Introduction

Rapid urbanization is consuming farmland globally, affecting not only in developed nations but also in rapidly developing regions such as Southeast Asia and Africa, where this phenomenon is especially pronounced1,2,3. China, as one of the fastest-growing countries in Asia over the past 30 years, exemplifies this trend, with the expansion of urban land serving as both a driver and a consequence of its rapid economic growth, complicating efforts to protect farmland4,5,6,7. Thus, urbanization is often viewed as a major threat to food security. However, this view often overlooks the nuanced impacts of urbanization on farmland and the role of policy in farmland protection8,9. Urbanization exerts pressure on farmland by converting prime agricultural land into urban built-up areas6,7. Conversely, it also relocates substantial populations from rural areas to cities, potentially freeing up rural residential land10,11.

To address the pressures on farmland, China has implemented policies like the Farmland Requisition-Compensation Balance (FRCB) policy of 1997, designed to maintain food security12,13. The FRCB mandates that any farmland used for urban development must be compensated by reclaiming an equivalent amount of land elsewhere, such as forests, grasslands, or wetlands14. Furthermore, government subsidies have encouraged the conversion of natural land to agricultural use in poorer regions, such as Northwest China. Thus, the overall impact of urbanization on farmland security hinges on the effectiveness of these policies15.

In this study, we aim to provide a comprehensive analysis of urbanization’s impact on farmland conservation and food security by: (a) quantifying and comparing direct urban land conversion with potential rural land savings from rural-urban migration; (b) evaluating the FRCB policy, including its spill-over effects on other land-use types and its effectiveness in achieving food security; and (c) simulating how different urbanization modes might affect China’s farmland conservation and food security goals.

Although the FRCB policy keeps China’s total farmland area constant through land-use conversions, previous studies have overlooked the ‘ripple effects’ of farmland reclamation on other land uses. These effects span large distances, often relocating fertile coastal farmland to less fertile northern regions, jeopardizing both food and ecological security. Moreover, the phenomenon of ‘double-dipping’—where rural populations decrease but rural built-up land increases due to the urban-rural dual structure and household registration policy—aggravates these ripple effects. Our study reveals that expanding rural built-up land amid a declining rural population result in low land-use efficiency. To mitigate these ripple effects, current land-use conversion practices must change to reduce long-distance cropland displacement. By integrating the Future Land Use Simulation (FLUS) model and the Global Agro-Ecological Zones (GAEZ) yield-estimation model (See Supplementary Fig. 1), we simulate the total area and potential yield of farmland that could be conserved by reducing rural built-up land in line with the urbanization of the rural population.

Results

Ripple effects of rapid urbanization

We begin by assessing the outcomes of farmland protection efforts over a 20-year period using high-resolution land-use data. Our quantitative analysis reveals that while China’s total farmland area remained relatively stable from 1990 to 2010, this equilibrium was maintained through complex land-use conversions across regions—a phenomenon we term the ‘ripple effect’. Specifically, farmland converted to urban use was compensated by reclaiming other types of land, such as forests, grasslands, and wetlands (termed ‘natural land’). As illustrated in Fig. 1, between 1990 and 2000, 1.54 million hectares of farmland were converted to urban areas, while 4.40 million hectares of other land types, including natural and unused land, were transformed into farmland. This resulted in a net increase of 2.86 million hectares of farmland and maintaining overall stability. Similarly, from 2000 to 2010, 2.03 million hectares of farmland were urbanized, with only 0.82 million hectares from other land types were converted back to farmland, leading to a net decrease of 1.20 million hectares. Despite this decline, the overall system still exhibited a degree of stability. The first decade primarily featured the replacement of farmland with other land types, whereas the second decade not only continued this trend but also resulted in a net loss of 1.21 million hectares of farmland. This conversion process underscores the tension between urban development and ecosystem health, as farmland acts as a critical interface between natural landscapes and urban expansion.

Fig. 1: Illustration of land-use change patches in China from 1990 to 2000 and 2000 to 2010, showing areas involved in land-use transitions.
figure 1

Land-use data was obtained on a national scale for 5-year periods from 1990–2010, using land-use change data from the Chinese Academy of Sciences (CAS)46.

Full size image

Our analysis further indicates that the FRCB policy, intended to preserve farmland and ensure food security, may inadvertently degrade farmland quality. The net changes in farmland area were +1.6, +1.7, −0.7, and −0.6 million hectares during the periods 1990–1995, 1995–2000, 2000–2005, and 2005–2010, respectively, with no overall decrease in the total farmland area (Fig. 2a). However, the newly reclaimed land, often of lower quality, replaced fertile farmland lost to urbanization. High-resolution datasets show that farmland loss primarily occurred in flat, fertile areas around cities, especially in the highly developed eastern coastal regions. This lost farmland was compensated by reclaiming land in barren and ecologically sensitive areas, particularly in the less-developed northern regions16,17. Long-distance land displacement thus compromises the FRCB’s effectiveness, as it results in the loss of fertile farmland and the degradation of sensitive natural lands—a trend noted in several studies18,19. In response, China introduced policies to prohibit farmland requisition-compensation between different provinces, aiming to reduce regional disparities and protect remote environments through local balance18,19. Despite these efforts, the updated policy has not fully addressed the insufficiency of newly claimed cropland.

Fig. 2: Comparison of the area and potential yield of reclaimed and lost farmland in China from 1990–2010 for every 5 years.
figure 2

a Total area of farmland loss (the red bar indicates farmland converted to built-up land and the gray bar indicates farmland converted to other land-use types) and reclaimed farmland (blue bar). b Average potential yield of farmland lost to built-up land and of reclaimed farmland.

Full size image

To quantify the productivity loss from long-distance farmland displacement, we used the GAEZ yield estimation model developed by the FAO and IIASA20. This model incorporates climate, soil, terrain conditions, and cropping systems to predict potential farmland yield (see Methods)21,22. The bar chart in Fig. 2b shows that the average potential yield of reclaimed farmland was significantly lower than that of farmland converted to urban use in each 5-year period between 1990 and 2010. Over the past 20 years, urbanization has resulted in an average productivity loss of 10,842 kg/ha for the lost farmland that has been converted. In contrast, the average potential yield of reclaimed farmland is only 6771 kg/ha, representing a 37.5% reduction compared to the yield of lost farmland. This indicates that the productivity of the lost farmland is 1.6 times greater than that of the reclaimed farmland, suggesting that an increase in farmland area by a factor of 1.6 is required to offset the loss in agricultural production.

Further negative impacts include biodiversity loss and environmental degradation of natural systems, driven by spatial tele-coupling effects that undermine the “local nature” of distant ecosystems. As shown in Fig. 3, new farmland, mainly converted from natural land, is concentrated in humid to semi-humid Northeast China and arid to semi-arid Northwest China—both ecologically fragile regions. Farmland loss predominantly occurs in the middle and lower reaches of the Yangtze River due to urban expansion and insufficient farmland reserves. This geographical displacement facilitated by the FRCB endangers China’s ecological security, as tele-coupling leads to resource loss and ecosystem function changes. These changes also exacerbate unequal access to ecosystem services and poverty23,24,25,26.

Fig. 3: Spatial pattern of three types of land-use transition under the ripple effects over 20 years.
figure 3

First, we identified the three types of land-use transition based on the 30 m resolution land-use dataset. They are built-up land encroaching farmland (Frl. to Btl.), farmland reclaimed from natural land (Nat. to Frl.), and natural land supplied by unused land (Unu. to Nat.). Then, the area proportion of the three types of transition in each 10 × 10 km pixel cell was calculated and set as the components of tricolors to show the spatial pattern of land-use transition. The figure highlights farmland loss in the eastern coastal developed areas and reclaimed in the northern part of China.

Full size image

The natural state of the land in these regions makes it unsuitable for reclamation. Northern meadows are vulnerable to soil erosion, and northeast wetlands have critical ecological functions. The indirect costs of reclamation are high, including landscape fragmentation, biodiversity loss, climate change, and desertification. These conversions mainly occur in less-developed areas with poor living conditions and fragile ecosystems, exacerbating regional inequality in China27,28. This study underscores the need to reconsider current land-use conversion practices to mitigate the adverse ripple effects of the FRCB policy and promote sustainable urbanization and farmland protection29.

The hidden threat to farmland: rural expansion

While urbanization is often blamed for farmland loss, a less obvious but significant factor is the rapid expansion of rural built-up areas, despite a decreasing rural population. This phenomenon, known as “double-dipping”, presents a complex challenge to land use and food security in China. Between 1990 and 2010, China’s urban and rural built-up areas expanded significantly, leading to substantial farmland losses. Surprisingly, rural built-up land increased across the country, from remote regions to highly developed areas, even as the rural population declined by 20% (see Fig. 4).

Fig. 4: Comparing built-up land and population in urban and rural China during 1990–2010.
figure 4

a Urban population growth and urban built-up land increased in China during 1990–2010. The pale blue rectangle denotes the urban built-up land in 1990, the blue and brown patches show the new increased urban built-up land from farmland and other types of land respectively. b Rural population declined and urban built-up land increased in China during 1990–2010. The pale blue rectangle denotes the rural built-up land in 1990, the blue and brown patches show the new increased rural built-up land from farmland and other types of land respectively. c Here we see the change of urban and rural built-up land per capita during 1990–2010. The figure shows the built-up land use in urban areas become more and more compact, while the opposite was true in rural areas. Rural population decreased but rural built-up land increased rapidly. This can be called the “double-dipping” of rural land in China.

Full size image

Several institutional systems in China contribute to these issues: (1) Urbanization Trends and Economic Development: As the economy expands, a growing number of individuals migrate from rural areas to urban centers in search of improved opportunities and a modern lifestyle. Urban governments face the dual challenge of accommodating the housing needs of these new residents while also responding to economic incentives tied to land financialization, which directly fuels urban land expansion. (2) Land Quota System and Spatial Planning: This stringent land management system imposes strict limits on the amount and transformation of urban land. While it aims to regulate land resource development and prevent chaotic growth, it inadvertently drives up urban land prices and causes housing demand from the “floating population” to spill over into rural areas. (3) Dual Land-Tenure System: This fundamental framework creates a clear division between rural and urban land ownership, with urban land typically state-owned and rural land collectively owned. This division not only affects the economic value and developmental potential of the land but also shapes its management and planning. The lax oversight of rural land often leads to its inefficient or informal utilization. (4) Household Registration Policy: This policy reinforces the divide between urban and rural residents by limiting access to urban public services for rural inhabitants. As a result, many rural workers, often referred to as the “floating population,” continue to maintain their rural homes despite relocating to urban centers for employment, driven by these restrictions and economic necessities. (5) Cultural Factors: In many rural communities across China, there exists a deep cultural attachment to the land. This strong connection often motivates individuals to pursue the construction of homes in rural settings, further complicating the dynamics of urbanization.

Rural-urban migrant workers often face restrictions in accessing urban public services, including housing, education, and healthcare. As a result, they tend to work temporarily in cities without permanent settlement. Meanwhile, they also maintain or expand their rural homes for family members left behind, even they have already settled in the cities. Occupying land bot in urban and rural areas, leading to inefficient land use30,31. This situation has led to inefficient use of rural built-up land that could otherwise be used for farming32. The expansion of rural built-up areas, combined with urban growth, poses a significant threat to China’s farmland and food security. Addressing this issue will likely require reforms to the current institutional systems. These reforms should aim to: (1) to improve access to urban services for migrant workers; (2) encourage more efficient use of rural land; (3) balance urban development with the preservation of farmland. By tackling the root causes of “double-dipping”, China can work towards more sustainable land use practices and better safeguard its agricultural resources.

Turning the tide: migrant-inclusive urbanization

Improving land-use efficiency necessitates a more inclusive urbanization strategy that fully integrates rural migrants into urban environments. The rapid expansion of rural built-up areas, driven by migrants’ insecurity in cities, exacerbates land inefficiency (Fig. 5). Despite significant urban population growth and economic development, rural areas are expanding up to four times faster than urban regions. Aligning the built-up land in rural areas with the actual rural population could significantly free up land resources, thereby supporting both the land requirements for urban development and farmland conservation goals. An inclusive approach to urbanization could facilitate the reclamation of urban-encroached farmland by releasing land in rural areas, mitigating adverse ripple effects, and providing an alternative to long-distance displacement for farmland preservation. Although migrant-inclusive urbanization is not a panacea for the complexities of urban-rural tele-coupling effects, it does afford rural migrants the right to purchase urban homes, thereby reducing their settlement costs from a policy perspective. Furthermore, by utilizing policies such as the Farmland Reclamation and Conservation Border (FRCB), this approach can increase urban land supply, potentially alleviating urban property prices and further lowering housing costs for migrants. This strategy not only assists migrant workers in urban living but also diminishes the need for repeated rural home construction.

Fig. 5: Migrant-inclusive urbanization help to slow down the ripple effects.
figure 5

a Pre-conversion land use pattern before land use changes. b Tele-connections of land use change, illustrating transitions from farmland to urban land, from natural land to farmland, and from unused land to natural land (ripple effects). c Inclusive urbanization helps to slow down the ripple effects: reduced land use conversion area and the restoration of rural built-up land to farmland.

Full size image

In recent years, rural land consolidation in China has transitioned from concept to practice. The primary objective of this initiative is to optimize land use, addressing critical issues such as urban-rural land imbalances, farmland preservation, and land abandonment. Key strategies employed include land transfer, consolidation, and optimization, all aimed at reducing farmland waste, promoting intensive land use, and enhancing productivity through technological innovations. For instance, in regions conducive to agricultural transformation, land transfer initiatives encourage farmers to lease or transfer their land to cooperatives or companies, with compensation mechanisms in place to protect their interests. This approach not only safeguards farmers’ livelihoods and social security but also mitigates the risk of large-scale land-related unemployment. Successful implementations of these strategies have been observed in provinces such as Sichuan and Gansu, demonstrating the potential for effective rural land consolidation to improve agricultural efficiency and sustainability. To evaluate the effectiveness of migrant-inclusive urbanization, we simulated the potential reduction of rural built-up land using a cellular automat-based model, called Future Land Use Simulation (FLUS) (see Methods). We estimated the quantity and distribution of rural built-up land that could be converted back to farmland. By calculating the hypothetical rural built-up land area required per province based on population ratios and assessing justifiability using variables such as proximity to rivers and roads, we identified patches of land for potential restoration.

The simulation revealed significant areas of rural built-up land that could be reclaimed as farmland. Figure 6 compares the spatial distribution and potential yield of farmland lost to urban expansion (red patches) with farmland restored in rural areas (blue patches). While urban-encroached farmland loss is concentrated around large cities, the more substantial yet scattered loss in rural areas often goes unnoticed. In many provinces, the restorable rural built-up land could offset the farmland lost to urbanization. For instance, in provinces such as Jiangsu, Shandong, Hebei, and Anhui, the area of restorable rural land is four to five times greater than that of urban-encroached farmland. Migrants are more likely to relinquish their rural land when they can more easily settle in cities or towns. Our findings indicate that if this can be achieved, a substantial area of high-quality prime farmland could be restored—approximately 5.77 million hectares, which is equivalent to a potential yield of 60.23 million tons of food. In this regard, the migrant-inclusive urbanization mode is proposed as a solution for sustainable urban land expansion.

Fig. 6: Comparison of farmland lost to urban built-up land expansion and potential farmland that can be restored from rural built-up land under migrant-inclusive urbanization from 1990–2010.
figure 6

The simulated restorable farmland was estimated based on the assumption that rural built-up land will decline in line with the declining rural population and using the same per capita land use as in urban areas. Six typical sites were selected to show the farmland loss to urban built-up land (red patches) and farmland that can be restored from rural land (blue patches). The bar chart shows the total potential yield of farmland lost to urban built-up land (red bars) and the farmland that can be restored from rural built-up land (blue bars).

Full size image

Discussion

This research quantifies the ripple effects of urbanization under the FRCB policy over a 20-year period using a GAEZ yield-estimation model based on high-resolution land-use and population data. We observed that the ripple effects of farmland degradation and declining ecosystem health are pronounced in areas experiencing land conversion, driven by long-distance land-use displacement under the FRCB policy. The phenomenon of dual land waste—occurring simultaneously in both urban and rural China—represents a significant negative externality of the current urbanization model. Our study reveals that the expansion of rural built-up land accounts for 50% of farmland loss, coinciding with a 20% decrease in the rural population during the same period. Our findings indicate that the current urbanization model and FRCB policy have led to land resource waste in both urban and rural areas. We propose a migrant-inclusive urbanization strategy and evaluate its effectiveness using the FLUS model and the GAEZ model to simulate potential reductions in rural built-up land use (see Methods). Our estimates suggest that reclaiming 5.77 million hectares of rural built-up land could yield 60.23 million tons of food, nearly 2.5 times the yield lost to urban sprawl between 1990 and 2010.

In this paper, we explored the relationship between urbanization and farmland protection from an indirect perspective. While urban land expansion is frequently perceived as a significant threat to food security due to the loss of high-quality farmland, it can also play a role in enhancing farmland security by preventing excessive development of rural land—a benefit that has often been overlooked. Public discourse surrounding farmland preservation, particularly in the context of the Farmland Reservation and Conversion to Building (FRCB) policy, tends to focus on the negative impacts of urbanization, resulting in stringent controls on urban development. Although these policies have stabilized total farmland area, they have also inadvertently led to adverse consequences, including remote land conversions, farmland degradation, declines in ecosystem health, and imbalanced urban-rural development. The novel contribution of our study lies in rethinking the indirect impacts of urbanization on food security in China. Unlike most previous research that positions urbanization and farmland protection as fundamentally opposing forces, we propose a long-term perspective aimed at reconciling urbanization with farmland preservation. Specifically, due to China’s unique land quota system, urbanization not only directly encroaches upon farmland but also indirectly contributes to farmland loss in rural areas through remote connections. This study broadens the understanding of the urbanization-farmland security dilemma by considering rural land-use changes and advocating for migrant-inclusive urbanization as a means to achieve sustainable urban growth while effectively preserving farmland.

The perspective we wish to convey is that the impact of urbanization on farmland preservation is multifaceted. Previous studies often portray urbanization as detrimental to food systems and ecological security—an outlook rooted in a longstanding tradition of urban development discourse33. As early as the 1950s and 1960s, Western countries began employing planning tools to limit urban spatial expansion exemplified by Howard’s “Garden City” concept in England. Similarly, American urban planning practices, along with those in Australia, reflected a shift towards centralized planning aimed at curbing urban sprawl34,35,36. In contrast, China’s strict urban land quota system and the policies governing cultivated land occupation and compensation, while intended to restrict urban expansion, have inadvertently led to increased loss of rural arable land through remote coupling. Several factors contribute to this phenomenon, including urbanization trends, economic development dynamics, the land quota system, the dual land-tenure system, the household registration policy, and cultural factors, all of which interplay to create this complex issue. To address these challenges, we advocate for a migrant-inclusive urbanization model that help remove barriers for rural labor and unlock the potential of farmland resources in rural areas. This approach aims to reconcile urban growth with effective farmland preservation, promoting a more sustainable future for both urban and rural communities.

Several concerns arise regarding the difficulty and side effects of policy implementation. Firstly, the affordability of new urban housing for immigrants is a significant issue. The question is who will cover the costs of rural migration, as local governments typically lack the capacity to bear this financial burden. Even if new immigrants can afford urban houses, they may be reluctant to give up their rural land freely. Chongqing, a city in Central China, proposed a solution in 2008 with the “transaction of land ticket” system37. In this system, farmers obtain land tickets from the local government by re-cultivating land, which they then sell at public auctions. Anecdotal evidence suggests that auction prices reached as high as 764,700 RMB per mu38. This system allows farmers to profit from high land auctions and addresses the issue of abandoned rural land in remote areas.

Even if the costs of land restoration are covered, labor shortages in agriculture and environmental stresses due to increased urban population density and agricultural industrialization remain concerns. The environmental issues challenge city management, but evidence suggests that environmental problems can be handled more efficiently in densely populated urban areas compared to scattered rural settlements. Agricultural industrialization also poses environmental risks, such as soil health degradation and greenhouse gas emissions39. These problems may be managed more effectively in a centralized system. China must cautiously address these side effects while promoting agricultural industrialization.

Collective efforts are essential to implement migrant-inclusive urbanization. Reforms in household registration policies, equalization of basic public services, and free labor flow are necessary. Optimal allocation of labor and land resources should be the primary goal. Recent reforms in rural land-use policies signify a transition from local balance strategies to a framework that allows farmland requisition and compensation between provinces. Initiatives such as the “Link the increase in urban construction land with the decrease in rural construction land” (the Link Policy), have the potential to optimize labor and land resource allocation40,41. This approach aims to enhance overall efficiency in land use while addressing the competing demands of urban expansion and rural preservation.

This study raises the question of whether urbanization can progress while protecting farmland security. Developing countries, such as Southeast/South Asia, with high population densities and low urbanization levels must devise policies that balance urban expansion with farmland preservation42,43. Our research is a starting point for exploring this balance in different developing countries, where the dilemma of urbanization versus prime farmland protection requires further investigation.

Methods

We begin by quantifying land use changes across different types from 1990 to 2010, using the National Land Use/Cover Dataset of China (NLCD-C). We compare the contributions of urban expansion and rural built-up area changes to farmland loss and assess the FRCB’s spill-over effects on other land uses. To evaluate the policy’s impact on food security, we estimate and compare the average potential yield of newly reclaimed farmland against the loss of farmland due to urban expansion. Additionally, we simulate the effects of migrant-inclusive urbanization for the future by using FLUS model and GAEZ model (See Supplementary Fig. 1).

Quantify the ripple effects in land-use change

We used the NLCD-C from 1990–2010 to identify land-use changes in this study. It was the most authoritative and widely-used land use/cover data in China, produced by the Chinese Academy of Sciences coordinated a nationwide land use/cover classification project involving eight research institutions and ~100 scientists44,45,46. A hierarchical classification system of 25 second-level land use/cover classes were grouped into six first-level classes (Supplementary Table 1) from 1980s to 2015 at a spatial resolution of ~30 m47. According to the data report, the accuracy of the selected first-level types exceeds 95.41%, with producer’s accuracy for urban land and rural settlement at 91.86% and 91.67%, respectively48. This dataset is the primary source for studies on land use/cover changes in China since the 1980s44,45,46 and provides crucial data for global environmental change and carbon cycle research49,50,51,52. Compared to global land-cover classification systems, the NLCD-C emphasizes land utilization by humans. For example, built-up land is detailed into urban and rural settlements and other types of built-up land.

We categorized the detailed land-use classes in the databases into four primary types: natural systems (forests, grassland, and water bodies), farmland, built-up land (urban and rural), and unused land (including marginal types such as abandoned land, saline-alkali land, and rocky flatland, as per the Chinese land-use classification). We calculated the total area for each type of land use for the years 1990, 2000, and 2010 using the datasets. To assess changes in land use between these years, we employed the transformation matrix method, which performs spatial overlay analysis to quantify transformations between various land-use types over 10-year intervals53. This method enabled us to locate and quantify land use/cover transformations in China during 1990–2000 and 2000–2010, revealing the ripple effects described in the study.

Figure 3 illustrates the flow process of these ripple effects. Using visualization technology, we depict the quantity and direction of land change. Three colors represent different types of land change: red for farmland to built-up land, green for natural land to farmland, and blue for unused land to natural land. Figure 3a, b show the spatial distribution of land-use changes for the periods 1990–2000 and 2000–2010, respectively. We calculated the area of the three types of land change within each 10 × 10 km grid and determined the area ratio for each type of land change. These ratios were then used to create a color image, with the dominant land change type in each grid represented by its corresponding color.

Compare built-up land expansion in urban and rural area

We compared the urban and rural built-up land change and population dynamics as shown in Fig. 4, where the change in urban and rural built-up land was analyzed based on the national land use/cover database of China. The urban and rural population dynamics data are from the National Bureau of Statistics of China54. According to the definition used by the National Bureau of Statistics of China, the permanent population includes (1) registered permanent residents living in the city, (2) unregistered permanent residents who have lived in the city for more than half a year, and (3) registered permanent residents of the city who have lived outside the city for less than half a year55. We then calculated the built-up land per capita in urban and rural areas. Figure 4b shows the disproportionate amount of rural built-up land relative to the rural population. Figure 4c illustrates the increasing trend of built-up land per capita in rural area, showing that despite the population migration out of rural areas, rural built-up land has actually increased rapidly40,41. Supplementary Fig. 2 shows the spatial distribution of the rural built-up land increase between 1990–2010. An increase in rural built-up land was a widespread phenomenon across China, including both remote areas such as Xinjiang to highly developed regions such as Guangdong.

Estimate potential yield based on GAEZ model

This study employs the GAEZ model to estimate the potential yield of farmlands. Developed in the 1970s and updated in 2010 by the FAO and IIASA, the GAEZ model uses a step-wise optimization approach20. Initially, it calculates the photosynthetic thermal yield based on radiation and growing-season length. Subsequently, it incorporates temperature constraints to compute the photo-temperature yield. The model then adds water constraints to analyse the climatic production potential, and finally, it estimates the agricultural production potential by considering agricultural input levels and various management methods.

To account for the impact of climate change on potential yields, we used the 30-year average from 1980 to 2010 for climatic conditions. Meteorological data, including monthly maximum and minimum air temperatures, precipitation, relative humidity, wind speed at 10 m, and sunshine hours, were sourced from the Chinese Meteorological Administration’s data centre. These data were interpolated into raster surfaces at a 1 km resolution using ANUSPLIN software. We considered two water-limit scenarios: rain-fed and irrigated. We quantified the effect of irrigation as follows:

$${{Yield}}_{{total}}={{yield}}_{{rain}-{fed}}\times \left(1-i\right)+{{yield}}_{{irrigated}}\times i$$
(1)

Where \({{Yield}}_{{total}}\) denotes the farmland’s potential yield for a certain crop under a particular cropping system in an area (a raster cell); \({{Yield}}_{{rain}-{fed}}\) is the potential yield under the rain-fed condition, determined by the local climate; \({{Yield}}_{{irrigated}}\) is the potential yield in a situation where the water is entirely from irrigation (we assume that the water in this situation is not a limiting factor to the yield); and \(i\) is the ratio of irrigated area to the total area of farmland in a region. The irrigated area for each county in China was obtained from the China Statistical Yearbook (National Bureau of Statistics) for each year.

In line with the GAEZ model requirements, we incorporated soil conditions, terrain, cropping systems, and farmland distribution (see Supplementary Table 2). Soil data, including soil type, effective soil depth, and water-holding capacity, were obtained from the Data Centre for Resources and Environmental Sciences at the Chinese Academy of Sciences. Terrain data were sourced from the Shuttle Radar Topography Mission56, with a digital elevation model at a 90 m resolution. We accounted for multiple cropping systems using the multiple cropping index to calculate the highest potential yield under various systems, including single, double, and triple cropping per year.

We applied the GAEZ model to several key crops, including wheat, maize, rice, sweet potato, and soybean, which together account for 97.7% of China’s total crop production. The GAEZ model estimates potential yields for these crops based on their specific growth requirements and local climate conditions. These estimated yields are then weighted according to the area planted for each crop. For example, in a region where wheat is the predominant crop, the yield for wheat will carry greater weight in the overall assessment. The model was executed on a 1 × 1 km grid to estimate potential yield. Then, the yield loss due to urban/rural expansion from 1990 to 2000 and 2000 to 2010 was obtained by overlaying potential yield maps with land-use change maps. Model validation was performed by comparing the estimated yields for 2010, using 2010 climate data, against actual crop yields from the China Statistical Yearbook, covering over 1,924 counties. The correlation coefficient between estimated and actual yields was 0.83 (See Supplementary Fig. 3). After validation, we used the 30-year average climate data from 1980 to 2010 to estimate the farmland potential yield in China. This approach mitigates the influence of extreme climate events and variations, providing a robust basis for subsequent analyses.

Estimate the required built-up land in rural areas

Based on the dynamics of the rural population and built-up land per capita, the restorable farmland in rural areas was simulated in the following steps. First, we calculated the theoretical required total area of built-up land in rural areas based on the 2010 provincial population and the ratio of per capita rural/urban built-up lands, assuming that it has remained the same in urban areas for each province since 1990. If allocating built-up land based on population size is considered reasonable, we can estimate the necessary area of rural built-up land by examining future rural population projections alongside the per capita rural built-up land area for each province. Given the significant regional variations in per capita rural land area, we assume it corresponds to the per capita urban built-up land area of each province. This assumption is grounded in the ratio observed in 1990, prior to the emergence of the urban-rural dual increase in built-up land. Consequently, in Supplementary Table 3, we calculate the per capita built-up land for both urban and rural areas and estimate the corresponding areas that should be preserved for urban and rural built-up land, informed by anticipated population trends. Through this method, we estimated the area of rural construction land that will decrease with the future emigration of rural population, which will have the potential to be restored to farmland.

Simulate restorable rural built-up land and estimate potential yield

To identify patches of rural built-up land that should be restored to farmland, we adapted a methodology from urban land change simulations based on the FLUS model. This model operates under the principle that “everything that exists has its reason,” simulating multi-class land use changes. It posits that spatial driving factors—such as proximity to urban centers, transportation networks, water bodies, and environmental features—determine the probability of land use changes. Machine learning techniques are frequently employed to establish relationships between these spatial drivers and the likelihood of specific land use occurrences. The simulation does not aim to predict the suitability of an area for farming; rather, it assesses whether certain areas should transition away from rural built-up land use. This approach contrasts with simulations focused on urban land expansion. By analyzing the environmental characteristics of existing villages, we estimate the potential area and locations of rural built-up land that can be restored and predict the likelihood of these land patches being converted back to farmland. In our study, we treat rural built-up land as a distinct land use type and simulate its conversion back to farmland. We emphasize environmental factors such as elevation and proximity to rivers, administrative centers, and roads. Specifically, we consider a total of seven spatial driving factors: distance to city centers, distance to county centers, distance to rivers, distance to provincial roads, distance to highways, distance to railways, and distance to national roads. This comprehensive methodology, inspired by the FLUS model, aims to provide clearer insights into the restoration of rural built-up land.

Step 1: We begin by calculating the probability of occurrence for each rural settlement patch using logistic regression (Eq. 3). This probability reflects the suitability of each settlement area for farmland restoration. Subsequently, we assess the overall “justifiability” of each settlement patch by incorporating its area (Eq. 4). To train the model, we randomly sampled 10,000 points nationwide—5000 from rural built-up areas and 5000 from non-built-up areas. We employ logistic regression to establish the relationship between various spatial drivers and the probability of rural built-up land occurrence. In this analysis, spatial environmental variables serve as the independent variables (X), while land use types (rural built-up or non-built-up) function as the dependent variable (Y). Logistic regression analysis will be conducted on these training samples. This method has been extensively used in previous studies to simulate urban land expansion; we adapt it herein to prioritize the conversion of rural built-up land with low suitability probability back to farmland.

$${logit}\left({p}_{{occ}}\right)={\beta }_{0}+{\beta }_{1}x1+{\beta }_{2}x2+{\beta }_{3}x3+{\beta }_{4}x4+{\beta }_{5}x5+{\beta }_{6}x6+{\beta }_{7}x7$$
(2)

where \({logit}\left({p}_{{occ}}\right)\) represents the probability of occurrence for each patch. The variables \(x1\) through \(x7\) correspond to the following spatial environment factors, with the corresponding regression coefficients \({\beta }_{0}\) to \({\beta }_{7}\) being 1.931, −2.139, −7.337, −0.746, −1.183, −0.597, −4.377, and −0.328.

Supplementary Table 4 presents the results of the logistic regression used to evaluate the probability-of-occurrence for rural built-up patches. Supplementary Fig. 4 illustrates the spatial distribution of the probability-of-occurrence for rural built-up land that is likely to continue being developed. While these factors are essential for understanding the spatial patterns of land use, we recognize that additional influences on rural land use change exist. These include socio-economic contexts, land prices, agricultural practices, and various policy-related factors. Unfortunately, due to data limitations and the specific focus of our study, we were unable to incorporate these additional factors into our analysis.

Scale is another crucial factor in evaluating the justifiability of maintaining rural built-up land. Small patches usually have a scattered distribution, poor infrastructure access, increase landscape fragmentation, and cause land waste. According to the study from and Li57 and Zhang58, in traditional Chinese rural areas, many small hollowed villages require consolidation because they contribute to inefficient land use and hinder rural development. Therefore, our assumption that smaller patches are less justifiably maintained is grounded in these observed trends. We measure the scale suitability of rural built-up land based on patch size, considering small patches less justifiable for maintenance. The overall justifiability of a rural built-up land patch (P) is calculated using the following formula:

$${W}_{{size}}=\left\{\begin{array}{ll}\quad 1\quad ,\quad & \mathrm{if\; patch\; size}\ge 1\quad {{km}}^{2}\\ {Normalized\; size}, & {otherwise}\quad \end{array}\right.$$
(3)

where \({W}_{{size}}\) is a weight determined by the patch area:

$${P}_{\mathrm{justifiability}}={p}_{{occ}}* {W}_{{size}}$$
(4)

Step 2: The overall justifiability values are then input into the FLUS model to simulate the spatial transformation of rural settlements into farmland. Instead of directly converting areas with lower probabilities into farmland, we utilize FLUS for the simulation. This approach mirrors classic urban cellular automata methods, where low-probability areas do not necessarily transform; randomness is included to reflect the dynamics of social development. Moreover, farmland restoration, akin to urban land expansion, exhibits spatial dependencies: as adjacent areas are restored to farmland, the likelihood of conversion in surrounding areas increases. This spatial relationship is effectively captured through the FLUS model, enhancing the realism of the simulation. In the FLUS model, patches with the lowest justifiability will have a higher probability of being restored until the total area of rural built-up land in the province aligns with the theoretical value. These patches, which have been unnecessarily developed, can potentially be reclaimed as farmland. The key parameters used in the FLUS model are as follows:

Maximum number of iterations: 300

Neighborhood size: 3

Acceleration factor: Default value of 0.1

Number of threads: 1

Additionally, the base land use data for rural built-up land from 2010, along with future land use demands (calculated from Supplementary Table 3), are incorporated into the model. The transformation cost matrix is set to unrestricted, allowing for maximum flexibility in the simulation.

Step 3: Finally, the outputs from the GAEZ model are overlaid to assess the potential productivity of the land. The FLUS model simulates which areas can be restored to farmland, while the GAEZ model evaluates the potential productivity of these lands. Supplementary Fig. 1 illustrates the comprehensive framework for assessing farmland productivity restoration.

It is crucial to emphasize the limitations of relying solely on the selected seven spatial environmental factors, as other influences could also impact land use changes. Our evaluation model primarily considers two key elements: the location environment and the sizes of rural built-up land patches, but it does not factor in the costs associated with restoring rural built-up land. Due to a lack of data on land restoration costs across different regions, we cannot currently assess the economic feasibility of such restoration efforts. Thus, our assessment is limited to evaluating the potential of reclaimable land from the perspective of agricultural suitability. We explicitly acknowledge the uncertainties in the estimated yields and the simulation of rural built-up land restoration, noting that these estimates are based on idealized assumptions. Further research could provide more nuanced insights.

We appreciate the importance of addressing confidence in the estimated yields and the simulated transformations of rural built-up land. As mentioned earlier, this study is founded on several assumptions, particularly regarding the restoration of farmland and the suitability of rural built-up land for conversion. The GAEZ model, while one of the most authoritative large-scale crop yield models, was applied under idealized climate conditions. We recognize that climate change and other variables (such as technological progress and input levels) can significantly affect agricultural productivity, which is not accounted for in our model. Moreover, the simulation of rural built-up land restoration involves numerous assumptions regarding land transfers, compensation mechanisms, and regional economic conditions, which can vary widely across China. Therefore, our results should be interpreted as rough estimates under ideal conditions.