Penny wise, pound foolish: Substitution cost of cropland lost to urbanization


 Urbanization has appropriated millions of hectares of cropland1, and this trend will persist as cities continue to expand2. Here we estimated the substitution cost by comparing the yield potential between the converted and newly cultivated land as determined by climate and soil properties. To do so, we used robust spatial upscaling techniques, well-validated crop simulation models, and soil, climate, and cropping system databases 3–5, focusing on populous countries exhibiting high rates of land conversion. We found that productivity of new cropland is substantially lower than the land it replaces, which means that projection of food production potential must account for expected cropland loss to urbanization and the lower productivity of new land that replaces it. Policies that protect existing farmland from urbanization would relieve pressure on expansion of agriculture into natural ecosystems and reduce the associated greenhouse gas emissions and biodiversity loss.

versus yields on cropland distant from cities 2 , which often confounds the substitution cost with regional differences in sophistication of crop and soil management practices employed by farmers and their access to inputs and supporting services. Other estimates of substitution costs are generic and do not differentiate among major food crop species 1,13 . Estimating substitution costs speci c for each major food crop is important because diets in developing countries are typically comprised of one or two major staple crops, which, in turn, have speci c requirements in terms of the climate and soils to which they are best adapted. Finally, most previous studies are based on data organized within coarse gridded spatial frameworks that have been found unreliable for estimating crop production potential on existing cropland at sub-national spatial scales 14 . Approaches used to estimate substitution costs to evaluate future global food production capacity are summarized in Supplementary Table 1.
This study overcomes the limitations of past studies by utilizing robust spatial upscaling techniques, well-validated crop-speci c simulation models, and soil, climate, and cropping system databases at ner spatial resolution, using primary data sources as much as possible, to estimate yield potential and yield stability of current and newly developed croplands 3,4 . To determine the crop production "cost" of this substitution, we compare yield potential and yield stability of cropland in regions with contracting or expanding production area in the rst decade of the new millennium for rice in China (irrigated) and Indonesia (rainfed and irrigated), and rainfed maize in Nigeria. These countries and crops were selected because they: (1) have large populations and associated food demand, (2) are projected to undergo rapid land use change due to urbanization 15 , and (3) the evaluated crops represent major staples in national diets and account for a large proportion of total farmland in each country 16 . To avoid the confounding in uence of differences in sophistication of crop and soil management and use of inputs on converted versus new cropland, we evaluate differences in yield potential instead of current average yields, as the former assumes use of best management practices in both expanding and contracting regions.
Substitution costs estimated in this way are therefore dependent on differences in endowments of climate, soil, and access to irrigation. A spatial framework that delineates areas with similar crop production potential and yield stability based on climate was used to upscale substitution cost estimates from sub-national to national scale 4,5 .

Results
Population growth and urbanization are key drivers of land use change worldwide. For example, in China, total production area of seven major staple food crops decreased substantially in regions surrounding the most rapidly growing cities during the 2000-2010 period (Fig. 1). In contrast, cropland expansion occurred in central and northeastern China where urban population growth was much slower.
One reason for differences in total grain production between cropland lost to urbanization and new cropland is cropping intensity (i.e. number of crops grown each year on the same piece of land). In China, irrigated rice area has been decreasing in regions surrounding mega-cities such as Shanghai (current population: 27 M 7 ), Guangzhou (13 M), and Hangzhou (8 M) where warm climate and long growing season allow production of two rice crops per year on the same eld (called double cropping). Expansion of irrigated rice production occurred mostly in central and northeastern regions where only a single rice crop can be grown each year given a cooler climate and shorter frost-free period (Fig. 2a). National average yield potential of converted rice land was 15.2 t ha − 1 compared with 11.2 t ha − 1 for newly established rice land (Fig. 2b). While yield potential for a single crop is highest in central and northern provinces, total annual production potential per hectare is about 70% greater in south and southeast of China due to annual double cropping. Taking into account all areas undergoing cropland conversion or expansion, the area-weighted national substitution cost in China is 1.3 hectares (Table 1), which means that, on average at national scale, proportionally more cropland (1.3x) is required to replace the productive potential of one hectare of rice land lost to urbanization. Not accounting for differences in cropping intensity would lead to the (wrong) conclusion that new crop area in China has higher rice yield potential than crop area lost to urbanization. Yield stability on new and converted cropland, as quanti ed by the inter-annual coe cient of variation (CV), is similar in both cases because rice is produced with irrigation, which avoids yield losses from drought and greatly increases yield stability compared to rainfed crop production 19 . A similar situation occurs in Indonesia, where highly productive irrigated rice area in West and Central Java, the island with fastest population growth (+ 122 km − 2 y − 1 ; in 2000-2010 20 ), has been converted to other uses 21 while rice area expanded mostly into more marginal agricultural regions, with slower population growth, such as in South Sumatra (+ 14) and South Kalimantan (+ 17) (Fig. 2c). Total annual yield potential is about two-fold greater in irrigated double (or even triple) rice systems in West and Central Java compared with those marginal regions, where single-crop tidal and ood-prone rice systems are dominant (Fig. 2d). In these harsher environments, rice is typically grown during the wet season and water supply depends exclusively on ocean tides and rainfall, which allows only one rice crop per year in most cases and reduces yield potential from exposure to both ooding and drought stress during the same growing season 22 . Considering all land conversion and expansion throughout the country, national average substitution cost for rice in Indonesia is 1.3 hectares (Table 1).
In Nigeria, the greatest reduction in rainfed maize area occurred in southern coastal regions with humid tropical climate around Port Harcourt (+ 0.7 M population increase, 2000-2010). Most new maize area came from northward expansion into the more sparsely populated Guinea Savanna region, which has lower annual rainfall and greater year-to-year variation in rainfall amounts (Fig. 2e). As a consequence, rainfed yield potential of new maize land is considerably lower and much less stable than the converted land it replaced, with a national average substitution cost of 1.4 hectares ( Fig. 2f; Table 1). In contrast to Indonesia and China, farmers in most of Sub-Saharan Africa lack adequate access to inputs and extension services. As a result, the difference in potential productivity between new versus converted land reported here would not have been captured if the analysis was based on the very low current yields attained by maize farmers throughout the country (1.8 t ha − 1 ), which would lead to the conclusion that converted land has a substitution cost near unity (Supplementary Table S2).
Sub-national estimates of annual yield potential for new and converted cropland show enormous variation due to endowments of climate and soil. For example, across rice producing regions in China, total annual yield potential ranges from 10 to 19 t ha − 1 in both new and converted croplands (Fig. 2). Similarly, wide ranges of annual production potential can be observed across rice and maize producing areas in Indonesia and Nigeria, respectively. Hence, national average substitution costs based on areaweighted estimates of annual yield potential, as given in Table 1, hide enormous variation in sub-national estimates of annual production potential (Fig. 2). As a result, using a xed substitution cost to estimate the impact of land conversion on crop production at subnational levels can give misleading input to inform national agricultural and land-use policies, including prioritization of investments in agricultural research and development.
Accuracy of substitution cost estimates are highly sensitive to data quality, precision of cropland distribution maps, spatial scale at which the data are analyzed and aggregated, and reliability of crop yield potential simulations. We have con dence in the spatial framework used for upscaling results from sub-national to national scale because it has proven to be robust in estimating yield potential at subnational to national scales for a number of crops and countries across a wide range of soils and climates 5,14,23 . Likewise, crop simulation models used to evaluate crop yield potential have been widely validated in China, Southeast Asia, and Sub-Saharan Africa 14,24,25 . While we attempted to use the best available sub-national data sources for cropland distribution, cropping systems, climate, and soil properties as described by Grassini et al. 3 , data quality is always a concern for long-term weather records and soil properties, which are input to the simulation models and, thus, may be a source of uncertainty 26 .
Similarly, crop models may not account for all possible factors limiting crop production. For example, currently available rice simulation models do not account for the negative effect of alternate cycles of drought and submergence, which are frequent in tidal and ood-prone systems of the new Indonesian rice areas but less common in regions with irrigated production 19 . Similarly, the best available rice models have limited ability to simulate the effects of cold sterility 27 , which may be important for estimating yield potential in high-latitude temperate environments as found in northeastern China where rice production area is expanding. In both cases, inclusion of these factors in simulating yield potential would tend to increase estimated substitution costs as found in this study.

Discussion
Global land-use trends document that increases in crop production area now contribute more to global supply of staple food crops than the rise in crop yields, which reverses trends of previous decades when crop yield increases were more prominent 8 . Reliance on conversion of new land to meet increasing food demand is ampli ed by loss of existing farmland to urbanization. Hence, accurate estimation of the impact from these trends on food production capacity provides critical input to development of agricultural and land-use policies at national and global scales to achieve appropriate balance between food security and environmental goals. Using new methods to make such an assessment, as reported herein, with ner spatial resolution and more robust simulation of crop yields and yield stability than previously possible, we nd that average national substitution costs range from 1.3 for rice in China and Indonesia to 1.4 for maize in Nigeria. Despite relatively little variation in these national averages, there was enormous variation in yield potential of both converted and new land at sub-national scales in all three countries. Hence, xed ratios for estimating substitution costs, as employed in some studies 12 , should be used with caution as input to strategic national land-use plans. Similarly, use of current yields 2 , rather than yield potential, underestimates substitution costs by a large margin when average farm yields in both converted and new land are limited by lack of inputs and technologies to overcome nutrient de ciencies, weeds, and pests, which is the case for maize in Nigeria. In addition, year-to-year variation in Nigerian rainfed maize yield potential is two-fold larger on new rainfed maize land than on farmland lost to urbanization (Table 1), which means food production on new land is much less reliable than on converted land. Similar assessments are possible for other countries that have sub-national data on changes in population 7 , crop production area 17 , and crop production systems, soils, and climate 26 .
Conversion of cropland for urban use can be penny wise when substantial pro ts accrue from such land development. But these conversions can also be pound foolish for several reasons when new cropland has substantially lower yield potential, less yield stability, or both. First, a substitution cost greater than one increases pressure to further expand cropland area to meet food demand through conversion of rainforests and grasslands at the expense of biodiversity and other ecosystem services provided by natural habitat. Second, deforestation and conversion to agricultural land use accounts for 17% of global greenhouse-gas emissions contributing to climate forcing 28 . Third, in rainfed systems, reduced yield stability makes it riskier to invest in fertilizer and other inputs to raise yields in new production areas, which in turn contributes to slower rates of increase in crop yields 3,29 . And fourth, while we assessed substitution costs based solely on differences in annual crop yield potential, the overall cost would be higher if one also considers the greater production costs (fertilizer, labor, transportation) and required investments in infrastructure (roads, canals, drainage systems) associated with establishing crop production in remote areas where expansion typically occurs. We conclude that in countries where land substitution costs are large, as found in this study, there is strong justi cation for land-use policies that seek to conserve prime farmland at the periphery of urban areas 2 supported by agricultural development and land-use policies that seek to accelerate yield gains on existing farmland through sustainable intensi cation while also ensuring conservation of natural ecosystems 8,30 . Continuing current land-use trajectories undermines progress towards the tripartite goals of food security, conservation of natural resources, and addressing the threat of climate change.

Competing interest declaration
The authors declare no competing interests.

Estimation of land productivity
Our analysis of farmland substitution costs is based on comparison of annual crop yield potential of converted and new croplands rather than on differences in current farm yields of both land categories. As noted in the main text of our paper, the latter approach can mask differences in the inherent productive capacity of agricultural land, as determined by soil quality and climate, due to differences in sophistication of crop and soil management practices or access to inputs and markets, all of which can limit yields 31 . In many developing countries, and especially at the frontiers of current agricultural areas, farmers have limited access to inputs, equipment, supporting services and technologies. However, we also evaluated substitution costs based on current average yields and the results are presented in Supplementary Table 2 although we believe these results are less useful. For example, substantial funding is allocated by government agencies and charitable foundations (e.g., Bill & Melinda Gates Foundation, CGIAR, USAID--Feed the Future Initiative) to improve farmer access to markets, technologies, and information in developing countries. Therefore, an analysis of land substitution costs to inform national policies concerning agricultural development and land use policies based on current yields would not only mask the potential cost of cropland substitution based on use of modern farming practices, but it would also quickly become outdated as farmers gain access to markets, technologies, and information.
Yield potential is the yield of a crop cultivar when grown with water and nutrients non-limiting and biotic stress effectively controlled 11,32 . Under these conditions, crop growth rate is determined by solar radiation, temperature, atmospheric CO 2 , and genetic traits that govern the length of growing period and light interception by the crop. For rainfed crops, rainfall amount and distribution and soil water holding capacity also impose an upper limit to crop productivity. Hence, yield potential is the most relevant parameter for estimating crop production potential of irrigated crop systems, while water-limited yield Three countries undergoing rapid urbanization during the last few decades were selected as case studies 36 . Crop area distribution in 2000 (average for 1999-2001) and 2010 (2009-2011) 17 from MAPSAM was used to estimate net change in crop area for that 10-year period in each buffer and CZ, which in turn was used to identify areas of rapid crop expansion or contraction for maize (Nigeria) and rice (China, Indonesia). In the case of Nigeria, maize is grown under rainfed conditions, which means crop growth depends on stored soil water at sowing and in-season rainfall to meet its water requirements. In the case of rice, nearly all rice production in China occurs on irrigated land, while both irrigated and lowland rainfed rice are grown in Indonesia.
Locally calibrated crop models were employed to estimate yield potential of rice or maize in each buffer within each country 14,24,25 . We used Hybrid-Maize 37 for maize simulations in Nigeria and Oryza V3 38 for rice simulations in China and Indonesia. Ten or 15 years of weather data were employed for yield potential and water-limited yield potential, respectively, as per GYGA methodology 26 . Soil and crop management data, including cropping intensity within each buffer were collected with the assistance of local agronomists. Current yields were obtained from o cial statistics at the lowest administrative level at which they are available within each buffer, for the most recent ve years. Using a longer time period would bias estimation of current yields due to in uence of a technology trend 39 . Details on the methodology followed to estimate yield potential in each country and data sources can be found in Supplementary Table S3 and elsewhere 13,21,24 .
Current yields and yield potential, as well as crop intensity and yield stability, in buffers experiencing large cropland substitution of rice or maize were compared with those at buffers where cropland is currently expanding (Figure 2). We estimated yields on an annual basis to account for the higher crop intensity in those regions where two or even three crops were produced each year on the same piece of land (rice in Indonesia and southern China). Then, for each country, we calculated the average annual yield (either current or potential) in CZs with expanded or contracted crop-speci c area, weighted by the crop area net balance within each CZ (2000-2010). National average substitution costs were computed as the ratio between the weighted average yield in CZs with contracted area versus the weighted average yield in CZs with expanded crop area in the study period. For comparison, substitution costs were estimated separately based on either current yields (Supplementary Table 2) or potential yields as reported in Table   1. In this study, yields are reported at 15.5 and 14% seed moisture for maize and rice, respectively, which correspond to the commercial yield reporting standards for these crops.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. Figure 1 Changes in cropland area for seven major staple food crops (rice, maize, wheat, soybean, barley, sorghum, and cassava)17,18 and change in population of major cities from 2000 to 20107. Labeled cities correspond to urban centers with population growth larger than 2 million inhabitants during that period. Note: The designations employed and the presentation of the material on this map do not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.