Global modeling of SDG indicators related to small-scale farmers: testing in a changing climate

Some indicators used to track the progress of the Sustainable Development Goals (SDGs) suffer from a lack of reported data, and therefore need estimates to fill the data gaps. Using crop model outputs and global cropping system datasets, we present a modeling of small-scale farmer productivity and agricultural output (conceptually similar to the formal SDG 2.3.1 and 2.3.2 indicator, respectively). We analyze the responses of the indicators for 106 low- and middle-income countries for the periods 2051–2060 and 2091–2100, relative to 2001–2010, to various scenarios of climate, socioeconomic development, cost-free adaptation, and irrigation expansion. The results show the potentials of modeling in gap-filling of reported national data, and that the agricultural output indicator indicates the positive effect of climate mitigation to small-scale farmers. The contributions of adaptation are evident when agricultural output indicator is used but are no longer visible, or even wrongly interpreted, when productivity indicator is used, underling the importance of selecting robust indicators to track SDG goals in a changing climate. Also discussed are the caveats identified in the SDG 2.3 indicators that enable the design of indicators more aligned with the other development goals, such as poverty eradication.


Introduction
Indicators used to track the progress of the UN's 17 Sustainable Development Goals (SDGs) often suffer from a lack of reported national data (Annan 2018). SDG target 2.3, which calls for doubling the agricultural productivity and incomes of small-scale food producers by 2030, is no exception (FAO 2021). In this case, the productivity and income of small-scale farmers have been designated for use in tracking progress. However, these indicators are classified as Tier II (as of 4 February 2022), meaning that the 'indicator is conceptually clear, K but data are not regularly produced by countries' (UN 2022). Therefore, it is worth examining potentials of modeling to fill the data gaps of SDG indicators as a supplementary tool.
Small-scale farmers are the main target in Goal 2. This is reasonable, given the facts that small-scale farmers account for 84% of farmers worldwide (Lowder et al 2021); 77% of the small-scale farmers operating in low-and lower-middle-income countries live in water-scarce regions (Ricciardi et al 2020); even in developing countries, many small-scale farmers are poor, and therefore agricultural development is an essential engine of poverty eradication as targeted in Goal 1 (Morton 2007, Carletto et al 2015; and these challenges are subject to the negative impacts of climate change depending on the mitigation and adaptation efforts employed since smallscale farmers are one of vulnerable groups and often at higher risk of malnutrition, livelihood loss, rising costs and competition over resources (Bezner Kerr et al 2022).
However, few studies shed light on modeling of small-scale farmers at the global scale, although many have presented modeling of hunger risk, another target in Goal 2 (Hasegawa et al 2018, Fujimori et al 2019, Fuso Nerini et al 2019, Soergel et al 2021. Household surveys have played a role as a source of local data, but reliable data on small-scale farmers across the globe are still lacking despite efforts to identify their number, production share, and geographic distribution in global agricultural systems (Samberg et al 2016, Ricciardi et al 2018, Lowder et al 2021, Su et al 2022. This makes global modeling of SDG indicators related to small-scale farmers challenging.
In this study, we show the global modeling of small-scale farmer productivity and agricultural output. The productivity and agricultural output analyzed here are conceptually similar to the formal SDG 2.3 indicators. Then we examine response of the indicators to various scenarios of climate, socioeconomic development, costfree adaptation, and irrigation expansion to test if the modeled indicators are appropriate for use even in a changing climate. Such a modeling exercise is expected to contribute to improved indicator design in the real world where climate change has occur.

SDG indicators related to small-scale farmers
We analyze two SDG indicators: small-scale farmer productivity and agricultural output (conceptually similar to the formal SDG indicator 2.3.1 and 2.3.2, respectively; Supplementary text S1). Crop production costs are not considered in the agricultural output used here, which is the difference with the formal SDG indicator 2.3.2. The agricultural output is annual crop production in U.S. dollars (USD) per household. The productivity is derived by dividing the agricultural output by annual number of agricultural labor days.
The data used to estimate the average agricultural output for the period 2001-2010 are summarized in table S1. In short, we combined yield and harvested area maps, farm field size category map, producer prices, and crop calendars, all of which were obtained from public databases, as well as per household acreage obtained from the literature. Furthermore, we used crop model outputs which projected yields and crop durations from sowing to harvesting for the near and distant future (averages of 2051-2060 and 2091-2100, respectively). The subsequent subsections provide additional details. Figure 1 shows the data flow for calculating the current small-scale farmer agricultural output at the country level. The procedure starts with extracting the grid area harvested and yields for 2000 [10-km resolution; Monfreda et al (2008)] by overlaying the farm field size category map for 2005 ( figure 1, figure S1). For this, we re-gridded the global 1-km farm field size category map (Fritz et al 2015) to 10-km resolution by taking the 1-km mode category within a 10-km grid cell. The extracted grid area harvested and yield were multiplied to obtain the production volume for each grid cell, of which the typical farm field size category was either 'very small (0.5 ha)' or 'small (0.5-2 ha).' We assumed that these two categories represent the small-scale farmer-dominated landscape. The calculated grid production volume (tonnes) was then summed to the country level and multiplied by the FAO country producer price (the average of 2011-2015) to derive the national annual value of the small-scale farmer production of a crop (USD). These calculations were repeated for the six crops considered (maize, rice, wheat, soybeans, millet, and sorghum), which together accounted for 50% (785 Mha) of the global cropland area in 2020 (FAO, 2022). Cereals account for a large portion (>85%) of the caloric production by small-scale farmers with farm size of <10 ha (Ricciardi et al 2018). We therefore assume that the six major cereals represent a predominant portion of agricultural output generated by small-scale farmers. The national total harvested area (hectares) of the six crops raised by small-scale farmers was calculated and then divided by per household acreage for 2013 as collected from the literature (Ricciardi et al 2018) to determine the number of small-scale farm households operating within a country, resulting in the annual agricultural output per household (USD per household) featured in the study.

Current agricultural output
The crop duration from sowing to harvesting (days) was determined for each crop using the global crop calendars for 2000 [Portmann et al (2010); the middle day of the sowing and harvest months was used] (figure 1). The days in which any of the six crops fell into the crop duration were considered agricultural labor days. Although agricultural labor days are also required outside the crop duration for tillage, seed preparation, postharvest processing, selling, and so on, we assumed that the crop duration was a good representation of agricultural labor days for cereal-producing small-scale farmers. Although this assumption may be unrealistic for some regions, data on on-farm and off-farm labor days are not available at the global scale. Using the multicrop-combined crop duration, the annual agricultural output per household described above was converted into productivity (USD per household per agricultural labor day).
The reference year of the variables used here varied by dataset, and some of the values may be outdated, especially for developing countries where rapid agricultural transitions have occurred. However, these were the best available datasets at the moment of writing. To reduce any errors associated with the use of different reference years, a 10-year period (2001-2010) was used as the base period. Table 1 summarizes the scenarios used to derive future projections of yields and crop durations from crop model. Climate mitigation, socioeconomic development, cost-free adaptation, and irrigation expansion were translated into crop yields and then agricultural output, or into crop durations and then agricultural labor days and productivity. The effects of climate mitigation, socioeconomic development, and cost-free adaptation on yields were considered within the crop model simulations described later, whereas the effects of irrigation expansion were considered when aggregating the simulated grid-cell yields under rainfed and irrigated conditions to establish the country average yields of the small-scale farmer-dominated landscape. Projected food price changes were considered when calculating agricultural output and productivity. The harvested area, farm field size category map, and per household acreage were held constant at their current state.   a The climate scenarios and corresponding atmospheric CO 2 concentrations are inputs to the crop model. The climate input includes daily mean, maximum, and minimum air temperatures, precipitation, downward shortwave radiation flux, relative humidity, and wind speed. b N application rates and agricultural knowledge stock, which are also inputs to the crop model, increase as per capita GDP increases and are referred to as SSP-based agronomic technology and management scenarios. An increase in knowledge stock increases the crop tolerance to stresses and represents higher adoption rates of high-yielding cultivars, improved input-use efficiency, and corresponding management. c Sowing date shifts and cultivar switching are considered within the crop model. d The irrigation expansion scenarios are considered when crop model outputs for rainfed and irrigated conditions are spatially aggregated to the country level.  1, table S2). These RCPs were also used to consider the CO 2 fertilization effect on yields in the crop model (451 and 1,074 ppm for the distant future under SSP126 and SSP585, respectively).

Socioeconomic development
We used SSP-based scenarios on N application rates (figure S2) and the agricultural knowledge stock (figure S3) derived using the method described in Iizumi et al (2017). The knowledge stock is an economic indicator calculated as the sum of governmental annual research and development expenditures for agriculture since 1961, with a certain obsolescence rate; it represents the average technology and management level among farmers in a particular country. A greater agricultural knowledge stock value leads to higher crop tolerance to stresses (heat, cold, water deficit, and water excess) and better input-use efficiency (moderating the N deficit), which represent the high-yielding cultivars and related management practices currently in use in high-income countries, by adjusting the form of the stress functions for the knowledge stock level ( figure S4). To derive these SSP-based agronomic scenarios, country gross domestic production (GDP) and population data were taken from O'Neill et al (2014): SSP1, sustainability, medium-high economic growth and low population number; SSP3, fragmentation, slow economic growth and very high population number; and SSP5, conventional development, high economic growth and low population number (table 1, figure S5).

Cost-free adaptation
Sowing date shifts and cultivar switching do not require additional production costs (Iizumi et al 2020) and are therefore considered to be a cost-free adaptation. The sowing date shifts depend on changes in thermal and moisture regimes, whereas cultivar switching depends solely on thermal regime changes. The crop thermal requirement from sowing to harvest, which represents the length of the crop duration, is parameterized as a function of the long-term average annual growing-degree days (GDD) (Deryng et al 2011). Using this method, longer-season cultivars, which help prevent yield reduction due to the shortened crop duration, are gradually adopted in the model along with the warming. In the scenario without cost-free adaptation, these functionalities were deactivated in the crop model, and therefore the sowing dates and crop total thermal requirements were kept at their year 2000 levels (table 1). In total, the crop model simulations were performed for 60 different scenarios, consisting of 2 RCPs, 5 GCMs, 3 SSPs, and 2 adaptation settings. In CMIP6, the radiative forcing levels and corresponding climate projections are associated with a specific SSP, such as SSP585 and SSP126 (O'Neill et al 2016). Some combinations, such as SSP3-RCP2.6, are unlikely (O'Neill et al 2020). However, we considered such combinations to examine, for instance, how the small-scale farmer agricultural output differed among the socioeconomic assumptions with the same level of warming. Changes in the 0.5°grid annual yields and crop durations of the six crops for the period 2001-2100 simulated by the crop model, relative to the simulated averages of 2001-2010, were calculated. The simulated changes were then averaged over the period 2091-2100 (2051-2060) and combined with the average yields multiplicatively and crop calendars additively for 2000 to estimate the agricultural output in the distant (near) future.

Irrigation expansion
When converting the simulated yields and crop durations under rainfed and irrigated conditions into the agricultural output, we considered two levels of irrigated area (the current level and the current level +5%) (table 1). This was done to shed light on the contribution of irrigation expansion since crop production in waterscarce regions is economically, socially, and infrastructurally vulnerable (Carrão et al 2016), and four-fifths of the small-scale farmers operating in lower-income countries live in water-scarce regions (Ricciardi et al 2020). The extent of irrigated and rainfed areas for 2000 (Portmann et al 2010) was used as the current level. The current irrigated areas were then increased by +5% based on the reported difference in the current irrigation adoption rate between small-and non-small-scale farmers in the water-scarce region (37% and 42%, respectively; Ricciardi et al 2020). Although climate change will likely affect irrigation water availability and the possible expansion of irrigated areas in the future, the lack of irrigation in Africa is mostly due to limited institutional and economic capacity rather than hydrologic constraints (Rosa et al 2020a), and irrigated cropland expansions of a magnitude of several to ten percent are likely to be feasible for many regions of the world in +3°C warming, with the exception of Eastern Europe and Central Asia, the Middle East and North Africa, and South Asia (Rosa et al 2020b). Therefore, the irrigated area expansion of +5% used here is a conservative assumption for most regions even under climate change.

Validation
We compared the estimated current productivity of a small-scale farmer with multi-year averages of the reported data obtained from UN (2021). The UN database included 20 samples from 11 countries for the period 2005 to 2016. The maximum sample number was 5 countries per year and 4 years per country (as of 16 February 2021). The data from high-income countries (Canada) were excluded from the analysis, leaving 10 countries (Burkina Faso, Ethiopia, India, Malawi, Mali, Niger, Nigeria, Panama, Tanzania, and Uganda) to be considered. These countries are in either the low-or middle-income groups. We estimated current productivity of smallscale farmer for each source of the reported country-specific per household acreage, Lowder et al (2016) and Samberg et al (2016), and compared with the reported data.

Relative contributions of individual factors to agricultural output
To quantify the relative contribution of each factor to small-scale farmer agricultural output, we calculated the difference between pairs of scenarios. For instance, we compared the agricultural output estimated under RCP8.5 and RCP2.6 with the same GCM, socioeconomic development, cost-free adaptation, and irrigation scenarios when assessing the contribution of climate mitigation. The difference was calculated for each GCM to show the uncertainty in climate projections among the five GCMs. In assessing the contribution of socioeconomic development, we compared rapid development scenarios (SSP1 and SSP5) and slow development (SSP3), while the contribution of cost-free adaptation was evaluated using comparisons with and without cultivar switching or sowing date shifts. For consistent comparisons, we used some unlikely SSP-RCP combinations (O'Neill et al 2020), such as SSP1-RCP8.5 for the calculation of climate mitigation contributions and SSP3-RCP2.6 for the calculation of socioeconomic development contributions. Without the combinations, both the climate change impact and socioeconomic condition change simultaneously, and it makes impossible to address the contributions of the individual factors. For the calculation of the contributions of climate mitigation, socioeconomic development, and cost-free adaptation, no distinction was made between waterscarce regions and non-water-scarce regions. Unlike these factors, the contribution of irrigation expansion compared the +5% scenario and current level for two cases, one for the water-scarce region and non-waterscarce region combined and the other for the water-scarce region only (see figure S6 for the water-scarcity classification map).

Current productivity of small-scale farmers
The estimates of small-scale farmer productivity were found to be comparable to the reported data for most of the countries examined here, with an exceptional error of 24.4 $ hh -1 d -1 for Panama and of -11.9 $ hh -1 d -1 for Nigeria (figure 2). Current productivity across the 106 low-or middle-income countries was estimated to be 0.1-68.9 USD per household per agricultural labor day ($ hh -1 d -1 ) (figure S7), which was reasonably wider than the reported ranges of 3-13.5 $ hh -1 d -1 among the 9 countries (FAO 2021). The literature reports that farms in the <20 ha farm size classes grow more cereals as well as roots and tubers than fruits, oil crops, pulses, tree nuts, vegetables, and other crops (Herrero et al 2017, Ricciardi et al 2018. This difference in main crop type between farm size classes is in line with additional evidence that cereals constitute 30-70% of the total dietary calories in developing countries (Burke and Lobell 2010). These facts explain why small-scale farmer productivity could be plausibly estimated based on the crop model outputs for only six crops.

Future projections
The projected changes in small-scale farmer productivity and agricultural output for the near (2051-2060) and distant (2091-2100) futures relative to the current levels (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010) were characterized by a large spread associated with the use of different scenarios. When only likely SSP-RCP combinations were used, productivity increases, on average, to 15.6 $ hh -1 d -1 , with a minimum-maximum range among the 20 scenarios of 13.5-16.6 $ hh -1 d -1 in the near future along with socioeconomic development, from the current levels of 13.3 (11.7-14.4) $ hh -1 d -1 (figure 3(a)). Even though agronomic technology and management continues to improve in the scenarios ( figure S2, figure S3), in the distant future, productivity showed two distinctive patterns, with a range of 9.7-19.6 $ hh -1 d -1 (an average of 15.9 $ hh -1 d -1 ): a 'continuous increase' for the mitigated climate (RCP2.6) and a 'turn to a decrease' for the non-mitigated climate (RCP8.5). Current productivity differed depending on the GCM and the presence or absence of cost-free adaptation, indicating that impacts on productivity from recent warming and adaptation efforts have already occurred.
The average annual number of agricultural labor days was projected to decrease from the current level of 343 (346-340) days to 338 (344-327) days in the near future and to 322 (332-302) days in the distant future ( figure 3(b)). The decrease in agricultural labor days was smaller in the mitigated climate or with adaptation and larger in the non-mitigated climate or without adaptation. In the crop model, temperature increases led to rapid growth and shortened crop duration, and cost-free adaptation (cultivar switching and sowing date shifts) moderated the reduction in crop duration.
The average annual agricultural output changes from the current level of 4,640 (4,055-4,887) $ hh -1 to 5,235 (4,383-5,673) $ hh -1 in the near future and to 5,163 (2,953-6,391) $ hh -1 in the distant future ( figure 3(c)). When the paired scenarios were compared so that the adaptation setting was the only difference, a contrasting tendency was found between productivity and agricultural output. Productivity without adaptation was slightly higher than with adaptation ( figure 3(a)), which was the opposite of agricultural output (figure 3(c)). Note that the shortened agricultural labor days rather than the increase in agricultural output were the reason that productivity without adaptation was apparently higher than productivity with adaptation. This outcome suggests that there is a risk of misinterpretation, i.e., incorrectly concluding that small-scale farmer productivity would not benefit from adaptation when analyzing the productivity indicator. Consequently, we used agricultural output in our subsequent analysis.

Socioeconomic development contributions
The results revealed that, when comparing the paired scenarios (including ones recognized to unlikely occur), on average, climate mitigation had the greatest impact on agricultural output, followed by socioeconomic development, cost-free adaptation, and irrigation expansion ( figure 4, figure S9). Agricultural output was found to be higher under rapid development (SSP1 and SSP5) than under slow development (SSP3) (figure 4, figure  S9). This same tendency was seen for yields as well (figure S10). Among the income groups, the low-income countries would benefit the most from socioeconomic development ( figure 4). In low-income countries where crop yields are currently low, there is more room than in middle-income countries for adopting the highyielding technology and management currently used in high-income countries.

Mitigation contributions
Climate mitigation from RCP8.5 to RCP2.6 would lead to a substantial avoidance of a reduction in agricultural output in the long run (figure 4). Agricultural output for the distant future under RCP2.6 would increase by 51.0-62.4% relative to the current levels due to the improved agronomic technology and management in the developing countries (this commonly occurred in SSP1, SSP3, and SSP5 to differing degrees) and avoidance of the negative impacts of climate change. These values were, on average, 52.7% higher than under RCP8.5 ( figures 5(a), (b)), indicating an overall benefit from climate mitigation to the agricultural output. For the distant future, the mitigation contributions were greater in the lower-middle-income countries (an average difference of +69.1% under RCP2.6 relative to RCP8.5) and less in the upper-middle-income countries (+39.5%). The low-income countries fell roughly in the middle of the two income groups (+52.2%). Importantly, the mitigation contributions for the near future were still substantial ( figure 5(b), figure S9).

Adaptation contributions
Overall, the impacts of cost-free adaptation on agricultural output were found to be positive but small compared to the contributions from socioeconomic development and climate mitigation described above ( figure 4). Interestingly, the positive impacts were more prominent in a high warming scenario (RCP8.5) than in the lower warming scenario (RCP2.6), with a mean difference between the with and without adaptation cases of +6.4% for RCP8.5 (figures 6(a), (b), (d)) and +2.4% for RCP2.6 (figures 6(a)-(c)). The adaptation contribution in the low-income countries was not the highest among the different groups considered here. This tendency was common between the mitigation levels (figures 6(c), (d)). Such a counterintuitive outcome can be explained as follows. Replacing current cultivars with those grown in warmer regions occurred in the crop model. This modeled cultivar switching was only marginal in the low-income countries since they are often located in the low latitudes where seasonal temperatures are already high (figure S11). Novel cultivars beyond those currently grown in warmer regions were not considered. Although a sowing date shift was also considered, it did not greatly help increase agricultural output in the low-income countries when the cultivars remained largely unchanged. Figure 4. Contributions to agricultural output by factor and income group. The factor-by-factor contributions to agricultural output for the distant future are distinguished by comparing the paired scenarios: mitigated climate (RCP2.6) relative to non-mitigated climate (RCP8.5); rapid development (SSP1 and SSP5) relative to slow development (SSP3); adaptation relative to no adaptation; and +5% irrigation relative to current irrigation (the water-scarce and non-water-scarce regions together). Boxplots indicate the median (horizontal line in a box), average (the cross), 25-75th percentile (box), and minimum-maximum range (whiskers) between the 5 GCMs (the results are averaged for each GCM). Figure 5. Mitigation contributions. Projected changes in agricultural output for the near and distant futures relative to the current levels under RCP2.6 and RCP8.5 (a) and the differences in agricultural output between RCP2.6 and RCP8.5 by income group (b). The notation for the boxplot is the same as in figure 4, but the results for the 3 SSPs and 2 adaptation settings are averaged for each GCM.

Irrigation expansion
The contributions of irrigation expansion to agricultural output were found to be the smallest among the factors considered here for the combined water-scarce and non-water-scarce regions (figure 4). However, these contributions turned out to be larger than the contribution from cost-free adaptation when the focus is on water-scarce low-income countries. If small-scale farmers could increase their irrigation use to the current nonsmall-scale farmer level, the agricultural output in the water-scarce low-income countries would be 10-11% higher than what would occur under the current irrigation level ( figure 7(b)). The projected gains in agricultural output in the water-scarce low-income countries associated with irrigation expansion were substantial compared to those in the water-scarce middle-income countries (3-5%). Nevertheless, the shift to a decreasing pattern prevailed under RCP8.5. This result suggests that the returns to agricultural output from investments in irrigation expansion for small-scale farmers in the water-scarce region would disappear if accompanied by a non-mitigated climate, while the returns would be maintained if a mitigated climate could be achieved.

Discussion
Our results indicate that climate mitigation and adaptation have positive effects on SDG 2.3 indicators in the long run beyond the current time frame set to 2030. Returns to agricultural output from socioeconomic development and associated improvements in agronomic technology and management would be maintained under a mitigated climate (RCP2.6, which corresponds to the 2°C target of the Paris Agreement) but that progress towards improved small-scale farmer livelihoods would stagnate or even retreat under a non-mitigated climate (RCP8.5). Cost-free adaptation would work to some degree. However, its contributions are rather dependent on warming levels and less effective in low-income countries where seasonal temperatures are already high. High-cost technology, namely, irrigation expansion, is likely unavoidable. In water-scarce lowincome countries, even closing the gaps in irrigation adoption rates that exist between small-and non-smallscale farmers would be substantially beneficial. Therefore, irrigation expansion is expected to be a firmer option Figure 6. Adaptation contributions. Projected changes in agricultural output for the near and distant futures relative to the current levels by adaptation setting under RCP2.6 (a) and RCP8.5 (b), and the differences between with and without adaptation by income group for the distant future under RCP2.6 (c) and RCP8.5 (d).
for climate resilient development in water-scarce low-income countries in terms of agricultural output, as long as water availability and the cost-benefit ratio allow adoption of this option.
By ameliorating the data gaps that prevalent in reported national data, the estimates of small-scale farmer agricultural output presented in this study enable one exploring the relationships between climate action and development goals related to small-scale farmers in the future research. However, the modeling needs to be improved to provide more accurate estimates. The estimates are less reliable when the farm size map is likely not representative due to the small extent of national cropland area (ex. Panama; figure S12). For interested readers, opportunities for model improvement, for instance, the availability of production cost and labor days data, use of other crop model outputs, and the consideration of other crops, livestock and fish products, are discussed in Supplementary text S3. In addition, information on the quality of the reported data would help identify sources of error in the estimates, since the small-scale farmer productivity reported for Nigeria is suspected to be too high (it is 3-4 times higher than the productivity in other African countries considered here).
Our results also identify the requirements that will be important in designing indicators that will properly work in a changing climate and that will enhance our understanding of the long-term associations involving climate action and the goals related to small-scale farmer livelihoods. The importance of selecting appropriate indicators is underlined. While small in size, the contributions of adaptation are evident when agricultural output is used as an indicator but are no longer visible, or even wrongly interpreted, when the productivity indicator is used.
In addition, a more strictly defined unit is desired. Apart from the ambiguity and complexity in defining small-scale farmers across countries (Samberg et al 2016, FAO 2018, the modeling exercise presented here revealed that the current unit of the productivity indicator (both ours and the formal SDG indicator 2.3.1) can be either the value of agricultural production per person or per household. The fact that the same reported value of this indicator should be interpreted differently if the number of household members, the per household acreage, or the annual number of agricultural labor days is different hinders consistent comparisons between countries and between regions within a country. The same is true for the agricultural output indicator or the formal SDG indicator 2.3.2. With the current definition, it is impossible to determine whether the reported agricultural output (or more precisely income) of a small-scale farmer is higher or lower than the official international extreme poverty line of $1.90 per person per day in 2011 purchasing power parity (World Bank 2018). Clarifying the units of these indicators enables an alignment of the multiple indicators used for different goals (e.g., Goal 2 and Goal 1) so that they become consistent as a whole.
(JPMEERF20202005) of the Environmental Restoration and Conservation Agency Provided by the Ministry of Environment of Japan. T I was partly supported by Grants-in-Aid for Scientific Research (22H00577 and 20K06267) from the Japan Society for the Promotion of Science.

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

Additional information
Supplementary information accompanies this manuscript.