated with low-103 carbon energy technologies). It would be smaller than others except for AIM/Hub which shows 104 additional 29.0 million people become under the risk of hunger (Supplementary Figure 4). 105

Drivers of food price increases 113
Afforestation for the purpose of sequestering carbon from the atmosphere to the terrestrial system 114 is incentivized by carbon pricing on the carbon sink above-and below-ground. The CO 2 emissions 115 drastically decreases in the mitigation scenarios (Figure 2a) and become negative 3.80 (0.20-13.74) 116 Gt CO 2 in 2050. Accordingly, forest area increases by 11.1% (1.7%-24.7%) in 2050 relative to 117 baseline scenarios and these forest area expansion put an additional land demand pressure on overall 118 agricultural activities (Figure 2e). The land rent can also increase by pricing on land carbon sink and 119 both factors would increase the average land rent by 366% in 2050 (from AIM/Hub model). 120 Non-CO 2 emissions mitigation is the second largest contributor to the price increases associated 121 with mitigation measures. There are basically two factors to increase the agricultural production prices. which simply add up the agricultural production cost particularly in livestock products (Supplementary 124 Figure 6). Second, in contrast to CO 2 emissions which can be negative value, non-CO 2 are thought to 125 be difficult completely gotten rid of and some potions such as 68.6% (56.5%-84.6%) remain as 126 residual emissions (Figure 2bc), which is slightly larger than existing literature but possibly due to the 127 sectoral coverage 30, 31 . This would become a penalty of carbon pricing (e.g. carbon tax imposition). 128 Interestingly, the maximum emissions reductions almost reach under a certain carbon price which 129 would imply that further higher carbon prices that are primarily determined by energy system side in 130 IAMs would increase the penalty of the carbon prices from that point (Figure 2h). 131 Finally, the energy purpose biomass crop can compete with current food crops which pushes the 132 land demand pressure on the land market. Current model estimates show the bioenergy crop area is 133 196 (62-494) million ha in 2050 under the full mitigation policy, which accounts for 11.7% of the 134 current cropland area (Figure 2d). 135 Although afforestation and bioenergy both need large amount of land and therefore might 136 compete with land for food production, results suggest that the effect of afforestation is larger than 137 bioenergy. That's because afforestation requires more land than bioenergy, possibly due to the higher 138 carbon sink capacity of bioenergy crops particularly combined with CCS. In the bioenergy scenario,  relationship between carbon price and forest area, bioenergy area and non-CO 2 emissions reduction 154 rates (Non-CO 2 emissions is CO 2 equivalent value using GWP2100 in AR5). 155 156

Regional implications 157
The global trend in terms of the composition shares of three mitigation measures in risk of hunger 158 is in principle similar across regions. Risk of hunger in most regions except for Sub-Saharan Africa is 159 projected to decrease overtime mainly driven by income growth as global results, which might not be 160 the case for short-term due to COVID-19. China and India have relatively high-income growth and 161 thus the risk of hunger rapidly decreases, which are 43.6 (21.4-70.7) and 79.0 (22.8-100.0) million, 162 change ratios associated with total mitigation measures would be more or less similar across regions 166 and the absolute population changes would depend on the baseline projection except for African region.

Discussions and conclusions 215
We have identified the three main causes of food security and agricultural changes associated 216 with climate change mitigation measures; namely afforestation, bioenergy expansion and non-CO 2 217 emissions abatement. Afforestation turned out to be the primary driver of making adverse-side effects 218 on food security followed by non-CO 2 . We confirm this similar implication under different 219 socioeconomic assumptions with multiple global agricultural economic models. We further 220 demonstrate that specific extreme models do not lead our conclusion. Regionally, Sub-Saharan Africa 221 is most vulnerable to these shocks. Our results indicate the complexity and challenges in the AFOLU 222 sector's climate mitigation policy from multiple angles. 223 We summarize the logical chains of the causes and effects of climate change mitigation measures 224 and agricultural price increases in Figure 5 which we should think how to cut off any chains linking 225 to the cost increases for each. Most stringent climate stabilization scenarios heavily rely on negative 226 emissions technologies such as afforestation and BECCS, and non-CO 2 emissions would be more 227 important under low or net zero emissions conditions 4, 34 . The carbon pricing on land carbon stock productivity, the stronger this incentive is. This afforestation induces the cropland decreases and 230 production cost upwards. Bioenergy increases can trigger the similar effects. In this case, energy crop 231 land is the competitors for food crops. The non-CO 2 effect has a slightly different way to increase the 232 cost, which directly hit the food crop production by the technological implementation of non-CO 2 233 emissions abatement and carbon price imposition to the residual emissions. Once climate policy would 234 give incentives to these measures, it might be difficult to cut-off the left arrows in Figure 5. One 235 possibility to prevent this situation would be transforming the societal structure completely (e.g. 236 reducing energy demand drastically 35 and lifestyle changes 36 ). Although there are possibilities that the 237 society move forward to such directions, it would be too optimistic to only bet on that. 238 The second left arrows in Figure 5 would be able to be somehow cut off by policy. For example, 239 even if large scale afforestation and bioenergy expansion occur, land rent could be controlled by policy. 240 To prevent those land demands invading cropland for food, the strong regulation on the cropland for 241 food cultivation might also work. Note that without the carbon pricing on land sink, there would be 242 strong incentives to cultivate the land for gaining negative emissions and thus there must be some 243 policies specifically to deal with non-food land demand 37, 38 . Regarding the right-hand side arrows 244 linking from secondary effects to the production costs in Figure 5, policy roles would again be crucial, 245 but technology can also change the situation. As current agricultural policy conditions in many 246 countries, there are more or less supports for agricultural production directly or indirectly. This would 247 imply that the production cost increase could be managed by such policies similarly. For example, the 248 subsidy would be often used for the agricultural sector, and in this case, subsidy for the incremental 249 cost to off-set the price increases could be a possible solution. Then, the issue of this policy would be 250 the scale of the market distortion. The current estimates show the cost increases by around 30% relative 251 to baseline. If we aim to subsidize these cost increases, how to get the tax revenue as a source of 252 subsidy and getting social acceptance would be an issue. Note that carbon tax revenue might be a right 253 candidate for this purpose 39 . Technological progress in non-CO 2 emissions reduction and bioenergy 254 yield would mitigate the agricultural cost increases. While it is essential to encourage research and 255 development in those technologies, we should keep in mind that technological progress is essentially 256 uncertain. 257 Besides the supply side management, there can be demand-side transition to mitigate the adverse 258 side effects such as dietary shift reducing meat consumption 40, 41 , changing distribution of food for the 259 poor people and implementing subsidy for consumption 14,42 . In any cases, abovementioned measures 260 may not be effective with a single independent action. We would need holistic approaches for food 261 security under deep decarbonization transition. Trade would be thought as one of the measures to fill 262 the gap between supply and demand in general, but at least in our study's framework it would not be 263 a magical tool to resolve all issues because the climate policy (carbon price) is implemented globally. 264 Other than the above discussion points, there should be some more arguments. First, the model 270 uncertainty was not so small to indicate that the conclusion is always hold with any models in this 271 study and this model variation comes from several factors. One is the way how to implement emissions 272 reduction measures. For example, the models that show relatively high afforestation effects are AIM, 273 GCAM and MAGNET, which explicit have land rent representation that may be related to the outcome. 274 As reported earlier, land rent is currently relatively cheap, and carbon pricing on carbon sink would 275 drastically affect. In contrast, GLOBIOM has no explicit land rent representation, and the cost 276 increases in agricultural commodities would be caused by cropland shift from high to low productivity 277 area. These differences would be eventually crucial. Since the structure quite differs across models (at 278 least partial equilibrium models and general equilibrium models), it would be difficult to completely 279 harmonize the way of implementation of mitigation measures and the model ranges would not be 280 narrowed even if we put much efforts on that. 281 Second, in this study, we purely focused on agricultural and food security aspects, but there must 282 be side-effects of afforestation on ecosystem (e.g. biodiversity). If forest area is newly expanded and 283 revert to the natural forest land using such as native tree species, that would have additional 284 environmental co-benefit in regenerating habitat of the lives 43 . In contrast, if the afforestation purely 285 aims to sequestrate carbon from the atmosphere, the tree species for that purpose might be transplanted 286

Overall methodology 295
We carried out a scenario analysis to decompose afforestation, bioenergy and non-CO2 effects 296 on agricultural market and food security. Overall research framework is shown in Supplementary 297 Figure 1. Six state-of-the-art global agricultural economic or integrated assessment models, which 298 sufficiently represent agricultural sectors, and land use to assess the interaction between climate 299 mitigation and food security, are applied for this scenario exercise. Global economic models compute 300 the agricultural consumptions, production, land-use area and associated emissions by crops and 301 livestock as well as forestry. Food consumption is then fed into hunger tool which computes individual 302 countries' food consumption distribution and population at risk of hunger. To identify the magnitude 303 of three causes on agricultural market, we developed a sensitivity scenario protocol that systematically 304 switching on/off the mitigation options. Here we describe 1) scenario definition and protocol, 2) a 305 brief model overview for each agricultural model (a summary is in Supplementary Table 2), and 3) 306 hunger tool description. 307 308

Scenarios and experiment design 309
We developed a set of scenarios with combination of three socio-economic conditions and one 310 mitigation policy scenarios (and one baseline scenario) that are consistent with 2 °C goal stated in the 311 Paris Agreement or equivalent to RCP2.6 level emissions reduction 47 . For the socio-economic 312 assumptions, we used three SSPs from the internationally developed SSP framework designed to 313 conduct cross-sectoral assessments of climate change impact, adaptation, and mitigation 26 . The SSPs 314 are representative future scenarios, which includes both qualitative and quantitative information in 315 terms of challenges in mitigation and adaptation to climate change. In this study, we used three SSP 316 scenarios from the SSP framework, i.e., "sustainability" (SSP1) 23 , "middle of the road" (SSP2) 48 , and 317 "regional rivalry pathways" (SSP3) 18 to address the uncertainty of socio-economic conditions. 318 To isolate the effect of each land-based mitigation options (afforestation, bioenergy, and non-CO 2 319 emissions reduction + carbon price imposition), we used a recently developed 49, 50 and widely applied 320 methodology 27 that identifies the individual effects of an input factor with a limited number of model 321 experiments even in a complex system. In general, we could classify the mitigation options into four three scenarios with each applying only one of the land-related mitigation options, and one scenario 325 that apply all three options simultaneously, as shown in Supplementary Table 3 (for the models with  326 non-agricultural sectoral emission (e.g., energy sectors), including AIM, GCAM, and FARM, since it 327 is difficult to turn off these mitigation options, tax on non-agricultural sectoral emission was also 328 applied in these four scenarios). 329 We have also run sensitivity scenarios for the models with enough modeling ability, where most 330 mitigation measures are available and only one is switched off with AIM and GLOBIOM which 331 sufficiently represent all agricultural activities and GHG emissions(Supplementary Table 2  which applies an endogenous carbon price??) was imposed in these mitigation scenarios 48  production factors). Land is modelled as an explicit production factor described by a land supply curve, 377 constructed with land availability data provided by IMAGE. 378 IMAGE is a comprehensive integrated assessment framework, modelling interactions between the 379 human and natural systems 56 . The framework comprises a number of sub-models describing land use, 380 agricultural economy, the energy system, natural vegetation, hydrology, and the climate system. In this 381 study specifically the land component is applied which represents land use, crop production, 382 afforestation and the carbon cycle spatially explicitly at 5 arc-minutes resolution. In particular, GCAM assumes the carbon price is applied to carbon stocks held in the terrestrial system, 415 incentivizing land owners to increase these stocks. As a result, strong incentives exist to expand carbon 416 stocks under a climate policy, resulting in significant afforestation. The agriculture and land-use 417 component is connected to the climate through emissions (CO 2 and non-CO 2 ), which are produced by 418 the land system and passed into the climate system to calculate concentrations, radiative forcings, and 419 other climate indicators. 420 421 CAPRI (Common Agricultural Policy Regionalised Impact) modelling system is an economic large-422 scale, comparative-static, partial equilibrium model focusing on agriculture and the primary 423 processing sectors (www.capri-model.org). CAPRI comprises two interacting modules, linking a set 424 of mathematical programming models of EU regional agricultural supply to a spatial multicommodity 425 model for global agri-food markets. The regional EU supply models depict a profit maximizing has been extended in many ways beyond the "GTAP in GAMS" model described in Lanz and 449 Rutherford (2016) 66 : conversion from comparative-static to a recursive-dynamic framework; 450 conversion of the consumer demand system from constant-elasticity-of-substitution (CES) to the 451 Linear Expenditure System (LES); allowing for joint products in production functions; introduction 452 of land classes for agricultural and forestry production; and introduction of electricity-generating 453 technologies. Two markets are important for bioelectricity: the market for land and the market for 454 electricity. Bioelectricity must compete against crops, pasture, and forest for land, and must also 455 compete for a share of electricity generation. Land shifts among crops, pasture, and forests in response 456 to population growth, dietary preference, changes in agricultural productivity, and policies such as a 457 renewable portfolio standard or a carbon tax. Land competition is based on the land rent for each 458 competing use: land use is adjusted within agroecological zones until rents at the margin are equal. 459 Carbon dioxide capture and storage is available for electricity generated from fossil fuels and from 460  Each IAM reports the mean food calorie consumption per person per day (cal). We standardize 484 the base year calorie consumption to what FAO reports and take the change ratio of each year to the 485 base year for IAMs. We then compute the standardized calorie consumption to make a consistent 486 number for those at risk of hunger. In this process, since the IAM's are regionally aggregated values, 487 they are downscaled to the individual country level by taking the base year value reported FAO and 488 future change ratio from IAMs. The CV is an indicator of food security observed in a household 489 survey conducted by the FAO. It ranges from 0 to 1. FAO country data for CV are weighted on the 490 basis of population data in the base year and aggregated to regional classification to obtain the CV of 491 aggregated regions. The CV is changed over time with the consideration of income growth dynamics 492 as presented in Hasegawa et al. 24 . Note that there is an assumption that the future CV changes of 493 each region is based on the current regional value. 494