Future urban land expansion and implications for global croplands

Future urban expansion is highly likely to occur in areas currently under cultivation (Fig. 1). Globally, 46 Mha (medium scenario; range from low to high scenario: 43–55 Mha) of croplands in 2000 are located in areas that are expected to be urbanized by 2030, corresponding to 3.2% (3.0–3.8%) of existing cultivated land. However, urban agriculture is known to be significant in many cities. Hence, we account for urban agriculture by overlaying maps of urban areas and croplands for the year 2000, and find that, on average, 36% of all urban areas are used for crop production. We assume this percentage of urban agriculture to prevail when urban area expands but account for regional variation (for example, 41% in Asia and 32% in Africa; see Supporting Information for details). Accounting for these prevailing cropland fractions, total cropland loss amounts to 2.0% (1.8–2.4%) of the global total—around 30 Mha (27–35 Mha), with countries such as China, Vietnam, and Pakistan ranging between 5 and 10% (Table 1).

Although the aggregate impact of urban expansion on global cropland is modest, regional impacts will be acute and differentiated. In the medium urbanization scenario, Asia and Africa will experience around 80%, or roughly 24 Mha, of the total global cropland loss. The most affected regions in Africa include Egypt, Nigeria, and the region surrounding Lake Victoria Basin in Eastern Africa (Fig. 1). In Asia, the hot spots of cropland loss are river valleys and coastlines, many of which are in the vicinity of MURs, such as the Bohai Economic Rim and the Yangtze River Delta in China, or Java Island in Indonesia (Fig. 2).

One-fourth of total global cropland loss will occur in China. Urban expansion in China is taking place in the country’s most productive farmland and over large areas. Therefore, urban expansion could pose a threat to domestic crop production. In contrast, India, the United States, and Brazil will also experience high losses in absolute terms, but here urban expansion leaves large expanses of croplands untouched, and is therefore less likely to threaten domestic crop production (Table 1).

Future urban land expansion will continue to take place on prime agricultural lands. We observe a total loss of crop production of 3.7% (3.4–4.2%) due to urban expansion. On average, the cropland lost to urban expansion is 1.77 times as productive as the average global croplands. Our results hence confirm evidence from local case studies (7–9), indicating that urban agglomerations are surrounded by croplands with above average productivity.

Our analysis shows that 84% of global production losses are expected to occur in Africa and Asia (Table 1). The 3% cropland loss in Asia translates into a 6% production loss (Table 1). In Africa, the effects are tripled: a 3% cropland loss translates into a 9% crop production reduction, most of which will take place in Egypt and Nigeria. Only a few countries display urbanized cropland with below national average agricultural productivity, the United States being the most prominent example. China and India will continue to urbanize rapidly, but with different spatial patterns and development dynamics. China’s croplands are concentrated along the coastal areas and in the east of the country (Fig. 1). By 2030, most of the urban land cover expansion is expected to occur in that region. The analysis reveals relative cropland losses of 5–6% (8–9 Mha) and productivity losses of 8–10% (128–153 Pcal) between 2000 and 2030 (Table 1). Results for India are markedly different. Total urban extent in 2000 is an order of magnitude smaller than in China (3 Mha compared with 8 Mha), and absolute urban area expansion until 2030 is expected to cover one-half as much area as in China (3–4 Mha compared with 7–8 Mha). This difference in urban expansion is in large part explained by very different urbanization and urban expansion trends (3). Whereas China’s urban population exceeded its rural population in 2012 and is expected to be 75% of the total population by 2050, India’s urban population is currently less than one-third of the total population and by 2050 will just be over one-half. Furthermore, as of 2011, 79% of India’s total population resided in settlements of 100,000 or fewer, and 52% of the country lived in towns and villages with populations fewer than 5,000 (18). This is in stark contrast to China. Although cropland loss is currently not an issue in India (about 2% by 2030; Table 1), other studies corroborate that it is likely to become more significant in the future when the country’s urban expansion begins to accelerate (19).

In African countries, there will be significant variation in the geographic distribution and rates of cropland loss. Croplands in less arid zones are expected to be relatively less affected by urbanization. Nigeria, Africa’s most populous country, will experience high rates of urban expansion and 5–7% cropland loss (Table 1). Urban expansion will be concentrated along the continent’s coastlines, whereas the majority of cropland lies inland (Fig. 1). The region around Lake Victoria will experience the highest rates of urban expansion. In particular, for Burundi and Rwanda, the high rates of expected cropland conversion to urban (∼28 and 34%) reflect the limited availability of land in those countries.

Our disaggregated analysis for individual staple crops shows their relative importance in urbanizing areas. In 2000, 4% of maize, 9% of rice, 2% of soybean, and 7% of global wheat production were grown in areas that are forecast to be urbanized (Table S1). Although the results for Europe (range between 2 and 3%), the Americas (1–2%), and Australasia (all <1%) indicate low competition for these key staples, the findings for Asia and Africa suggest significant losses of specific crops. In Asia, 10% of maize, 9% of rice, 7% of soybean, and 13% of wheat production were produced in areas that will be urbanized by 2030. In Africa, these shares range from 11% of soybean production to 26% of the continent’s wheat production (14% maize, 19% rice).

We further analyzed cropland loss for a selection of MURs, defined as continuous urban regions with multiple urban centers and a combined population greater than 20 million, often expanding over 10,000 km². Prime agricultural lands are especially vulnerable to conversion in MURs with estimated cropland losses between 0.1 and 1.2 Mha for the 11 case studies (Fig. 2 and Table S2). With the exception of the US Northeast, the productivity of the cropland converted in MUR is higher than national averages (Table S2). Notably, in MURs of India, Bangladesh, and Indonesia, the relative productivity is >2 (Fig. 2 E–G). In Chinese MURs, the relative productivity is 1.05–2.05 (Fig. 2 A–C).

To understand agricultural production patterns around these evolving MURs, we analyzed the harvested area fraction (HAF)—the ratio of harvested area of a specific crop over the total harvested area—in competing areas of the abovementioned staple crops and a selection of cash crops specific to some of the MURs (cacao, oil palm, sugarcane; Table S2). The aggregated HAF for these crops is high in most of the MURs. In the Yangtze River Delta around Shanghai, for example, the combined HAF of rice and wheat accounts for 50% of total area harvested in competing areas. In contrast, the combined HAF is very low for the United States, Brazil, and Japan, indicating that these areas are used to grow other crops such as vegetables. HAF is also low for the Greater Ibadan Lagos Accra (GILA) corridor in Western Africa, where these crops only contribute marginally to diets. The prevalence of the cash crops analyzed is comparatively low (the exception is sugarcane around Delhi with HAF of 18%).

The spatial pattern of urban expansion plays an important role in cropland loss. MURs are often characterized by multiple urban centers, with productive cropland distributed throughout the urban fabric. Although the aggregate amount of cropland in these regions may be high, each patch of cropland is relatively small and thus vulnerable to urban envelopment (Pearl River Delta, Fig. 2A). In regions with a single dominant urban center, such as Greater Delhi (Fig. 2E), urban envelopment of cropland is still contained around the urban core, with little evidence of large-scale continuous urban fabric development. Cropland in these regions will continue to be converted (19), but not at the same magnitude as in multinodal urban regions.

As urban areas expand, the remaining croplands and farmers at the periurban interface experience greater competition for water and increased exposure to climate hazards. The urban expansion into the Ganges-Brahmaputra Delta, for example, has resulted in the loss of wetlands and water bodies that serve as flood protection (20). In addition, cropland conversion led to a sinking of the delta due to a combination of sediment loading, compaction, ground water extraction, and reduced aggradation. This makes the delta increasingly vulnerable to hazards associated with climate change, such as sea level rise (21), and threatens not only urban areas but also the remaining croplands that were largely used to feed the regional population with rice (HAF of rice >83%; Table S2).

Sea level rise and subsidence are also significant concerns for Greater Cairo, because a considerable fraction of the Nile Delta is already near or below sea level and expected to sink further (22). Diminishing sediment discharge due to dams in the south will increase the pressure on the delta, which will eventually decrease in size (23). Our results show that urbanization converts precarious croplands at high rates along the Nile even though they are important for maintaining food supply of the urban centers [combined HAF for wheat and maize, 49% (Table 1 and Table S2)]. Efforts to divert urbanization away from the fertile lands into the deserts are underway but have been less effective than hoped (23).

Our study shows that future urban expansion is expected to convert 27–35 Mha of croplands (1.8–2.4% of global cropland and 3.4–4.2% of the yearly production) globally between 2000 and 2030, adding an additional component to the emerging global consequences of land use (24). On average, this amounts to an annual land consumption of 1 Mha, which is almost a third of the annual agricultural expansion between 1961 and 2009 of 3.38 Mha⋅y⁻¹ (25). Our study is limited by the spatial resolution of the analysis; although higher-resolution data would generate more detailed insights, these results provide a global assessment of the patterns of likely cropland loss due to urban expansion.

We base our study on a spatially explicit urban area expansion probability dataset (4) and two gridded datasets on global croplands in 2000 (15) and 2005 (16). We use a dataset on gridded global crop yields in 2000 (17) to calculate the productivity of the displaced land. Yields of the 16 most important crops (listed in Supporting Information) are converted to calories and aggregated in a single dataset, weighted with area harvested. We supplement this with a disaggregated analysis of four staple crops (maize, rice, soybean, and wheat) and three cash crops (cacao, oil palm, and sugarcane). We assess the impact of urban area expansion by intersecting three distinct urbanization projections for the year 2030 with the cropland dataset for the year 2000. The resulting cropland and production loss scenarios are “low” (with a restrictive threshold including only grid cells exceeding 87.5% urbanization probability), “medium” (>75% urbanization probability), and “high” (>50% urbanization probability). As a “best guess,” we assume that all grid cells with >75% probability of becoming urbanized (medium scenario) will be affected by urbanization until 2030. Please see Supporting Information for a detailed description.

We base our study on a spatially explicit urban area expansion probability dataset (4), projected in Goode's Homolosine Equal-Area projection. The dataset uses urban extent in 2000 as baseline and forecasts urban area expansion for the year 2030. We use two gridded datasets on global croplands in 2000 and 2005 [from EarthStat (15) and from the International Institute for Applied Systems Analysis and the International Food Policy Research Institute, IIASA–IFPRI (16)]. The EarthStat cropland map by Ramankutty et al. (15) combines agricultural inventory and satellite-derived land cover data in a 5-min grid (∼10 km at the equator). We integrate a newly developed high-resolution map of global cropland from IIASA–IFPRI by Fritz et al. (16) (∼1 km) into our analysis. Both datasets are in GCS_WGS_1984 projection. We use the EarthStat dataset on gridded global crop yields by Monfreda et al. (17) in 2000 to calculate the productivity of the displaced land. Yields of 16 important food crops (barley, cassava, groundnut, maize, millet, oil palm, potato, rapeseed, rice, rye, sorghum, soybean, sugarbeet, sugarcane, sunflower, and wheat) are converted to calories and aggregated in a single dataset. We supplement this with a disaggregated analysis of four staple crops (maize, rice, soybean, and wheat) and three cash crops (cacao, oil palm, and sugarcane). The dataset uses GCS_WGS_1984 projection with a spatial resolution of 5 min. We reproject the urban expansion dataset to fit GCS_WGS_1984 projection. We aggregate the IIASA–IFPRI map to the same resolution as the initial urban expansion forecast by assigning the mean values of the ∼1-km pixels to the resultant ∼5-km pixel. We create separate shapefile layers for a selection of megaurban regions (MURs) by creating a 100-km buffer around the geographic centers of the most prominent urban centers of each region. For Java, we take the administrative boundaries of the island. We define MURs as continuous urban regions with multiple urban centers and a combined population greater than 20 million, often expanding over 10,000 km² or more (with the exception of Greater Cairo). Our selection is based on our cropland loss findings and mostly entails countries from developing or emerging regions. For comparison, we supplement them with MURs from developed regions (the United States and Japan). Note: this list of MURs is not comprehensive. We further include World Bank data on poverty (58) and FAOSTAT data from the Food and Agriculture Organization of the United Nations (25) to supplement our findings (Table S3).

We assess the impact of urban area expansion by intersecting three distinct urbanization projections for the year 2030 with the EarthStat cropland dataset for the year 2000 (at 5-min resolution). The resulting cropland and production loss scenarios are “low” (with a restrictive threshold including only grid cells exceeding 87.5% urbanization probability), “medium” (>75% urbanization probability), and “high” (>50% urbanization probability). As a “best guess,” we assume that all grid cells with >75% probability of becoming urbanized (medium scenario) will be affected by urbanization until 2030.

If the urbanization probability of a grid cell exceeds one of the predefined thresholds, it is assumed that it will be urbanized. Loss of cropland due to urban area expansion is only calculated for grid cells with a cropland area fraction >0%. The corresponding cropland areas (and corresponding production metrics) are aggregated and subsequent calculations made at national level (for the MURs at the corresponding regional level). We assume that cropland area is partially maintained within the predominantly urbanized regions to account for the potential of urban agriculture. For each country, we assume that cropland area is only lost to the extent that it exceeds the prevailing fraction of the initial cropland area. This prevailing fraction is estimated for each country by intersecting urban areas around the year 2000 with the cropland map for the year 2000. For example, if croplands cover 25% of urbanized areas in a country in 2000, we assume that 25% of the initial cropland area in the newly urbanized area is maintained, whereas the remaining cropland area is lost.

A similar procedure is applied for calculating the corresponding production losses. We use the aggregate production in urban areas in 2000 to calculate the average productivity in million calories per km² of croplands in urban areas in 2000. This average productivity is then used to compute the crop production on the prevailing cropland in competing areas. Loss of production due to urban expansion is the total production in competing areas minus the so-computed production on the remaining croplands. We assume that the production is spread equally over the cropland of a grid cell.

For the disaggregated crop specific analysis, we abstract from calculating actual losses. Instead, we use the amount of crop production [and the average harvested area fraction (HAF) for the MURs] in competing grid cells as an indicator of the relevance of a specific crop in a specific area. We intersected the urban area expansion dataset with the production and HAF datasets of the respective crops. The HAF represents the ratio of area harvested of a specific crop over the total area harvested in a grid cell. We aggregated the production in the grid cells with an urbanization probability >75% (medium scenario) and compare it to the total production of the country/region. For the HAF, we computed the average in all competing pixel per MUR.

We cross-check our cropland loss estimates by intersecting the urban area expansion forecasts with the generated cropland map at ∼5-km resolution (Table S5), finding little to no variation on aggregate (both 29.9 Mha of total cropland loss; medium scenario). Some variation was visible in Africa and Asia (6.1 vs. 5.6, and 17 vs. 17.9, Mha, medium scenario, high- to lower-resolution analysis), which can be explained by the differences between the two cropland datasets as discussed in detail by Fritz et al. (16).

We acknowledge the contribution of Ulf Weddige from the Mercator Research Institute on Global Commons and Climate Change, Graham Furniss from the University of Canterbury, and Sung Bae from the BlueFern Supercomputing Services. We thank the EarthStat team for support in data-related inquiries. Thanks to Stockholm Resilience Centre, University of Gävle, and Formas for funding S.B.; Stiftung der Deutschen Wirtschaft for funding C.B.d.; ERC-2010-stg-263522 LUISE for funding K.-H.E.; and NASA Grants NNX15AD43G and NNX11AE88G for supporting B.G. and K.C.S.