CLIMATE CHANGE AND HUNGER

This report assesses the cost of adaptation to climate change across a range of future climate scenarios and investment options. We focus on offsetting climate change impacts on hunger through investment in agricultural research, water management, and rural infrastructure in developing countries. We link climate, crop, water, and economic models to (1) analyze scenarios of future change in the agriculture sector to 2050 and (2) assess trade-offs for these investments across key Sustainable Development Goals (SDGs) for poverty, hunger, and water. Our reference projections show that climate change slows progress toward eliminating hunger, with an additional 78 million people facing chronic hunger in 2050 relative to a no-climate-change future, over half of them in Africa south of the Sahara. Increased investments can offset these impacts. Achieving this would require that annual investment in international agricultural research increase from US$1.62 billion to US$2.77 billion per year between 2015 and 2050. Additional water and infrastructure investments are estimated to be more expensive than agricultural R&D at about US$12.7 billion and US$10.8 billion per year, respectively, but these address key gaps to support transformation toward food system resiliency. Findings on ranges of costs and trade-offs and complementarities across SDGs will help policymakers make better-informed choices between alternative investment strategies. INTRODUCTION Climate change is a growing challenge for food systems around the world, presenting both changing trends and increased variability in temperatures and precipitation. These changes threaten agricultural productivity, food and nutrition security, health, and the potential for achieving key Sustainable Development Goals (SDGs) for poverty, hunger, and water, among others. Despite these challenges, however, appropriate investments in the agriculture sector can help people and food systems adapt by increasing productivity, resilience, and resource-use efficiency. To help policymakers and other decision-makers evaluate alternative adaptation strategies, this report assesses the cost of adaptation to climate change across a range of future climate scenarios and investment options, with a particular focus on adaptation to offset climate change impacts on hunger. Climate change, its impacts on agriculture, and measures to address those impacts are all highly complex phenomena. Estimating the cost of such measures is correspondingly difficult. Rather than attempting to estimate the total cost of adaptation, Nelson et al. (2009) focused on one particular dimension of these phenomena: the cost of selected agricultural investment strategies to offset the adverse impact of climate change on humans as proxied by future projections of childhood undernutrition. The assumption was that if we can avert impacts on the most vulnerable, then we will likely have taken care of the rest of the population as well. This report does the same, but with an updated methodology and a broader focus on an estimate of the total population at risk of chronic hunger. We draw on recent analyses by Rosegrant et al. (2017), Rosegrant et al. (2021), and Mason-D’Croz et al. (2019), which in turn built on Nelson et al. (2009). Other studies have used a variety of approaches to estimate the cost of climate change adaptation. Countryand sector-level studies in the 1990s assessed adaptation costs based on the estimated damages from climate change impacts (Tol 2005; Tol, Fankhauser, and Smith 1998). The first global estimates of adaptation costs shared a methodology developed by the World Bank (2006) that estimated the fraction of investments flowing to developing countries deemed sensitive to climate change impacts and then used a “markup” factor to represent the costs of “climate-proofing” those investments.1 These studies did not specify a particular time horizon or climate 1 The initial analysis estimated that 2–10 percent of gross domestic investment, 10 percent of foreign direct investment, and 40 percent of official development assistance were sensitive to climate change. The authors estimated that a markup of 10–20 percent would be needed to climate-proof these investments (World Bank 2006). 7 scenario, but the use of current investment flows as a starting point suggests that they focused on shortterm adaptation costs. Estimates vary widely due to the lack of empirical data on climate sensitivity and the costs of climateproofing. Studies by UNFCCC (2007) and Forstater and Zadek (2009) specified a time horizon of 2030 and included more detailed assessments of investment and financial flows for each sector. A 2010 review of published studies on the global cost of adaptation found that estimates ranged between US$25 and US$100 billion per year for the period 2015–2030 (Agrawala and Fankhauser 2008; Fankhauser 2010). These figures referred to the costs of adapting to a warming of 2–3 degrees Celsius and were estimated to rise exponentially for higher temperatures. A subsequent analysis by the World Bank (2010) estimated the cost at US$70–100 billion annually for developing countries for the period 2010 to 2050. These were the most recent comprehensive estimates as of the Intergovernmental Panel on Climate Change’s (IPCC) Fifth Assessment Report (Chambwera et al. 2014), and arguably the most robust (O’Garra and Mourato 2016). A recent study by Baldos, Fuglie, and Hertel (2020) focused on national public spending on R&D at aggregate regional levels; they estimated that climate adaptation costs that offset climate impacts on crop yields would cost a total of US$187 to US$1,384 billion from 2020 to 2040. In contrast to these studies, Nelson et al. (2009) focused more specifically on a detailed analysis of the impacts of climate change on: (1) agricultural productivity and related outcomes (including child undernourishment) to 2050, by country and commodity; and (2) the cost of selected agricultural investment strategies in developing countries (namely agricultural research, soil and water management, and rural infrastructure) to counter those impacts. More recent work by Rosegrant et al. (2017) and Mason-D’Croz et al. (2019) extended that approach, incorporating updated models and data and an emphasis on looking at trade-offs and complementarities across different investment strategies. Additional analyses led by colleagues from across the CGIAR broadened these two studies by taking deeper looks at particular commodities and development issues, such as fish, livestock, cereals, and roots, tubers, and bananas (Chan et al. 2019; Enahoro et al. 2019; Kruseman et al. 2020; Petsakos et al. 2019), and agriculture sector income and employment (Frija et al. 2020). 8 CLIMATE CHANGE AND HUNGER APPROACH AND METHODS MODELING FOOD SYSTEM FUTURES UNDER CLIMATE CHANGE AND ADAPTATION INVESTMENT SCENARIOS This analysis builds on the recent work by Rosegrant et al. (2017) and others by analyzing a more complete range of future scenarios along with sensitivity simulations on key costing parameters. The goal is to help establish a more robust foundation for the assessment of the cost of adaptation to climate change for the food system in the developing world. The core output of this study is a direct analog and update to the cost of climate change adaptation presented in Nelson et al. (2009). See Appendix A for the core details for comparison. To simulate changes in the global food system and assess investment scenarios for their effectiveness to offset climate change impacts out to mid-century, we used IFPRI’s International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) (Robinson et al. 2015). This modeling framework links climate, crop, water, and economic models to analyze scenarios of future change in agricultural production, consumption, prices, and trade at national, regional, and global scales. The methodology behind the IMPACT modeling framework has evolved and been refined over the past couple of decades in collaboration with other leading research groups around the world, particularly with partners across CGIAR and in the Global Economics group of the Agricultural Model Intercomparison and Improvement Project (AgMIP, https://agmip.org/global-economics/). The IMPACT modeling framework has been applied extensively in numerous peer-reviewed publications and decisionmaking fora, such as international development organizations, national governments, foundations and grant-making institutions, and across CGIAR itself (https://www.ifpri.org/project/ifpri-impact-model). IMPACT produces reference scenarios to 2050 assuming no climate change as well as different levels of climate change and socioeconomic assumptions from the IPCC, capturing a wide range of possible climate and socioeconomic futures (Table 1). In conjunction with these reference scenarios, we then analyzed scenarios of how plausible investment options — including investments in agricultural R&D, water management, and infrastructure (determined in collaboration with experts across the CGIAR) — could help offset the potential impacts of climate change. Many complex connections exist between climate change and hunger (Mbow et al. 2019; Beach et al. 2019; Nelson et al. 2018). This analysis focuses on a subset of those. In the most basic sense, climate change, by and large, negatively affects agricultural production by reducing yields. This reduced productivity then reduces availability and drives prices upward, ultimately resulting in increased hunger. The investments modeled here generate economic benefits through several pathways. Specific details are summarized in Appendix D, but in general, increased agricultural R&D boosts crop and livestock yields, reduces food prices, and increases farm income and economywide gross domestic product (GDP) through multiplier effects on the nonagriculture sectors. Increasing incomes, in turn, shift consumers toward different diets, with consumption of usually more food and more diverse food types. Irrigation and water use efficiency investments increase crop yields and reduce prices, and generating higher incomes. Enhanced rural infrastructure reduces postharvest losses and marketing margins, improving the profitability of farm production, and boosting supply to consumers for any given level of production. These effects also increase farm and economywide income, which increases food consumption and reduces hunger. While the investment alternatives modeled in this study are extensive, this list is not exhaustive and the focus in this study is on classes of investments that are commonly recognized across the agricultural development community and that can be modeled directly in IMPACT. 9 To expand the representation across the range of core drivers, we extended the original scenario set in Rosegrant et al. (2017) to show outcomes across different combinations of climate change assumptions (using Representative Concentration Pathways (RCPs), Global Circulation Models (GCMs), and a no climate change (NoCC) scenario), and scenarios of socioeconomic development (using Shared Socioeconomic Pathways (SSPs)) and agriculture sector investments, for a total of 36 scenarios. Detailed descriptions of RCPs, GCMs, and SSPs are available from Moss et al. (2010), O’Neill et al. (2017), and Navarro-Racines et al. (2020), respectively. These core scenarios are representative of possible futures for the global food system based on the range of core driver assumptions available in the IMPACT model. This analysis uses the number of people facing chronic hunger as the core indicator of climate change impacts and adaptation. Chronic hunger is calculated based on per capita calorie availability (including access via international trade) and minimum dietary energy requirements, following a methodology equivalent to the Food and Agriculture Organization’s (FAO) prevalence of undernourishment (PoU) indicator (details in Appendix D and Robinson et al. 2015). The target for adaptation in this analysis is to offset the effects of climate change by making investments that reduce the number of hungry people projected in 2050 to the same level that would be achieved in the absence of climate change. We detail here methods for the key advancements of this research; additional core methods used in support of this study are documented in detail in Rosegrant et al. (2017) and Mason-D’Croz et al. (2019). INVESTMENT COSTS Costs of investments in agricultural research, water management, and infrastructure are estimated as detailed in Rosegrant et al. (2017) and Mason-D’Croz et al. (2019). The lagged effects of investments in agricultural research are estimated using a perpetual inventory method (PIM) analogous to that used for physical capital (Esposti and Pierani 2003), with supporting data from a variety of recent resources. The underlying assumption behind R&D investment and Table 1 SUMMaRY DeSCRIPTION OF ReFeReNCe aND INVeSTMeNT SCeNaRIOS Scenario Description Abbreviation Reference With no climate change (NoCC; constant 2005 climate with various Shared Socioeconomic Pathways (SSPs)) REF-NoCC* With climate change (CC; combinations of SSPs and Representative Concentration Pathways (RCPs) across a range of General Circulation Models (GCMs)) REF-CC* Agricultural R&D Increased research and development (R&D) investment across the CGIAR portfolio plus faster and more efficient adoption of new technologies HIGH+RE Water Management Expansion of irrigated area coupled with increased water use efficiency IX+WUE Improved soil water-holding capacity SWHC Infrastructure Infrastructure improvements to improve market efficiency through the reduction of transportation costs and marketing margins (rail, road, port, and electrification) RMM Comprehensive A combination of the Agricultural R&D, Water Management, and Infrastructure scenarios described above COMP Note: *In this study, the combination that defines the REF-NoCC scenario is the Shared Socioeconomic Pathway “2” (SSP2) alone, while the REF-CC scenario is the combination of SSP2 with RCP8.5 via the UK Met Office Hadley Centre Earth System Model (HGEM) GCM. 10 CLIMATE CHANGE AND HUNGER productivity is that a flow of R&D investments creates a stock of knowledge that yields returns into the future. Information needed to construct such a stock includes the flow of gross R&D investment to determine how fast it builds the stock of knowledge, and how fast knowledge becomes obsolete, which is captured by a decay or depreciation rate. Unit cost estimates from existing literature (FAO 2016; Inocencio et al. 2007; IWMI 2012; Rosegrant et al. 2018) are used to calculate investments in water management (irrigation expansion, water use efficiency improvements, and improvements in rainwater harvesting and soil water holding capacity) and infrastructure (improvements to transportation infrastructure and increased rural electrification). With available current data, final cost estimates are evaluated at regional aggregations of East Asia and Pacific (EAP), South Asia (SAS), Africa south of the Sahara (SSA), Middle East and North Africa (MENA), and Latin America and the Caribbean (LAC). Only developing countries are included. To support this research and review the latest advances in the field, we held a technical consultation (Improving Cost Data and Methods for Evaluating the Cost-Effectiveness of Agriculture and Food Systems Interventions) in June 2019 with a range of subject matter experts in the field of costing investments in the agriculture sector. This helped us focus the sensitivity test of key parameters in the PIM that embody the greatest amount of uncertainty (reference and investment scenario elasticities for international and national investment levels to knowledge stock accumulation and knowledge stock depreciation rates). SENSITIVITY TESTING The primary suggestion from the expert technical workshop was to run sensitivity tests across different key dimensions of our methodology. Part of this was addressed using the 36 different scenarios across the range of IMPACT model drivers that show outcomes for the population at risk of hunger across a range of possible futures for the global food system, as noted above. Next, we conducted sensitivity simulations on key parameter assumptions in the costing models for agricultural R&D (using the PIM) and for water management and infrastructure development (unit costing). The sensitivity testing of the PIM costing methodology entailed running simulations with combinations of adjustments in three sets of core parameters (Table 2) that together determine the effectiveness and longevity of investments in agricultural R&D. Adjustments of +/−20 percent were made to: 1. elasticities relating investment levels to knowledge stock accumulation in both national and international agricultural research systems in the business-as-usual (BAU) scenario; 2. elasticities relating investment levels to knowledge stock accumulation in both national and international agricultural research systems in the investment scenarios; and 3. rates of knowledge stock depreciation in the BAU and investment scenarios. Table 2 PIM COSTING MeTHODOlOGY SeNSITIVITY SIMUlaTION SeTUP aCROSS CORe R&D COST-RelaTeD PaRaMeTeRS (as percentages of calibrated values for the SSP2 reference case) Cost Sensitivity Simulation Sensitivity Adjustments Knowledge stock Elasticities of investment level to knowledge stock accumulation in both national and international agricultural research systems Depreciation rate Reference scenario Investment scenarios All-1 80% 80% 80% All-2 (BAU) 100% 100% 100% All-3 120% 120% 120% DepRate-4 80% DepRate-5 120% ElasBase-6 80% ElasBase-7 120% ElasScnr-8 80% ElasScnr-9 120% APPROACH AND METHODS 11 Ranges of elasticities across the sensitivity simulations are similar to the bounding exercise conducted by Baldos, Fuglie, and Hertel (2020). In addition, sensitivity bounds for water and infrastructure investments were incorporated into this analysis. This was accomplished by making adjustments of +/−15 percent in the unit cost assumptions derived from the most up-to-date sources for investments in water management (irrigation expansion, water use efficiency, soil water-holding capacity) and infrastructure (road, rail, port, and electrification improvement/expansion). The range of reference and investment scenarios and cost sensitivity simulations included in this study enables presentation of a range of results. There is a reference value that corresponds to the calibrated “best” estimate based on available data and expert opinion, and a “plausible interval” that can be interpreted similarly to a confidence interval, so we are highly confident that the true value lies within the interval shown. For clarity, Box 1 includes an explanation of how we refer to the various scenarios and costing simulations in this study. ASSESSMENT OF TRADE-OFFS Policymakers faced with making investment decisions inevitably must consider trade-offs across multiple dimensions. This work incorporates a range of methods and indicators to enable quantitative assessments of potential trade-offs between competing and complementary multidimensional targets, such as the United Nations SDGs (UN 2015). We focus here on the SDGs for poverty (SDG 1), hunger (SDG 2), and water (SDGs 6 and 12), while further indicators are presented in Rosegrant et al. (2017). The SDGs and specific targets evaluated under the alternative investment trade-offs are: f SDG 1 — No Poverty, End poverty in all its forms everywhere • Target 1.1 — Eradicate extreme poverty bOx 1 ReFeRRING TO SCeNaRIOS aND SIMUlaTIONS IN THIS STUDY 1. The “reference” scenario is the core scenario against which most comparisons are made unless explicitly stated otherwise. In this report, this corresponds to the scenario using driver assumptions drawn from SSP2 and RCP8.5 with the UK Met Office Hadley Centre Earth System Model (HGEM) GCM. In certain cases, we add the term “climate change” for clarity. The abbreviation for this scenario is REF-CC. 2. There is also a “no climate change” version of the reference scenario that uses the SSP2 socioeconomic assumptions and a constant 2005 climate, but this is referred to explicitly in all cases and used only to estimate the impact of climate change. The abbreviation for this scenario is REF-NoCC. 3. In a few situations we need to specify scenarios that either include or do not include climate change but are not associated with the reference; these are referred to as CC and NoCC, respectively. 4. The investment scenarios build upon the reference scenario (with adjustments for alignment with specific SSPs and RCPs) and are referred to explicitly as shown in Table 1 with their appropriate SSP and RCP specifications. 5. The 36 scenarios that were run across the IMPACT model driver dimensions and investment scenarios (listed in Table 1) show ranges of results directly from the IMPACT model; in this case the primary result of interest is the population at risk of hunger, but other results are included to broaden the discussion. 6. The 8 sensitivity simulations run on the PIM costing method (noted in Table 2 along with the reference case “All-2”) show the ranges of costing results for the REF-CC and HIGH+RE scenarios. The HIGH+RE costing results are also incorporated in the COMP scenario. 12 CLIMATE CHANGE AND HUNGER • Target 1.2 — Reduce poverty by at least 50 percent f SDG 2 — Zero Hunger, End hunger, achieve food security and improved nutrition, and promote sustainable agriculture • Target 2.1 — Universal access to safe and nutritious food • Target 2.3 — Double the productivity and incomes of small-scale food producers • Target 2.4 — Sustainable food production and resilient agricultural practices f SDG 6 — Clean Water and Sanitation: Ensure access to water and sanitation for all • Target 6.4 — Increase water use efficiency and ensure freshwater supplies • Target 6.5 — Implement integrated water resources management f SDG 12 — Responsible Consumption and Production: Ensure sustainable consumption and production patterns • Target 12.2 — Sustainable management and use of natural resources SDG 2 has several targets that overlap with other SDGs, such as Target 2.3, which addresses incomes that are also an integral component of SDG 1, and Target 2.4 on sustainable food production, also addressed under SDG 12. We take a narrow view of SDG 2 to focus on hunger and agricultural production/supply and leave these overlapping targets to their respective SDGs (1, 6, and 12). In the Findings section below, we refer to key components that are explicitly modeled in the IMPACT framework and are aligned with SDG targets. These include: changes in aggregate country income (“GDP”) under SDG 1; total agricultural supply (“Ag Supply”), population at risk of hunger (“Hunger”) using an equivalent of FAO’s PoU calculation, and per capita calorie availability (“Per Capita KCAL Availability”) under SDG 2; and blue water use and efficiency measures and irrigation water supply reliability (“Blue Water Use,” “Irrigation Water Reliability,” and “Blue Water per KCAL”) under SDG 12. APPROACH AND METHODS 13 FINDINGS POPULATION AT RISK OF HUNGER ACROSS REFERENCE AND INVESTMENT SCENARIOS IN IMPACT MODEL In the absence of climate change, the number of people who face chronic hunger would be expected to decline by more than one-half between 2010 and 2050, led by EAP and SAS (Figure 1). Climate change is projected to slow this progress, with an additional 78 million people facing chronic hunger in 2050 (relative to the REF-NoCC scenario) — over one-half of them in SSA. Impacts vary across the range of climate and socioeconomic scenario specifications run in this analysis with high population, low income growth, and rapid climate change being particularly impactful, resulting in an increase above 2010 levels under climate change in all regions except EAP. Levels of hunger can rise even without climate change in some regions and scenarios, for example, under high population growth coupled with low income growth (SSP3), especially in SSA. In LAC and MENA, hunger rates are low enough that population growth can drive an increase in the number of people at risk of hunger regardless of climate change. Further progress against chronic hunger in LAC and MENA is dependent on drivers outside of IMPACT’s food system model of the population at risk of hunger (such as health and education systems). As modeled here, increased investments in the agriculture sector can more than compensate for climate change effects on the number of hungry people in the world. Improvements in the global situation for the population at risk of hunger under a NoCC future lead to at least an 18 percent reduction in the number of hungry people in 2050 compared to 2010 and, at best, a two-thirds reduction. Under the CC scenarios, the change in the total population at risk of hunger compared to 2010 ranges from an almost 10 percent increase — in the worst situation — to a three-fifths reduction under SSP1. Comparing the CC and NoCC scenarios in 2050 (instead of in 2010), the increase in the population at risk of hunger in 2050 due specifically to climate change is at least 16 percent, and up to one-third greater. SSA and SAS are the regions where the effects of climate change on the population at risk of hunger vary most widely across socioeconomic and climate scenarios (that is, they have the greatest minimum/maximum ranges among the aggregate regions presented). The range around the reference scenario in SAS is −28 percent to +174 percent, while in SSA it is −48 percent to +94 percent. The COMP investment scenario more than offsets the climate effects modeled here, with an 18 percent to 13 percent reduction in the global population at risk of hunger compared to a future without climate change (NoCC). It is useful to separate the effects of the various socioeconomic, climate, and investment assumptions underlying the different scenarios. Socioeconomic assumptions play a critical role in scenario outcomes (Figure 2a). Population growth (slowest in SSP1 and fastest in SSP3) is an important driver, but per capita income growth (fastest in SSP1 and slowest in SSP3) is an even stronger force for improving well-being across the different regions. SSP2, which is the “BAU” future used in this study, shows large reductions in hunger across the globe, especially in EAP and SAS, but persistence of hunger in SSA. Under SSP2 the global population at risk of hunger is cut in half between 2010 and 2050. SSP1 is a more favorable set of assumptions, leading to a nearly two-thirds reduction in the global population at risk of hunger over this same period, primarily by reducing incidence in SSA and across Asia. Under SSP3 hunger increases across the globe in comparison with SSPs 1 and 2, tied to slower growth in per capita incomes and accelerated population growth, especially in SSA, which sees a 36 percent increase in the number of hungry people from 2010 to 2050. Compared to 14 SSP2, the socioeconomic conditions in SSP3 lead to 277 million more people at risk of hunger in 2050 while the SSP1 scenario leads to a 130-million-person reduction. Across the five GCMs used to implement the RCP8.5 emissions pathway (Figure 2b), the global change in population at risk of hunger in 2050 compared to a REF-NoCC scenario in 2050 ranges from a 19 percent increase to a 4 percent decrease (78 million more hungry to 15 million fewer) in UK Met Office Hadley Centre Earth System Model (HGEM) and Geophysical Fluid Dynamics Laboratory (GFDL), respectively. SSA is the region most sensitive to the different conditions encountered across the GCMs, accounting for the majority of change when compared to the REF-NoCC scenario. Not surprisingly, results also vary significantly across climate pathways. Impacts under RCP4.5 are about half as severe as those for RCP8.5 under SSP2 as modeled with the HGEM GCM (Figure 2c). RCP4.5 sees only a 9 percent increase in the population at risk of hunger, compared to 19 percent under RCP8.5. The impact of RCP8.5 on the number of hungry people ranges from a 16 percent increase to a full one-third increase under SSP1 and SSP3, respectively. Under the socioeconomic conditions in SSP3, the world appears more susceptible to climate change impacts on food security. SAS, for example, FIGURe 1 POPUlaTION aT RISK OF HUNGeR IN 2010 aND 2050, globally and by region in the reference (with and without climate change) and investment scenarios Note: World includes all countries; regions include only developing countries. Data shown are a reference estimate (columns) with the minimum and maximum of a plausible interval (whiskers) using methods described. 1,000 900 800 700 600 500 400 300 200 100 0 World East Asia and Pacific Latin America and the Caribbean Middle East and North Africa South Asia Africa south of the Sahara Po p ul at io n at r is k o f h un g er (m ill io n) 201


INTRODUCTION
Climate change is a growing challenge for food systems around the world, presenting both changing trends and increased variability in temperatures and precipitation.
These changes threaten agricultural productivity, food and nutrition security, health, and the potential for achieving key Sustainable Development Goals (SDGs) for poverty, hunger, and water, among others. Despite these challenges, however, appropriate investments in the agriculture sector can help people and food systems adapt by increasing productivity, resilience, and resource-use efficiency. To help policymakers and other decision-makers evaluate alternative adaptation strategies, this report assesses the cost of adaptation to climate change across a range of future climate scenarios and investment options, with a particular focus on adaptation to offset climate change impacts on hunger.
Climate change, its impacts on agriculture, and measures to address those impacts are all highly complex phenomena. Estimating the cost of such measures is correspondingly difficult. Rather than attempting to estimate the total cost of adaptation, Nelson et al. (2009) focused on one particular dimension of these phenomena: the cost of selected agricultural investment strategies to offset the adverse impact of climate change on humans as proxied by future projections of childhood undernutrition. The assumption was that if we can avert impacts on the most vulnerable, then we will likely have taken care of the rest of the population as well. This report does the same, but with an updated methodology and a broader focus on an estimate of the total population at risk of chronic hunger. We draw on recent analyses by Rosegrant et al. (2017), Rosegrant et al. (2021), and Mason-D' Croz et al. (2019), which in turn built on Nelson et al. (2009).
Other studies have used a variety of approaches to estimate the cost of climate change adaptation. Country-and sector-level studies in the 1990s assessed adaptation costs based on the estimated damages from climate change impacts (Tol 2005;Tol, Fankhauser, and Smith 1998). The first global estimates of adaptation costs shared a methodology developed by the World Bank (2006) that estimated the fraction of investments flowing to developing countries deemed sensitive to climate change impacts and then used a "markup" factor to represent the costs of "climate-proofing" those investments. 1 These studies did not specify a particular time horizon or climate scenario, but the use of current investment flows as a starting point suggests that they focused on shortterm adaptation costs.
Estimates vary widely due to the lack of empirical data on climate sensitivity and the costs of climateproofing. Studies by UNFCCC (2007) and Forstater and Zadek (2009) specified a time horizon of 2030 and included more detailed assessments of investment and financial flows for each sector. A 2010 review of published studies on the global cost of adaptation found that estimates ranged between US$25 and US$100 billion per year for the period 2015-2030 (Agrawala and Fankhauser 2008;Fankhauser 2010). These figures referred to the costs of adapting to a warming of 2-3 degrees Celsius and were estimated to rise exponentially for higher temperatures. A subsequent analysis by the World Bank (2010) estimated the cost at US$70-100 billion annually for developing countries for the period 2010 to 2050. These were the most recent comprehensive estimates as of the Intergovernmental Panel on Climate Change's (IPCC) Fifth Assessment Report (Chambwera et al. 2014), and arguably the most robust (O'Garra and Mourato 2016). A recent study by Baldos, Fuglie, and Hertel (2020) focused on national public spending on R&D at aggregate regional levels; they estimated that climate adaptation costs that offset climate impacts on crop yields would cost a total of US$187 to US$1,384 billion from 2020 to 2040.
In contrast to these studies, Nelson et al. (2009) focused more specifically on a detailed analysis of the impacts of climate change on: (1) agricultural productivity and related outcomes (including child undernourishment) to 2050, by country and commodity; and (2) the cost of selected agricultural investment strategies in developing countries (namely agricultural research, soil and water management, and rural infrastructure) to counter those impacts. More recent work by Rosegrant et al. (2017) and  extended that approach, incorporating updated models and data and an emphasis on looking at trade-offs and complementarities across different investment strategies. Additional analyses led by colleagues from across the CGIAR broadened these two studies by taking deeper looks at particular commodities and development issues, such as fish, livestock, cereals, and roots, tubers, and bananas (Chan et al. 2019;Enahoro et al. 2019;Kruseman et al. 2020;Petsakos et al. 2019), and agriculture sector income and employment ).

MODELING FOOD SYSTEM FUTURES UNDER CLIMATE CHANGE AND ADAPTATION INVESTMENT SCENARIOS
This analysis builds on the recent work by Rosegrant et al. (2017) and others by analyzing a more complete range of future scenarios along with sensitivity simulations on key costing parameters. The goal is to help establish a more robust foundation for the assessment of the cost of adaptation to climate change for the food system in the developing world. The core output of this study is a direct analog and update to the cost of climate change adaptation presented in Nelson et al. (2009). See Appendix A for the core details for comparison.
To simulate changes in the global food system and assess investment scenarios for their effectiveness to offset climate change impacts out to mid-century, we used IFPRI's International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) . This modeling framework links climate, crop, water, and economic models to analyze scenarios of future change in agricultural production, consumption, prices, and trade at national, regional, and global scales. The methodology behind the IMPACT modeling framework has evolved and been refined over the past couple of decades in collaboration with other leading research groups around the world, particularly with partners across CGIAR and in the Global Economics group of the Agricultural Model Intercomparison and Improvement Project (AgMIP, https://agmip.org/global-economics/). The IMPACT modeling framework has been applied extensively in numerous peer-reviewed publications and decisionmaking fora, such as international development organizations, national governments, foundations and grant-making institutions, and across CGIAR itself (https://www.ifpri.org/project/ifpri-impact-model). IMPACT produces reference scenarios to 2050 assuming no climate change as well as different levels of climate change and socioeconomic assumptions from the IPCC, capturing a wide range of possible climate and socioeconomic futures (Table 1). In conjunction with these reference scenarios, we then analyzed scenarios of how plausible investment options -including investments in agricultural R&D, water management, and infrastructure (determined in collaboration with experts across the CGIAR) -could help offset the potential impacts of climate change.
Many complex connections exist between climate change and hunger (Mbow et al. 2019;Beach et al. 2019;Nelson et al. 2018). This analysis focuses on a subset of those. In the most basic sense, climate change, by and large, negatively affects agricultural production by reducing yields. This reduced productivity then reduces availability and drives prices upward, ultimately resulting in increased hunger. The investments modeled here generate economic benefits through several pathways. Specific details are summarized in Appendix D, but in general, increased agricultural R&D boosts crop and livestock yields, reduces food prices, and increases farm income and economywide gross domestic product (GDP) through multiplier effects on the nonagriculture sectors. Increasing incomes, in turn, shift consumers toward different diets, with consumption of usually more food and more diverse food types. Irrigation and water use efficiency investments increase crop yields and reduce prices, and generating higher incomes. Enhanced rural infrastructure reduces postharvest losses and marketing margins, improving the profitability of farm production, and boosting supply to consumers for any given level of production. These effects also increase farm and economywide income, which increases food consumption and reduces hunger. While the investment alternatives modeled in this study are extensive, this list is not exhaustive and the focus in this study is on classes of investments that are commonly recognized across the agricultural development community and that can be modeled directly in IMPACT.
To expand the representation across the range of core drivers, we extended the original scenario set in Rosegrant et al. (2017) to show outcomes across different combinations of climate change assumptions (using Representative Concentration Pathways (RCPs), Global Circulation Models (GCMs), and a no climate change (NoCC) scenario), and scenarios of socioeconomic development (using Shared Socioeconomic Pathways (SSPs)) and agriculture sector investments, for a total of 36 scenarios. Detailed descriptions of RCPs, GCMs, and SSPs are available from Moss et al. (2010), O'Neill et al. (2017), and Navarro-Racines et al. (2020, respectively. These core scenarios are representative of possible futures for the global food system based on the range of core driver assumptions available in the IMPACT model. This analysis uses the number of people facing chronic hunger as the core indicator of climate change impacts and adaptation. Chronic hunger is calculated based on per capita calorie availability (including access via international trade) and minimum dietary energy requirements, following a methodology equivalent to the Food and Agriculture Organization's (FAO) prevalence of undernourishment (PoU) indicator (details in Appendix D and Robinson et al. 2015). The target for adaptation in this analysis is to offset the effects of climate change by making investments that reduce the number of hungry people projected in 2050 to the same level that would be achieved in the absence of climate change.
We detail here methods for the key advancements of this research; additional core methods used in support of this study are documented in detail in Rosegrant et al. (2017) and Mason-D'Croz et al. (2019).

INVESTMENT COSTS
Costs of investments in agricultural research, water management, and infrastructure are estimated as detailed in Rosegrant et al. (2017) and Mason-D'Croz et al. (2019). The lagged effects of investments in agricultural research are estimated using a perpetual inventory method (PIM) analogous to that used for physical capital (Esposti and Pierani 2003), with supporting data from a variety of recent resources. The underlying assumption behind R&D investment and productivity is that a flow of R&D investments creates a stock of knowledge that yields returns into the future. Information needed to construct such a stock includes the flow of gross R&D investment to determine how fast it builds the stock of knowledge, and how fast knowledge becomes obsolete, which is captured by a decay or depreciation rate. Unit cost estimates from existing literature (FAO 2016;Inocencio et al. 2007;IWMI 2012;Rosegrant et al. 2018) are used to calculate investments in water management (irrigation expansion, water use efficiency improvements, and improvements in rainwater harvesting and soil water holding capacity) and infrastructure (improvements to transportation infrastructure and increased rural electrification). With available current data, final cost estimates are evaluated at regional aggregations of East Asia and Pacific (EAP), South Asia (SAS), Africa south of the Sahara (SSA), Middle East and North Africa (MENA), and Latin America and the Caribbean (LAC). Only developing countries are included. To support this research and review the latest advances in the field, we held a technical consultation (Improving Cost Data and Methods for Evaluating the Cost-Effectiveness of Agriculture and Food Systems Interventions) in June 2019 with a range of subject matter experts in the field of costing investments in the agriculture sector. This helped us focus the sensitivity test of key parameters in the PIM that embody the greatest amount of uncertainty (reference and investment scenario elasticities for international and national investment levels to knowledge stock accumulation and knowledge stock depreciation rates).

SENSITIVITY TESTING
The primary suggestion from the expert technical workshop was to run sensitivity tests across different key dimensions of our methodology. Part of this was addressed using the 36 different scenarios across the range of IMPACT model drivers that show outcomes for the population at risk of hunger across a range of possible futures for the global food system, as noted above. Next, we conducted sensitivity simulations on key parameter assumptions in the costing models for agricultural R&D (using the PIM) and for water management and infrastructure development (unit costing).
The sensitivity testing of the PIM costing methodology entailed running simulations with combinations of adjustments in three sets of core parameters ( Table 2) that together determine the effectiveness and longevity of investments in agricultural R&D. Adjustments of +/−20 percent were made to: 1. elasticities relating investment levels to knowledge stock accumulation in both national and international agricultural research systems in the business-as-usual (BAU) scenario; 2. elasticities relating investment levels to knowledge stock accumulation in both national and international agricultural research systems in the investment scenarios; and 3. rates of knowledge stock depreciation in the BAU and investment scenarios. All-2 (BAU) 100% 100% 100% All-3 120% 120% 120%

ElasScnr-9 120%
Ranges of elasticities across the sensitivity simulations are similar to the bounding exercise conducted by Baldos, Fuglie, and Hertel (2020).
In addition, sensitivity bounds for water and infrastructure investments were incorporated into this analysis. This was accomplished by making adjustments of +/−15 percent in the unit cost assumptions derived from the most up-to-date sources for investments in water management (irrigation expansion, water use efficiency, soil water-holding capacity) and infrastructure (road, rail, port, and electrification improvement/expansion).
The range of reference and investment scenarios and cost sensitivity simulations included in this study enables presentation of a range of results. There is a reference value that corresponds to the calibrated "best" estimate based on available data and expert opinion, and a "plausible interval" that can be interpreted similarly to a confidence interval, so we are highly confident that the true value lies within the interval shown.
For clarity, Box 1 includes an explanation of how we refer to the various scenarios and costing simulations in this study.

ASSESSMENT OF TRADE-OFFS
Policymakers faced with making investment decisions inevitably must consider trade-offs across multiple dimensions. This work incorporates a range of methods and indicators to enable quantitative assessments of potential trade-offs between competing and complementary multidimensional targets, such as the United Nations SDGs (UN 2015). We focus here on the SDGs for poverty (SDG 1), hunger (SDG 2), and water (SDGs 6 and 12), while further indicators are presented in Rosegrant et al. (2017 3. In a few situations we need to specify scenarios that either include or do not include climate change but are not associated with the reference; these are referred to as CC and NoCC, respectively. 4. The investment scenarios build upon the reference scenario (with adjustments for alignment with specific SSPs and RCPs) and are referred to explicitly as shown in Table 1 with their appropriate SSP and RCP specifications. SDG 2 has several targets that overlap with other SDGs, such as Target 2.3, which addresses incomes that are also an integral component of SDG 1, and Target 2.4 on sustainable food production, also addressed under SDG 12. We take a narrow view of SDG 2 to focus on hunger and agricultural production/supply and leave these overlapping targets to their respective SDGs (1, 6, and 12). In the Findings section below, we refer to key components that are explicitly modeled in the IMPACT framework and are aligned with SDG targets. These include: changes in aggregate country income ("GDP") under SDG 1; total agricultural supply ("Ag Supply"), population at risk of hunger ("Hunger") using an equivalent of FAO's PoU calculation, and per capita calorie availability ("Per Capita KCAL Availability") under SDG 2; and blue water use and efficiency measures and irrigation water supply reliability ("Blue Water Use," "Irrigation Water Reliability," and "Blue Water per KCAL") under SDG 12.

POPULATION AT RISK OF HUNGER ACROSS REFERENCE AND INVESTMENT SCENARIOS IN IMPACT MODEL
In the absence of climate change, the number of people who face chronic hunger would be expected to decline by more than one-half between 2010 and 2050, led by EAP and SAS ( Figure 1). Climate change is projected to slow this progress, with an additional 78 million people facing chronic hunger in 2050 (relative to the REF-NoCC scenario) -over one-half of them in SSA. Impacts vary across the range of climate and socioeconomic scenario specifications run in this analysis with high population, low income growth, and rapid climate change being particularly impactful, resulting in an increase above 2010 levels under climate change in all regions except EAP.
Levels of hunger can rise even without climate change in some regions and scenarios, for example, under high population growth coupled with low income growth (SSP3), especially in SSA. In LAC and MENA, hunger rates are low enough that population growth can drive an increase in the number of people at risk of hunger regardless of climate change. Further progress against chronic hunger in LAC and MENA is dependent on drivers outside of IMPACT's food system model of the population at risk of hunger (such as health and education systems).
As modeled here, increased investments in the agriculture sector can more than compensate for climate change effects on the number of hungry people in the world. Improvements in the global situation for the population at risk of hunger under a NoCC future lead to at least an 18 percent reduction in the number of hungry people in 2050 compared to 2010 and, at best, a two-thirds reduction. Under the CC scenarios, the change in the total population at risk of hunger compared to 2010 ranges from an almost 10 percent increase -in the worst situationto a three-fifths reduction under SSP1.
Comparing the CC and NoCC scenarios in 2050 (instead of in 2010), the increase in the population at risk of hunger in 2050 due specifically to climate change is at least 16 percent, and up to one-third greater. SSA and SAS are the regions where the effects of climate change on the population at risk of hunger vary most widely across socioeconomic and climate scenarios (that is, they have the greatest minimum/maximum ranges among the aggregate regions presented). The range around the reference scenario in SAS is −28 percent to +174 percent, while in SSA it is −48 percent to +94 percent. The COMP investment scenario more than offsets the climate effects modeled here, with an 18 percent to 13 percent reduction in the global population at risk of hunger compared to a future without climate change (NoCC).
It is useful to separate the effects of the various socioeconomic, climate, and investment assumptions underlying the different scenarios. Socioeconomic assumptions play a critical role in scenario outcomes (Figure 2a). Population growth (slowest in SSP1 and fastest in SSP3) is an important driver, but per capita income growth (fastest in SSP1 and slowest in SSP3) is an even stronger force for improving well-being across the different regions. SSP2, which is the "BAU" future used in this study, shows large reductions in hunger across the globe, especially in EAP and SAS, but persistence of hunger in SSA. Under SSP2 the global population at risk of hunger is cut in half between 2010 and 2050. SSP1 is a more favorable set of assumptions, leading to a nearly two-thirds reduction in the global population at risk of hunger over this same period, primarily by reducing incidence in SSA and across Asia. Under SSP3 hunger increases across the globe in comparison with SSPs 1 and 2, tied to slower growth in per capita incomes and accelerated population growth, especially in SSA, which sees a 36 percent increase in the number of hungry people from 2010 to 2050. Compared to SSP2, the socioeconomic conditions in SSP3 lead to 277 million more people at risk of hunger in 2050 while the SSP1 scenario leads to a 130-million-person reduction.
Across the five GCMs used to implement the RCP8.5 emissions pathway (Figure 2b Not surprisingly, results also vary significantly across climate pathways. Impacts under RCP4.5 are about half as severe as those for RCP8.5 under SSP2 as modeled with the HGEM GCM ( Figure 2c). RCP4.5 sees only a 9 percent increase in the population at risk of hunger, compared to 19 percent under RCP8.5. The impact of RCP8.5 on the number of hungry people ranges from a 16 percent increase to a full one-third increase under SSP1 and SSP3, respectively. Under the socioeconomic conditions in SSP3, the world appears more susceptible to climate change impacts on food security. SAS, for example,

FIGURe 1 POPUlaTION aT RISK OF HUNGeR IN 2010 aND 2050,
globally and by region in the reference (with and without climate change) and investment scenarios Note: World includes all countries; regions include only developing countries. Data shown are a reference estimate (columns) with the minimum and maximum of a plausible interval (whiskers) using methods described. sees a 48 percent increase in the population at risk of hunger compared to a less than 10 percent increase under SSPs 1 and 2. Finally, the investment scenarios contribute to offsetting the effects of climate change as represented by the number of hungry people (Figure 2d). The COMP scenario fully offsets the impact of climate change on the population at risk of hunger, driven primarily by investments in R&D in the international agricultural research system, that is, CGIAR. In these scenarios, we assume that investments in national agricultural research systems (NARS) remain at the levels modeled in the reference scenario. While water and infrastructure investments make only small contributions to reducing hunger on their own, they provide a critical foundation of environmental and economic sustainability that underpins longer-term and broader food system resilience.
Underlying the total population at risk of hunger seen in Figure 2 is the share of the population at risk of hunger, which is directly modeled (see Appendix B). While the shares of the population at risk nearly all decline from 2010 to 2050, in a few regions population growth will overwhelm this progress and the projections show increasing numbers of hungry people.

THE COST OF ADAPTATION
Thus far we have described the impact of alternative investments on hunger in the context of socioeconomic and climate change. We are not able to model those impacts directly as functions of dollars invested. Instead, investments are characterized in terms of their intermediate impacts on agricultural productivity (via increases in international agricultural research and improvements in water management) and on marketing costs (via infrastructure improvements). We focus here on R&D in the international agricultural system and set aside investments in NARS, which would require complementary development efforts and increased investments beyond the levels specified in the reference scenario. Next, we turn to estimating the costs of achieving those in-creases in international agricultural research and improvements in water management and infrastructure.
First, we consider agricultural research. Investment in international agricultural research via CGIAR increased from US$454 million per year in 2005 to almost US$900 million per year in 2015. Assuming in our reference scenario that that historical progression continues as detailed in Rosegrant et al. (2017) Table 3).
While investments in agricultural R&D go a long way toward offsetting the impact of climate change on hunger, complementary investments in water and infrastructure are necessary for robust, resilient development. This is congruent with Nelson et al. (2009), who argue that the best investments for adapting to climate change are simply good development investments, which already address resilience. Table 3 presents the estimates for the cost of R&D, water, and infrastructure investments. Total public investment in these three areas in developing regions is projected to average US$42.56 billion per year between 2015 and 2050 in the reference scenario, with an additional US$25.47 billion per year needed (in the COMP scenario) to offset the impacts of climate change on hunger. As noted above, most of this reduction in hunger is achieved by increased investment in agricultural R&D. Public agricultural R&D spending is projected to average US$8.08 billion per year between 2015 and 2050 in the reference scenario, with an additional US$1.97 billion per year needed to help achieve hunger reduction goals.
Most global agricultural R&D spending, both in the reference case and especially in the investments incremental to reference levels, is focused on SSA. The COMP scenario effectively doubles the CGIAR budget globally (with a range of a 92-148 percent increase); most of the increase is in SSA, where the budget could nearly triple from the reference scenario in the most expensive sensitivity simulation. The CGIAR budget increases the least in EAP (about a 25 percent increase over the reference scenario), while the rest of the world (SAS, MENA, LAC) hovers between a 40 percent to 70 percent increase over the reference scenario. The cost sensitivity simulations for agricultural R&D investments show a range around the reference and COMP scenario cost estimates of +9 percent to −6 percent and +8 percent to −11 percent for the projected value of CGIAR and NARS investments, respectively, in the reference scenario and +41 percent to −24 percent for the projected value of CGIAR investments in the COMP scenario.
Over the longer term, and going beyond the hunger metric, complementary investments in water management and supporting infrastructure are also necessary. Additional water and infrastructure investments are estimated to be more expensive than agricultural R&D, at US$12.7 billion and US$10.8 billion per year, respectively, but they address key gaps that expand over time -such as overused and undermaintained water delivery and drainage systems and inadequate and degrading municipal transport and electrical services -to ensure food system sustainability, efficiency, and reliability. Investments to expand irrigated area and improve water use efficiency and soil water-holding capacity will become increasingly important as climate change makes water resources scarcer in some areas and more erratic in others. Infrastructure investments will be needed to increase the efficiency and reliability of the food system and ensure that rural poor populations are not left behind in areas where the effects of climate change will be felt more severely. Water and infrastructure investments dominate the cost of the COMP scenario, while infrastructure accounts for much of the reference costs. Reference spending on water investments is mostly in SSA, EAP, and SAS, while incremental investments are concentrated in EAP and SAS. Infrastructure investments are especially focused on EAP, with some in SAS. Infrastructure spending is quite low for SSA in the reference scenario, so increases there will be important.
The sensitivity testing range for water and infrastructure investments is +/−15 percent, which is aligned with the range we see in the costing sensitivity simulations for the agricultural R&D results. Total water investments effectively double from the reference to the investment scenarios, up from a range of US$9.40 to US$12.72 billion per year by an increment of US$10.79 to US$14.59 billion per year in the investment scenario. Infrastructure investments in the COMP scenarios increase by US$9 to US$12 billion over the reference scenario, with most of the Note: Regions only include developing countries. Data shown are a reference estimate with the minimum and maximum of a plausible interval (in parentheses) using methods described. spending in EAP and SAS, though SSA sees the largest increase (by a factor of 10) from a smaller starting point.

TRADE-OFFS
The alternative investment scenarios entail a wide range of costs and generate a wide range of outcomes for the SDGs. In some cases, the impacts on different objectives are complementary; in others, significant trade-offs arise between objectives. Table 4 summarizes key impacts of each scenario across SDGs compared to the REF-CC scenario. Investments in agricultural R&D in the HIGH+RE scenario are relatively low cost and generally offer moderate improvements in income by 2030, but larger improvements by 2050. The HIGH+RE investments make important contributions to increasing food availability and reducing hunger, while also improving the water situation around the globe. The scenario combining irrigation expansion and increased water use efficiency (IX+WUE) offers greater reductions in water use (with improved efficiency) at a greater cost, but comes with smaller improvements in income, food supply, and food security. Improved market access through reduced marketing costs (RMM) increases income, supply, and food security, but at the cost of water with no improvements to use and decreased irrigation water reliability. These outcomes highlight the importance of a mixed portfolio of investments that combines productivity enhancement with improved water resource management and market access. The comprehensive scenario (COMP) achieves significant improvements in all outcome areas, particularly in 2050, but comes at a significantly higher cost.
The plausible ranges of scenario impacts (Table  5) also include important details to consider. The potential income gains across all the investment scenarios are quite large, with around a 10 percent increase by 2030 and more than a 30 percent increase by 2050; however, at the lower end of this range we see rather modest income improvements to even slightly negative effects with IX+WUE. The HIGH+RE carries with it important improvements for SDG 2 even at the lowest end of the plausible range. The other scenarios have a lot of potential to improve the hunger situation, but also show that these investments might not result in strong reductions in hunger. Water use and efficiency (SDGs 6 and 12) is clearly improved under the IX+WUE investment scenario, but HIGH+RE makes important contributions in this realm, too. The other particular scenarios (SWHC and RMM) show mixed improvements to water use and efficiency.
Beyond the SDGs, results are varied for the different commodities for regions across the different scenarios, which may be of interest to policymakers (Appendix C).

DISCUSSION AND CONCLUSIONS
KEY DIFFERENCES FROM OTHER STUDIES Nelson et al. (2009) previously analyzed the cost of adapting to climate change by increasing investments in agriculture and food systems in the developing world. They estimated those additional annual costs to be around US$1 billion for agricultural research and US$3 billion each for irrigation and rural roads. The current analysis builds on the same general approach but generates corresponding estimates roughly 2-4 times higher: around US$2 billion for agricultural R&D, US$13 billion for water management, and US$11 billion for rural infrastructure. These differences arise for several key reasons: • The current analysis uses an updated modeling specification in conjunction with a variety of updated modeling assumptions, including updated versions of the IMPACT model parameter specifications (for supply/demand elasticities, embedded technological progress assumptions, and geography specifications, among others), and related modeling methods within the IMPACT framework (crop and water models, in particular).
• The current analysis uses more recent climate 2 and socioeconomic data 3 based on the IPCC's Fifth Assessment Report (IPCC 2014).
• The current analysis uses an updated methodology based on the perpetual inventory method to capture the lagged effects of R&D investment on productivity.
As noted at the beginning of this report, earlier studies estimated the cost of adaptation to climate change at US$75-100 billion. This is considerably higher than our estimates, but we emphasize that our analysis looks only at the cost of adaptation in a single dimension -in terms of offsetting the adverse impacts of climate change on hunger.
Other studies have explored investments to reduce hunger, particularly in Africa south of the Sahara. These include studies by Laborde et al. (2016) and FAO, IFAD, and WFP (2015). Direct comparison is difficult because these other analyses focused on different aspects of the question, used different methods, and did not address the effects of climate change. Laborde et al. (2016) considered the minimum cost to eliminate hunger in vulnerable households; FAO, IFAD, and WFP (2015) assessed the cost to lift all households above the poverty line assuming it will be equivalent to a bulk cash trans-fer. Both studies focused on a reference case that is equivalent to our REF-NoCC, while our analysis specifically focuses on climate change effects. Other key differences between the various analyses are summarized in Fan et al. (2018) and .

CAVEATS, LIMITATIONS, AND FUTURE ADVANCES
While Rosegrant et al. (2017) made important advances in the estimation of costs and impacts of investments for agriculture, it is important to recognize several limitations of the current study and the opportunities to further advance the analysis of agriculture and food system investments for climate change adaptation.
The methodology underlying this analysis is still evolving. Both specific algorithms and parameterization of various factors are active fields of study and will likely see improvements in the future.
The results arrived at here might underestimate full costs. For example, trade plays an important role in adaptation in these analyses but is modeled in a relatively frictionless way in the IMPACT model framework (Janssens et al. 2020). Trade barriers will increase adaptation costs, especially for net importers. Further work is needed to better reflect the reality of evolving trade policies and trade interrelationships.
While the IMPACT modeling framework used here included feedback from a full economy model on total GDP, modeling improvements are needed to more fully integrate feedback from other sectors of the economy (energy, labor, etc.), private sector investment, and other complementary development strategies -such as social safety nets -for any particular climate scenario. These dimensions could either increase or decrease the total cost of adaptation; a fully integrated model is required to determine the direction and magnitude of change attributable to these different factors.
This analysis focuses on projected changes in climate trends (specifically temperature and precipitation) but does not capture the effects of changes in climate variability or extreme events (or other aspects of climate change such as sea level rise or melting glaciers).
This analysis focuses on just one dimension and indicator of adaptation to climate change: the number of people facing chronic hunger. Though we added breadth to this in our additional SDG tradeoff assessment, other SDG targets could be assessed with further development of modeling tools that can explicitly address the multitude of SDG dimensions. 4 A full analysis of adaptation to climate change in agriculture would require including many other social, economic, and environmental dimensions. Several dimensions exist in which to consider tradeoffs across the range of SDGs in any set of scenarios of future shocks and adaptation investments. In addition to the trade-offs presented here, results on several of these additional indicators and their tradeoffs are reported in Rosegrant et al. (2017). There are also many additional pathways of climate change impact on food systems and human well-being (Mbow et al. 2019;Myers et al. 2017;Crimmins et al. 2016), including altered production systems, vectorborne diseases, water source quality, and food safety, among many others.
Finally, this study is not designed to identify the minimum cost of adaptation to climate change. Rather, it examines several alternative investment strategies, their costs, and their impacts in relation to the selected indicator of adaptation. Further analysis is required to provide a more detailed picture of the range of possible options, including a broader set of investment alternatives and other policy approaches that aim at not only supply-side levers but also costeffective ways to affect levels of demand and diet preferences, food waste, and food system "greening," among others, which could ease food system stressors in the face of climate change.

KEY MESSAGES
Planning for an uncertain climate future will require an enabling environment for innovation in agricultural technologies and practices that can be applied across broad geographies and adapted to local requirements. Increased investment and innovation in agriculture need to be complemented by robust rural and pro-poor development strategies. The analysis presented here offers some key messages about the costs and trade-offs of investing in R&D, water management, and infrastructure for climate-change adaptation to help reach the goal of zero hunger: • Climate change is projected to slow progress in reducing hunger to 2050, particularly in Africa south of the Sahara.
• Increased investment in agriculture can support adaptation to climate change. R&D is most cost-effective in offsetting the impacts of climate change as measured by the hunger indicator, with US$1.97 billion in annual incremental investment needed for adaptation by this measure.
• Estimating ranges of results gives some context about the potential for investments to either under-or overshoot highlighted impacts, along with a better idea about costs that should be considered as part of the trade-offs.
• Increased investment in water management and infrastructure is essential to complement and sustain R&D-based gains, especially over the longer term. Other investments will be needed to achieve broader measures of adaptation.
• Alternative investment options involve different synergies and trade-offs: R&D investments offer the largest reductions in hunger but smaller improvements in blue water use and irrigation supply reliability; water management investments offer greater improvements in blue water use and irrigation supply but smaller reductions in hunger; and a comprehensive approach offers improvements in all goals but is much more costly.                    Income, kilocalories, and food security

MinKcal
It should be noted that due to the quadratic nature of this equation it is necessary to apply an upper and lower bound to the share at risk. The lower bound is defined as 0 and the upper bound is 100. Developed countries unsurprisingly have a low share at risk, so for simplicity we treat all countries with less than a 4 percent share at risk of hunger as if they had a 0 percent share of hunger. The relative availability of food was bounded to ensure realistic results on the quadratic curve: when the ratio of calories available to calories required, RelativeKCal, is greater than 1.7, we assume that the share at risk of hunger is effectively 0.

Scenario specifications
(summarized from full description in Rosegrant et al. 2017) Reference Scenario: Agricultural research Improvements in agricultural productivity in the reference scenario are represented by exogenous growth rates for each commodity and country, based on historical trends as well as expert opinion about future changes. We developed an R&D investment-yield model to assess the investment required to achieve projected growth in agricultural productivity.
Investments in research take time to bear fruit, as innovative ideas can take years to be developed and diffused widely. To capture these lags, the investment-yield estimation model is based on the perpetual inventory method (PIM), where research investments contribute to the stock of knowledge over time. Knowledge decays as older technologies become obsolete or irrelevant. Productivity grows if the stock of knowledge grows at a faster rate than the stock of knowledge decays. The lag structure in the PIM used here follows a gamma distribution, where R&D investments reach peak impact 10 years after initial investment and then decline over time to reach zero 10 years after peak impact. With regionally differentiated research elasticities and decay rates, these imputed lag structures would vary by region according to existing R&D capacity and the potential trajectories for each region.
Research capacity varies significantly by region. To reflect these differences, we use elasticities of productivity with respect to research investments from the literature and incorporate spillover effects to represent regional capacity to access and apply outside knowledge. To meet the productivity growth trajectories assumed in the reference scenario, required investment in agricultural R&D by the CGIAR is projected to average US$1.7 billion per year between 2015 and 2050 in real 2005 dollars, while annual national agricultural research systems (NARS) investment in developing countries averages US$6.4 billion (Table D -1). The largest investments are projected in Africa south of the Sahara (SSA) (US$2.2 billion per year) and Latin America and the Caribbean (LAC) (US$1.8 billion per year). In most regions, the larger contribution to agricultural research will come from investments from NARS, except for SSA, where about one-half of the investments will come from the CGIAR.

DVG
literature. While the largest expansion in irrigated area is projected in Asia, the largest investments will be needed in Africa south of the Sahara (SSA) due to the higher costs of expanding irrigation.
Similar to the cost of expanding irrigation, the cost for increased WUE is estimated at the hectare level and specified by region. Investments in improved WUE across the developing world are less than onethird of the investments projected for irrigation expansion, at US$2.2 billion per year. The majority of these investments are projected in East Asia and Pacific (EAP) and South Asia (SAS), which account for almost 80 percent and 77 percent of irrigated area in developing countries in 2010 and 2050, respectively. Baseline investments in soil water management (SWHC) technologies are from limited evidence but based on previous studies we arrive at a set of estimates for the developing world at US$1.3 billion per year.
Reference Scenario: Infrastructure The economic growth assumed in the reference scenario also includes investments in new and maintenance of existing infrastructure, such as electrification and roads. Based on the methodology in Rosegrant et al. (2015), we project reference investments across developing countries totaling US$23.4 billion per year.
Investment Scenarios: R&D for enhanced productivity The HIGH+RE scenario incorporates yield gains from increasing investments in CGIAR R&D and was developed in collaboration with all 15 CGIAR Centers through the Global Futures and Strategic Foresight program. As a starting point, each Center quantified potential yield gains for its respective commodities (including crops, livestock, and fish) in developing countries across major regions of the world (SAS, SSA, LAC, EAP, and MENA) with increased R&D investment. These yield gains were first expressed as potential changes in absolute yield levels and then translated into differential yield growth rates used in the IMPACT modeling framework. The final endogenous yields and output growth generated by the scenarios are functions of interactions between these growth rates and projected changes in prices, demand, and other factors. Increased CGIAR research efficiency accelerates and augments productivity gains so that the yield impact of investments is 30 percent higher and the maximum improvement is achieved earlier than in a business-as-usual (BAU) situation. Research efficiency is gained through advancement in breeding techniques including further advances in genomics and bio-informatics and high throughput gene sequencing; and more effective regulatory and intellectual property rights systems that reduce the lag times from discovery to deployment of new varieties. Table D -2 summarizes the additional investment needs for the productivity enhancement and other alternative investment scenarios, incremental to the reference costs for REF_CC in Table D -1. The additional investment costs associated with these scenarios vary by region and scenario.
Investment Scenarios: Improved water resource management Two alternative scenarios focus on investments and improvements in agricultural water resource management that affect crops and livestock directly through changes in the availability of water, and also livestock indirectly through changes in feed prices. They include accelerated investments in irrigation expansion combined with improved water use efficiency on all irrigated cropland (IX+WUE); and a focus on improvements in rainwater harvesting and soil water holding capacity (SWHC). These scenarios were developed in consultation with the International Water Management Institute (IWMI). The IX+WUE scenario simulates an expansion of irrigated areas in developing countries by 2030, relative to the REF_CC scenario, with changes thereafter following the growth rates in the REF_CC scenario. Irrigation expansion in the IX+WUE scenario is designed to have a neutral effect on total agricultural land across developing countries, with projected irrigated expansion of 20 million hectares roughly offset by a reduction of 22 million hectares of rainfed agriculture. In some regions, irrigation expansion will not completely displace rainfed area and will result in an overall expansion of land, as is the case in SSA, where total harvested area increases by 2.7 million and 4.3 million hectares compared to REF_CC by 2030 and 2050, respectively.
In IX+WUE, improving irrigation efficiency can increase agricultural output while conserving more water. Significant variation arises globally regarding the efficiency of water systems, ranging from highly efficient systems based on high efficiency technologies (for example, sprinklers and drip irrigation), to less efficient systems, such as furrow irrigation, and these system-level efficiencies do not necessarily translate to river basin efficiencies. IMPACT uses the concept of basin efficiency, defined as the ratio of beneficial water depletion (crop evapotranspiration and salt leaching) to total irrigation water depletion at the basin scale. Basin efficiency in future years is assumed to increase at a prescribed rate in a food production unit (FPU) depending on water infrastructure investment and water management improvement in the FPU. For the WUE scenario, basin efficiencies are assumed to increase by 15 percentage points by 2030 and then continue on previous trajectories.
The SWHC scenario simulates the benefits of technologies -such as no-till agriculture and water harvesting -that increase soil water holding capacity or otherwise make precipitation more readily available to plants (that is, effective precipitation). Improvements vary by region and represent the different levels at which these kinds of technologies are currently being applied in various regions, with a maximum increase in effective precipitation of 5-15 percent by 2045.
Each of these types of agricultural water management interventions are associated with different investment costs. These interventions include the cost of developing new technologies and infrastructure (canals, dams, etc.), as well as efforts to increase the capacity for water management. Across all developing countries, the additional cost of irrigation expansion amounts to US$3.5 billion per year. Because of differences in investment costs and assumed expansion of irrigated area, the cost of developing additional irrigation infrastructure is largest in SSA (US$1.2 billion per year). The region contributes about one-eighth of the projected irrigation expansion in hectares, but accounts for about one-third of the additional investment costs (Table D -2). For the IX+WUE scenario, the investment cost is composed of two portions, irrigation expansion plus improvements in water use efficiency. Improving WUE is about 30 percent more expensive than solely expanding irrigation using conventional methods. This is in large part because these improvements are implemented across a larger area, as the improved average water use efficiency is for all irrigated area (412 million hectares across developing countries), not just the newly expanded area (20 million hectares). The IX+WUE scenario requires US$8.1 billion per year in additional investments across developing countries.
The cost of implementing measures to enhance soil water capacity for crop production (SWHC) is estimated by applying a cost per hectare (US$179) to both rainfed and irrigated cropland across developing regions. The cost for the SWHC scenario is US$4.6 billion per year across developing countries.
Investment Scenarios: Improved infrastructure and market access The Reduced Marketing Margins (RMM) scenario assumes a mix of infrastructure improvements throughout the economies of developing countries, focusing primarily on improvements to transportation infrastructure (road building, road maintenance, and railroads) and increased rural electrification. These improvements enhance productivity along the value chain, increase the speed of moving commodities to markets, as well as improve storage capacity, all of which improve market efficiency by better matching supply and demand over time. These improvements are represented as a reduction in the cost of moving goods from the farm to market. In IMPACT, this is done by adjusting the price wedges between producer and consumer prices, reducing the margin from producer prices to consumer prices by 1 percentage point per year between 2015 and 2030. The average annual investment from 2015 to 2050 is US$10.8 billion across all developing countries.

Investment Scenarios: Comprehensive investment portfolio
The final alternative investment scenario considers the potential outcomes of a comprehensive investment portfolio (COMP) that combines the investments of four of the scenarios previously specified, representing a broad array of investments across different parts of the agriculture system. These include one of the productivity enhancement scenarios (HIGH+RE), two of the improved water management scenarios (IX+WUE and SWHC), and the investments in infrastructure to reduce marketing inefficiencies (RMM). The total additional cost for the COMP scenario is US$25.5 billion per year from 2015 to 2050.