Climate mitigation sustains agricultural research and development expenditure returns for maize yield improvement in developing countries

Governmental expenditure on agricultural research and development (R&D) has played a substantial role in increasing crop yields in recent decades. However, studies suggest that annual yield growth rates would decline in a warming climate compared to that in a non-warming climate. Here, we present how projected climate could alter maize yield gain owing to a US$ 1 billion increase in agricultural R&D expenditure (referred to as yield response) for 71 maize-producing countries using global gridded crop model simulations with socioeconomic and climate scenarios as inputs. For the middle of this century (2041–2060) under the low warming scenario (shared socioeconomic pathways: SSP126), the median yield response between countries is estimated to be the highest at 27.2% in the low-income group, followed by 6.6% in the lower-middle-income group, 1.0% in the high-income group, and 0.1% in upper-middle-income group. The projected median yield response for lower (the low- and lower-middle)-income groups under the high warming scenario (SSP585) was approximately half than that under the low warming scenario: 27.2% → 15.6% for the low-income, 6.6% → 1.7% for the lower-middle-income, and 1.0% → 0.6% for the high-income groups. For the upper-middle-income group, where there is limited room for adopting high-yielding technology and management already being used in higher (the high- and higher-middle)-income groups, the negative impacts of climate change cannot be offset and yields are projected to decline, even with continued R&D investments (0.1% → –0.2%). Even if the R&D expenditures increase at the same value, expected yield gains will depend on future warming levels. This finding suggests that climate mitigation is a prerequisite for maintaining the yield returns from agricultural R&D investments in developing countries.


Introduction
Global caloric yield increased by 125% for the period 1979-2018 (Zhu et al 2022). Yield growth (an increase in crop production per unit area), cropping frequency, and cropping area are increased by 73%, 18%, and 11%, respectively, which underlies the substantial role of yield growth in the production increase. Rainfed and irrigated agriculture together, global agricultural production is projected to increase by 1.1% per year over the next decade, with 80% and 15% of this increase coming from yield growth and cropland expansion, respectively (OECD and FAO 2022). In particular, rapid yield growth is expected to occur in low yield regions worldwide where crops are mostly grown in rainfed condition Pardey 2014, van Ittersum et al 2016).
However, studies indicate that climate change has reduced global agricultural total factor productivity by 20% (1961-2020), especially in the low latitudes, where many developing countries are located (Ortiz-Bobea et al 2021, IPCC 2022. A recent study reported that irrigated wheat yields in Sudan need to increase by 2.7% (0.3%) per year by breeding new heat-tolerant varieties to maintain the current yield level under a 4.2 • C (1.5 • C) warming in 2050, which challenges meeting the increasing national wheat demand (Iizumi et al 2021). The global population is projected to increase with the highest growth rate at the middle of the 21st century (United Nations 2022); therefore, the negative impacts of climate change on yield growth and resulting crop production raise concerns to ensure global food security for 9.8 billion people by 2050 (Wiebe et al 2015, Fujimori et al 2019. Continuous expenditures on agricultural research and development (R&D) by national governments as well as contributions from the private sector are a major driver of yield growth (Beintema et al 2012, Lobell et al 2013, Iizumi et al 2017a, Mason-D'Croz et al 2019, Salim et al 2020, Ortiz-Bobea et al 2021. Governmental agricultural R&D expenditures have contributed to yield growth in both developing (Griffith et al 2004, Pardy et al 2006, Nin-Pratt 2021 and developed countries Islam 2010, Andersen andSong 2013). However, yield growth has plateaued in some crops worldwide (Grassini et al 2013), and the negative impacts of climate change, especially temperature increases, have been identified as the major climatic factor affecting the observed yield stagnation (Moore and Lobell 2015, Iizumi et al 2018, Ray et al 2019, Ortiz-Bobea et al 2021. Moreover, agricultural R&D expenditure returns to yield growth are likely to decline with warming (Ortiz-Bobea et al 2021).
Aforementioned studies primarily focused on the relationships between yields and agricultural R&D expenditures in the past, and their long-term outlooks under projected climate remain unclear. Although the impact of climate change on yields, food prices, and undernourished populations have been widely investigated using agricultural economic models (e.g. Wiebe et al 2015, Fujimori et al 2019, yield growth returns from investments in agricultural R&D are treated as exogenous variables in those models and given from scenarios. Therefore, this study aims to present an outlook how projected climate affects yield growth fostered by governmental agricultural R&D expenditure for the middle of this century. We focus on maize, a major staple crop that is projected to be severely affected by climate change (Jägermeyr et al 2021), and 71 maize-cropping countries worldwide with different income groups.

Data and methods
The maize yield gain owing to a $1 billion (B) increase in governmental agricultural R&D expenditure (hereafter referred to as the yield response) was estimated for different warming levels and country income groups. The data used in the estimations were agricultural R&D expenditure scenarios derived from the shared socioeconomic pathways (SSPs) and yield projections under different warming and socioeconomic scenarios using the crop model (table S1). We used the outputs of the Crop Yield Growth Model with Assumptions on climate and socioeconomics (CYGMA) (Iizumi et al 2017a(Iizumi et al , 2018(Iizumi et al , 2021. The estimated yield responses were compared with reported national-level statistics for validation.

R&D scenarios
The annual agricultural R&D expenditures spent by the national government of each country were estimated as the product of three variables: gross domestic product (GDP, constant 2005 USD yr −1 ), GDP share of agriculture (% of GDP), and the sector-averaged GDP share for R&D (% of GDP). The national statistics available in the World Development Indicators database (World Bank 2022a) were used to obtain these variables for the historical period . The country GDP scenarios were obtained from the five SSP scenarios (SSP1: sustainability, SSP2: middle of the road, SSP3: regional rivalry, SSP4: inequality, and SSP5: fossil-fueled development) for the future period (2015-2060) (O'Neill et al 2014). The GDP share of agriculture and the sector-averaged GDP share for R&D were set as the averages of 2001-2010 for the future period. This setting may be unrealistic for some countries as the GDP share of agriculture generally decreases with economic development, which would likely overestimate agricultural R&D expenditure in the future period. Future policy could change agricultural R&D expenditure and therefore is another source of uncertainty. However, their future scenarios are not yet available. Furthermore, the agricultural R&D expenditure used here was not specific to maize but to agriculture. Crop-specific R&D expenditures are generally impossible to obtain, with a few exceptions, such as breeding projects and their benefit-cost analysis (Nalley et al 2018).
The agricultural R&D expenditure assumed in this study was the highest under SSP5 scenario and the lowest under SSP3 scenario (figure 1). For example, the anticipated agricultural R&D expenditure in 2060s (2051-2060) under SSP5 scenario is 3.7 times higher than that under SSP3 scenario in the low-income group. The relative increase in agricultural R&D expenditure in lower (the low-and lower-middle)-income group was scenario dependent but always larger than that in higher (the highand higher-middle)-income group due to low level of current R&D expenditure in the lower-income group in absolute term.

Yield projections
The yield projections were derived from the CYGMA crop model simulations (table S2). The simulated period is 1981-2005 and 2015-2060 for the historical and future period, respectively. Five socioeconomic scenarios (SSP1-SSP5) and two warming levels (low and high) were considered. For the sixth phase of the Coupled Model Intercomparison Project (CMIP6) atmosphere-ocean coupled general circulation models (GCMs), the low-warming scenarios were based on SSP126, while the high-warming scenarios relied on SSP585 (O'Neill et al 2016). In addition, we used the Representative Concentration Pathways (RCPs) 2.6 and RCP8.5 (van Vuuren et al 2011) for the low and high warming scenarios, respectively, in CMIP5 GCMs to test the robustness of our findings to different generations of GCMs (see supplementary for the results of CMIP5 GCMs).
The climate inputs to the CYGMA model were a bias-corrected global reanalysis called S14FD (Iizumi et al 2017b) for the historical period. For the future period, bias-corrected daily outputs of two CMIP6 GCMs (MPI-ESM1-2-HR and MRI-ESM2-0; Lange 2019) were used. The two GCMs were selected as a result of avoiding very high equilibrium climate sensitivity (Scafetta 2022). Additionally, five CMIP5 GCMs (GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC-ESM-CHEM, and NorESM1-M; Iizumi et al 2017b) were used (supplementary note). The different bias-correction method and reference meteorological forcing dataset, as described in Lange (2019), were used for the CMIP6 and CMIP5 GCMs used here. The atmospheric CO 2 concentrations were set to be consistent with the observations for the historical period (379 ppm in 2005) and with the values used in CMIP6 and CMIP5 for the future period (the CO 2 value was 474 ppm, 643 ppm, 442 ppm, and 604 ppm for SSP126, SSP585, RCP2.6, and RCP8.5, respectively, in 2060).
Governmental agricultural R&D expenditures were converted into agricultural knowledge stock by summing R&D expenditures since 1961 with a certain obsolescence rate, as described by Iizumi et al (2017a) (table S1). To set scenario on nitrogen application rate, per capita GDP and per capita agricultural area were considered. The per capita GDP changed over time based on the reported data for the historical period or the SSP scenarios for the future period. The per capita agricultural area changed with time based on the Food and Agriculture Organization of the United Nations (FAO) statistics for the historical period, whereas it was fixed at 2010 level for the future period since future scenario on crop-specific harvest area is not available (table S2). These simulation settings were the same as those used by Iizumi et al (2017a).
Furthermore, two adaptation practices, sowing date shifts and cultivar switching, were considered in the crop model. Sowing dates change gradually depending on the changes in thermal and moisture regimes. Cultivar switching assumes using longer-season cultivars to avoid yield reductions associated with shortened crop duration in warmer conditions, a widely used adaptation simulation setting in global gridded crop models (e.g. Deryng et al 2014). These adaptation practices neither incur additional production costs for farmers nor affect governmental agricultural R&D expenditures. The cultivar switching used here does not indicate the adoption of newly bred heat-tolerant cultivars, and rather indicates adopting existing cultivars already grown in warmer regions of the world.
In the middle of this century, the projected yields in the high-and upper-middle-income groups under the low warming scenario were the same as current levels for SSP2 and SSP4 scenarios, while they were 25% higher for SSP1, SSP3, and SSP5 scenarios (figure 1). In the lower-middle-income groups, the projected yield increases were found to be 50% for SSP1 and SSP5 scenarios, whereas those for SSP2, SSP3, SSP5 scenarios were on the similar levels to the current yields. Higher yields than current levels were projected for all SSP scenarios in the low-income group.
In the high-warming scenario, the increase in yield was projected to be suppressed, especially in the low-income and lower-middle-income groups. For the middle of this century and under the SSP5 scenario, the projected yield increase was 211.7% under the low-warming scenario and 164.2% under the high-warming scenario in the low-income group (figure 1), with a difference of 47.5% point. The large reduction of yield increase from the low-warming scenario to the high-warming scenario was also found in the lower-middle-income (149.5% → 126.4%; the difference of 23.1% point) and upper-middleincome (125.4% → 113.2%; 12.2% point) groups except for the small change in the high-income group (118.7% → 121.6%; 2.9% point). The reduction of yield increase under the high-warming scenario was common across the five SSP scenarios. These tendencies found in the CMIP6-based yield projections were similar when the CMIP5-based yield projections were analyzed (figure S1). Note that the yield projection to climate change largely dependent on individual crop models, and the CYGMA model tend to derive more pessimistic maize yield projections than many other crop models (Jägermeyr et al 2021).

Yield response
The yield response was computed for each country using the following procedure. For a given warming scenario (SSP126, SSP585, RCP2.6, and RCP8.5), annual yields for the period 2041-2060 (% relative to the GCM-based average yield of [1996][1997][1998][1999][2000][2001][2002][2003][2004][2005] were linearly regressed against annual agricultural R&D expenditures. The slope of the regression line indicated yield response. The sample size for a given country was 200 (=20 yr × 5 SSPs × 2 GCMs) in the CMIP6 projection for the warming scenario (for CMIP5, 500 = 20 yr × 5 SSPs × 5 GCMs). Figure 2 exhibits the calculated maize yield response for Colombia, where the yield gain due to a $1 B increase in agricultural R&D expenditure was found to be 24.4% (i.e. 0.61 t ha −1 ). The average (or median) value for a given income group was derived by using the calculated country yield response values. Four income groups were considered according to the World Bank classification (World Bank 2022b): high: gross national income (GNI) in 2020 > $12 535, upper-middle: $4046-$12 535, lower-middle: $1036-$4045, and low: <$1036 (figure S2). As mentioned above, the agricultural R&D expenditure used here was not specific to maize, and therefore we assumed that R&D expenditure for maize yield improvement increased proportionally with the total agricultural R&D expenditure.

Validation
To evaluate the estimated average yield response value per income group for the historical period, we calculated the yield response values for 1981-2005 using the reported yields and agricultural R&D expenditures from the FAO (FAO 2022) and International Food Policy Research Institute (IFPRI) databases (IFPRI 2022), respectively. This calculation was performed for 36 countries, whose average annual maize production exceeds 0.1 Mt during 1996-2005, and the reported agricultural R&D expenditure data were available. Although future yield response projections were made for 71 maize-producing countries, the report-based yield response values were only available for approximately half of them because of the lack of reported agricultural R&D expenditure data ( figure S2). The future R&D expenditure was estimated using future GDP projections available in SSP scenarios, which increased the number of analyzed countries for the future period. Currently, the R&D expenditures in developing countries is less than $1 B. In fact, $1 B corresponded to only 16.6% and 15.8% of the R&D expenditure in 2005 for highincome and upper-middle-income (including China) group, respectively, whereas $1 B did 156.6% and  Comparisons of the report-and model-based maize yield response to 1% R&D increase for the historical period . GLB denotes all income groups analyzed here (the number of countries is 36). HI, UM, LM, and LO represents the high-(13), upper-middle-(7), lower-middle-(13), and low-income groups (3), respectively. See figure S2 for the geographic distribution of the countries studied. The lines indicate the mean and 95% confidence intervals of average yield response per income group estimated using the bootstrap technique. 552.4% for lower-middle-income and low-income groups, respectively. For this large difference in R&D expenditure in absolute terms, we compared yield response per a 1% increase of R&D in 2005. The estimated and report-based country yield response values were bootstrapped 2000 times to calculate the mean and 95% confidence interval of the average response value for each income group.

Validation
The estimated yield responses to 1% R&D increase were consistent with the report-based responses, irrespective of the income groups (figure 3). The estimated average yield response value for each income group ranged from 0.14% (95% confidence  interval spanning from −0.37% to 0.62%) for the lower-middle-income group to 0.48% (0.19%-0.79%) for the upper-middle-income group. These estimated values were comparable to the reportbased values of −0.01% (-0.12%-0.10%) for the low-income group and 0.28% (0.02%-0.54%) for the lower-middle-income group. The yield responses were overestimated for some countries, such as Indonesia and Guatemala in the upper-middleincome group and Mali in the low-income group, where the reported trends in agricultural R&D expenditure were vastly different from those calculated from GDP, GDP share of agriculture, and sector-averaged GDP share for R&D. However, our estimates reproduced the historical average yield response per income group well.

Future changes
The projected yield response was larger in the low-income group than in the high-income group ( figure 4). Under the low-warming scenario (SSP126), the median value was the highest for the low-income group (27.2%), followed by the lower-middle (6.6%), high-income (1.0%), and upper-middle (0.1%) groups. The projected median yield response under the high-warming scenario (SSP585) was approximately half than that under the low-warming scenario for the lowincome (27.2% → 15.6%), lower-middle-income (6.6% → 1.7%), and high-income (1.0% → 0.6%) groups. For the upper-middle income group, the yield response under the high-warming scenario was projected to decrease (0.1% → −0.2%). The tendency of greater reduction of yield response under the high-warming scenario was the same as even when we focus on the top producers in each income group (

Discussion and conclusion
The future maize yield response in the high-and upper-middle-income groups is expected to be much smaller than that in the lower-income groups, irrespective of the warming levels ( figure 4). Increasing yields in high yield regions is difficult; as Fuglie (2018) reported, the yield response will be weaker in developed countries. Therefore, novel technologies and management to further increase yields in high yield regions were not considered in the scenarios used in this study, thus explaining the small yield response in the high-income group.
However, yield returns fostered by agricultural R&D expenditure will likely decrease under high warming scenarios compared to low warming scenarios. This suggests that even the same value of agricultural R&D expenditure increase would cause different percentage increase in yield, depending on the progress of climate mitigation, which will be more pronounced in the low-and lower-middle-income groups. In these lower-income groups, current yields are lower than in the higher-income groups; therefore, the expected future yield increase is larger than that in the higher-income groups. The current highincome countries were the center of agricultural production 50 years ago; however, their global relative importance has decreased (Alston and Pardey 2014), and the current global growth in agricultural productivity has been sustained by developing countries (Fuglie 2018). Therefore, maintaining yield growth in developing countries under warming will be important for global food production, indicating that efforts of emission reduction to achieve the low warning scenario are likely unavoidable not only for the higher-income groups but even also for the lowerincome groups with low adaptive capacity and high vulnerability.
Agricultural R&D expenditures of national governments as well as the private sector are major drivers of yield growth. However, returns from R&D investments under climate change depend on climate mitigation efforts and associated warming levels. This is most pronounced in the low-income countries, and the projected yield growth under high-warming scenarios will decline to approximately half than the anticipated yield growth under the low-warming scenarios. Therefore, climate mitigation is particularly beneficial for the low-and lower-middle-income groups, where high-yielding technology and management, already being used in high-income group, can be adapted. In this study, yield growth driven by agricultural R&D expenditure in high-income group was insensitive to warming levels. However, this does not indicate that yields in high-income group are unaffected by climate change, which is evident by looking at yields rather than yield responses (figure 1). This counterintuitive outcome for the high-income group is likely due to the limitation of the technology and management scenarios used. Future research is needed to more plausibly depict the yield response in high-income group under climate change. Moreover, the effect of governmental investment on agriculture would vary from crop to crop, but we used a simple assumption that R&D expenditure for maize is proportional to total agricultural R&D expenditure, as the first step. The information on crop-specified R&D expenditures is required to evaluate the effect of climate mitigation on yield responses in detail.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).